Welcome to the First Installment of Reader Questions!

We started Retailgentic 18 months ago now and across our various channels now have built to ~8000 subscribers across all the channels. For 16.5 of those months about every 2 weeks we’d get a question as a comment or into the various emails and other vehicles out there. In the last 45 days and really in the last two weeks, we have been getting a ton of great questions. If they are super-specific and we don’t feel would help the general reader, we handle them separately.

But now we have enough that we think are interesting to the majority of our audience that we thought it would be interesting to try something new - Reader questions!

There are three buckets of questions:

• Agentic Commerce Industry Questions

• Brand/Retailer Oriented Questions

• Misc - 2 Shopify questions and our most common pitch

Agentic Commerce Industry Questions

In this section we have 5 questions about some industry trends or news we’ve previously covered

Q1: What do you think of this Visa report?

• From: Marcus via LI DM

Answer: Thanks for the question Marcus! The Visa report was on my list to cover in a November round-up, but your question made me realize that with Thanksgiving and then all the BFCM news, we probably will have to skip November, so let’s cover it now. The Visa report in question is here - in it Visa ran 22 various AI models through a variety of agentic shopping tasks. The press release will take you to a whole micro-site. I’ve seen two analyses of the data that were great and really helped illustrate the data:

First there’s this graphic:

This graphic shows that Gemini 4 Argon is the top model completing all but 8 - 5 hard failed and 3 were ‘other outcome’ - it’s not entirely clear what this means, best guess is a CAPTCHA, last-step block - something like that.

Zooming out to 30,000 ft, you can visualize the ‘jagged edge’ of frontier models. Because it’s a new benchmark, we can’t look back in time too far, but, you can see back to Gemini 3.1 Pro which goes way way back to Feb of 2026 (haha- crazy right?). This also illustrates that as the models get better the frontier labs are pushing the edges out further and further - the last big OpenAI, Gemini, Muse and Anthropic models all made a clear push on browser use training, resulting in the leap forward you see here. Project the exponential curve forward, by this time next year, we can assume that the top models will be all blue and Visa will be re-doing the benchmark to make it harder because a 100% complete benchmark isn’t really a benchmark at all.

How the Visa benchmark works

It’s important to note that this benchmark was designed to focus on browser use. There’s a set of people that believe this is the path for the most completeness - if an agent can do 100% what a human can do, then that’s all of ecommerce (minus blocking). Another set of people believe that it’s better to have a set of machine-to-machine protocols (ACP, UCP, WebCP as examples) that speed up the transaction, add capabilities and take a lot of steps that can create problems from the process for an agent. I suppose there’s also a hybrid in there where you use a protocol for Plan A and browser use for Plan B.

Agentic Harnesses and Models

On top of the models are ‘agentic harnesses’ that add knowledge and capabilities. For example, Muse is the harness for Muse Wonder (Today, free Muse uses a less powerful, less expensive model, paid Muse uses Wonder). We’ve written that Google is imminently coming out with a shopping agent of some kind and it’s going to be interesting to see if they go totally browser based or a hybrid of UCP and browser, I am hoping we see that hybrid model it would be pretty killer IMHO.

But, what about cost?

The other graphic, built by A16Z, compares capabilities vs. cost:

Here you can see the frontier models currently spend $1-3 in token cost to achieve a complex shopping task - somebody has to pay for that (Merchant friends, you know who that is!). In a steady state World, these costs will decline, but we have a lot of weird things going on in AI and there are many pundits predicting a slow-down in data center buildout and thus a shortage in token supply with demand surging. This suggests prices are going to go up, way up.

Those were my observations from the data.

Q2: If Consumer Discovery with Agentic Commerce has Already Tipped, When Will We see the First Independent Agent Purchasing at Scale?

• From Unpursued AI, in reference to our article:

You’ve probably read, listened or seen me say that predictions are hard, not because of the topic of the prediction, but the timing. The exponential time-line messes with all of us. Here’s my thinking on this one:

• According to all the surveys, we just crossed 50% of consumers that use Agentic Commerce for discovery. I consider Agentic Commerce being born in Jan of 25, so let’s call that 2 years.

• We know from mobile that mobile got to 50% discovery quickly and then I remember all the years of adobe/salesforce/comscore data and eventually mobile got there on transactions.

• If we were on that timeline (2x ‘time to 50% discovery’ for ‘time to 50% purchase’) that would point to 4 yrs or 2031 (27→28→29→30 - puts us into 2031).

• BUT on the exponential, we can cut that at least in half. (27→28).

• We always see in digital retail, a Holiday surge in consumer behavior adoption that recedes a bit into Q1, but maintains a ‘new normal’ and we build back up by Q3 to the prior Q4, then in the new year Q4, set a record - rinse and repeat.

• Swirling all that together gives me as early as Holiday 2028 and as late as Holiday 2031. I’m going to go midpoint on this one and pick Holiday 2029

Short answer: I predict we’ll cross 50% of online transactions powered by Agentic Commerce by Holiday 2029. See you in 3yrs for when we revisit this one. This prediction is likely to blow the Agentic Naysayers’ minds, it will seem absolutely insane and unhinged to them. This implies discovery will be knocking on 75-80% I imagine.

Q3: How does a small brand with a modest budget stay visible once paid slots show up inside personal agents, and are you seeing early signs of what those ad formats will look like?

• From Sam, in reference to my Note announcing the AdGentic podcast launch (featuring Kiri Masters) — Oct 22

Sam good question. This one is hard to make a prediction on or give you any guidance on because, to your point in the question, we don’t know what the ad format or even the business model will be.

Business Model Thoughts

There are 3 potential business models:

• Advertising

• Consumer subscription

• Take-rate (% of sales model)

You can also mix and match these - ChatGPT currently has a hybrid sub and ad model - the sub ‘buys you out’ of the ads, like a Netflix or Hulu sub tier above the ad-plan.

In the marketplace world, Amazon 3P mixes all three - consumer subscription (Prime for free shipping), Take-rate - foundation of the 3P biz model and ads for enhanced discovery.

Amazon’s clever mix that was designed to align with consumers

The top reaction to Muse announcing a % of sale business model is “it’s over nobody will trust the agent because if you can pay more and the agent recommends you, then why would a consumer do that?”

Amazon looked at the Google ads and eBay’s v1 model and in both of those there is a misalignment with the shopper that goes like this. The merchant with the highest price has more margin and can afford more ads, so it’s a slippery slope to a forever degraded consumer experience because the shopper wants selection.

Amazon’s famous Bezos napkin flywheel diagram, has the customer experience right there. The Muse reaction makes one very bad assumption -that Meta/Facebook/Zuck are dumb. Instagram has ads and it’s a great experience, consumers vote with attention and with that KPI, there’s no arguing it’s popular and the ad load keeps going up. How’s that possible? The Ads add value.

What Amazon did is they said - how do we have ads, but not mess with that low price loop/CX in our flywheel?

They implemented a ‘price/quality’ gate: Once an offer was deemed to be the best for a particular SKU (what we popularized as ‘owning the buy box’), THEN that offer can be advertised. Then when the consumer shops for ‘hdmi cable’ or a category level search, they will only see SKU offers that own the buy box with ads on top of some to give them extra discovery juice.

The result: A balance between low price, quality merchant, high CX and ad revenue.

I imagine we’ll see something like this, in fact what’s convenient about product cards is you can almost do exactly the same thing, but it’s owning the Product Card, not the Buy Box. We’ll see, because there’s AI underneath, new models will be born that could be more sophisticated.

For example, a sophisticated merchant will pay more for a cart than an item, we should potentially be looking at % of margin vs. % of revenue. Looking further out, it seems like an economically rational merchant would pay more for a ‘net new’ customer transaction to acquire that customer and they may want to reward a very loyal customer. Imagine a very sophisticated ad system that projects CAC/LTV dynamics and that can be reflected in the economics.

Q4: Is anybody thinking about the consumer trust element — will Muse always work for the user, given Amazon is cheaper at least a third of the time?

• From Bharad, in reference to this Muse piece→

This question is a cousin of Q3, but I included it because it’s a clear example of someone assuming:

• Meta via Muse will do something dumb to screw up consumer trust - that’s 100% against their economic interests, so I don’t see it happening and as mentioned in Q3, charging $$ does not immediately ruin a CX.

• Also, Meta doesn’t have the power to make Amazon join the program, it’s not their fault they don’t include that inventory, that’s Amazon’s choice to not be where the shopper is.

The question also implies that the consumer is going to buy something from Muse and then suddenly realize they could have saved $ on Amazon. Back in 2018, Kasey from Deloitte came on the Jason and Scot show (Episode 114 here) and introduced us to the concept of the ‘great bifurcation’. To this day, 8 years later I think it’s still a megatrend that’s still playing out. The TL;DR on it is:

You see this all the time. We all have that friend that drives 15mins each way and sits 30mins in the Costco gas line to save $.25/gallon. We also know those people that Doordash everything including a $5 Starbucks that takes $25 all-in to get delivered to them. Which one of these people is wrong? The answer is: neither.

There’s an economic principle that we have to use to illustrate this called opportunity cost, let’s use that to evaluate each of these scenarios to explain how they can both actually make rational economic sense.

First, for the gas example, let’s say the friend uses 1hr to save $5 on gas. What we don’t know is the friend’s opportunity cost. Maybe our friend is doing this after hours at say 7pm ET, so that hour isn’t missing a hard currency, but there’s still a cost and that cost is the alternative for that hour. Maybe they would be spending money in that hour or sleeping or watching TV or ‘doom scrolling’ - they will economically be ahead in this scenario (+ $5) vs. the alternative. If they are missing their son’s 5th birthday, that’s vastly different math, and so on.

On the Starbucks example, perhaps the person is an AI developer and is currently billing $200-300/hr - a 20-minute Starbucks run has an opportunity cost of about $67–$100. Paying an extra $20 for delivery still leaves them about $47–$80 ahead.

Bringing it back to Q4: “once they realize Amazon is cheaper, won’t they lose trust?”

• First, I assume with estimated 100m Prime households in the US, nearly 100% of people are aware of Amazon and half the households have Prime.

• If a convenience-oriented person buys with Muse, Muse is very transparent in my shopping about what it’s looking at and not - and I can see Amazon isn’t in the list, but maybe I don’t care because I’m busy and want it to handle the discovery for me and, heck, handle the final piece so I don’t have to waste time looking everywhere. I want it off my list fast and I pay $2 more or it takes 2 days extra, for many items that’s not the priority convenience is.

• Every value-oriented consumer, by their very nature, will be acutely aware that Muse doesn’t include Amazon- in fact they are probably using Muse to price check, look for promo codes, etc. That’s one of Muse’s best use-cases - it saved me $40 on shoes the other day that weren’t even on Amazon because it’s a brand that has refused to sell 1P/3P on Amazon.

Conclusion: Shoppers are smart and will make economic decisions for themselves: convenience-oriented people will trust Muse because it saved them time, value-oriented will use it to price check vs. Amazon and only use it if it’s cheaper - they will be happy - everyone has a choice, when they make that choice they will be happy because they made it for a reason.

If Muse screws that up, nobody will go there, they will fail, Meta has a history and 4-6 properties over 1b MAUs to show they aren’t in the biz of shooting themselves purposely in the foot.

Q5: Will the costs of AI infrastructure start to erode value in the presentation layer?

• From Theo, in reference to a substack thread on Google/Amazon beating expectations and share prices dropping — Nov 3

Hey Theo, thanks for the question. I think what you are asking is, “if these models keep getting more and more expensive to build (train) and host (inference), will that make it hard for the economics to work?”

This question ‘pairs’ nicely with the Visa data - look at how far those models have come so fast - one year. By this time next year, the models will be able to shop and checkout on every site that doesn’t block them. By end of 2028, the late 2027 expensive models will slide down the Pareto Frontier curve (capability the same, costs come down) and so on.

The one fly in the ointment is the political orientation towards limiting and banning datacenters with some super far left politicians calling for a complete ban on AI. If we as an industry can’t build more datacenters or AI becomes basically illegal, then all bets are off. If any of that (DC slowdown→ AI Ban) happens, here’s what my Magic 8-ball of Agentic Commerce Predictions says:

Brand/Retailer Oriented Questions

We’re in a bit of a perfect storm here in November - most merchants are reviewing their Q3 data and seeing some wacky data:

• External searches (Google SEO/SEM) continue to rapidly slide

• Conversions on-site down - consumers are entering search terms the old search can’t handle

• Bot traffic > consumer traffic - this is causing all kinds of analytics chaos

The tipping point we’ve talked about from the Q3 Amazon and Google results is being felt in the iceberg part of our world - the vast surface area under the water.

Timing-wise, we’re now only 16 days from Thanksgiving and 45 days from Christmas, we are getting calls from brands and retailers alike that fall into two buckets - one urgent, one more 2027 looking→

Q1: What are the top things I can do between now and Christmas to convert some of this Agentic Traffic/Activity that is surging all of a sudden?

I’m going to keep this one short because we’ve covered all these in detail in these pages. The biggest things you can do in order of impact:

• Stop blocking bots - Many retailers we run into are blocking friendly shopper bots and don’t realize it. Their IT team or infrastructure layer have defaulted to ‘all bots bad’ and blocked them on your behalf. We have seen increasingly aggressive activity here where we had one retailer where we are 100% sure they are also blocking a large % of human traffic.

• (Retailers only) Send ChatGPT a feed - This is the single highest impact action you can take quickly and it will have an immediate (3 days or less) increase in sales. We have enough data now I can say this unequivocally. Bonus points for some basic speedy enrichment work here.

• (Brands) Upgrade PDP Page - This is super counter-intuitive for brands, but your lowly, vestigial PDP pages, even without transactions, have a ton of weight in the ‘mind’ of LLMs. If you take time to make sure all the basics are super clear in your meta data (ingredients, pairings, allergens, kosher status, gluten status, lactose status, etc.) this will make a huge improvement across EVERY agentic commerce surface, even Amazon AFS and Walmart Sparky. I realize this is hard to believe, we can prove it to you. Again, I can say this one unequivocally now - 100% a top priority high impact, almost zero cost thing you can do.

• Product Catalog Upscaling (aka enrichment) - This is overwhelming, I get it. Remember, we are NOT changing the human readable PDP. To narrow down the problem, do this:

  • Sort your SKUs by sales rank from highest to lowest

  • Pick a % - recommend 10%, 5% if that’s too much and just start with those to show ROI

  • Enrich them to what we call level 3. Level 1 isn’t enough, you need to get at least into the conversational attributes and shopper context. This includes reviews and a basic FAQ from a first round search for shopper context.

Example: You have 20k SKUs and this project feels insurmountable and you pitched management a $500k budget and they 🤮 all over that.

Take 5% - that’s 1000, manually approve the first 100 to get a feel, then come up with rules for auto-enrichment. Publish the first 1000 and measure that for 30/60/90 - you will then have a very clear ROI case to do the others and add more resources. This solves the chicken-and-egg problem as well as getting you to an ROI fast and also having the highest impact (you’d rather your top 1000 sellers have an X% bump in sales than your lowest sellers).

Learn more about Product Catalog Upscaling here→

Q2: I’m in my 2027 Budget Planning - how should I budget for Agentic Commerce?

This one can go long, but you’ll notice it dovetails nicely with the answer above - I would say at a minimum what you want to do is start with these initiatives:

Brand:

• Don’t block bots (free!)

• Upscale your PDPs (very low cost and can be automated - it’s meta data not human data!!!)

• Level 3 enrichment on your PDPs and pushed/avail to your retail partners (it’s meta data not human readable!) This is where the expense is at a team time level.

• AI shelf monitoring - Also expensive in team time.

Retailer:

• Don’t block bots (free!)

• Agentic Commerce optimized datafeeds to ChatGPT, Google Gemini, Copilot

• Level 3 enrichment on your catalog (not human readable!)

• Upscale your PDPs (not human readable!)

• ACO Monitoring

If your budget cycle looks at ROI, that’s a longer story and so variable that it’s hard. Perhaps once through the holiday crazy, we’ll look at ‘the ROI of Agentic Commerce’ - We actually have a very detailed model, it’s not what I’d call substack ready, but if you are interested, ping me and I can walk you through it and send it over.

We also have lots of resources on how to execute this. Our most popular started here in strategic question 1→

This one FAQ generated so many follow-ups we created this guide called ‘Building your ACO organization’.

It lives here.

The top trend we’re seeing on this topic of organizing for ACO - create a TIGER team that has the remit to just go knock this out.

Misc: 2 Shopify Questions and Our Most Common Pitch

Q1: Shopify’s Agentic page credits real sales to ChatGPT — have you seen other platforms give sellers a view like that yet?

• From Sam referring to this news roundup where we showed more about this Shopify feature→

Sam - thanks for the question. At the SMB platform level, this is necessary because SMBs can’t patch together their analytics, infrastructure layer, tracking, etc. Most larger retailers and brands are very good at this and have 20yrs of perfecting it with all kinds of attribution models and what-not.

With that background in mind, and I’ll say that focusing in and understanding Agentic Commerce at the level I do, leaves me a blind spot on the platform layer that folks like my friend Rick Watson cover more deeply. That being said based on what I know:

• Shopify - gold standard on this tracking

• BigCommerce - no idea, they seem to be struggling, so I doubt there is a lot of innovation going on there. My bet is YOYO - You’re On Your Own.

• Magento - Also struggling, YOYO.

• Salesforce - YOYO - most customers use GA4 or Adobe

• CommerceTools - same

• SAP Hybris - Same

If anyone knows more or wants to correct me on this, LMK.

A new rebel emerges

There is a new challenger to Shopify that’s coming up in a lot of conversations that I need to learn more about and want to have on the podcast (reach out please!)

It’s called Scayle and customers are raving about the AI capabilities and back-end agentic layers and we're seeing them win some big deals against Shopify.

Beware DACT!

That all being said, we’re seeing large retailers and brands are now tracking a lot more of these Agentic Commerce traffic and sales signals, but Dark Agentic Commerce Traffic is still a problem, this has to get fixed next year and I have faith that it will because ads won’t work without it. More on that topic→

Q2: I’m increasingly hearing that the relationship between Shopify and Stripe is strained and Shopify is looking at alternatives to Stripe for the Shop Pay Infrastructure - do you think Shopify and Stripe will ‘break up’?

This is a long one, here’s my shortest answer. Let’s start looking at the stakes best represented by market cap→

• Shopify: ~$200b

• Stripe: $160b (rumored as they are private)

• PayPal: $50b

I believe if Stripe was public, their valuation would be north of that $160b, especially after the acquisition of OpenRouter and all their smart Agentic Commerce moves.

I mention this because it takes a lot of things off the table - these companies are too large for one to buy the other or to merge. PayPal is very buyable and in fact, it was rumored Stripe took a run at them. I mention that because Braintree would be a fast way for Shopify to replace and compete directly with Stripe.

All that being said, it’s well known in the tech founder circles that Tobi, the Collison brothers and DHH are BFFs and spend a lot of time hanging out drinking Guinness, racing cars, talking tech, nerding out, etc. (Examples here and here and here ).

That’s a long way of saying I don’t see it happening.

But as a thought exercise, here’s my 2 best shots at how it could happen:

Scenario 1: Shopify is tilted at Stripe implementing Link, and now Link pay that is a lot like Shop Pay one step checkout. They decide - why are we built on a company competing with us - let’s go control our own destiny - DIY or buy PayPal. They could also want more economics in payments and Stripe has to be making some margin - Stripe’s margin is Shopify’s opportunity. This could also happen if Shopify is under margin or growth pressure.

Scenario 2: We know that Shopify is unbundling Shop Pay as a stand alone service. It’s going to naturally compete with Stripe/Link. Let’s say Shopify starts winning deals and Stripe starts to feel pain (in the form of slowing growth in revs/TPV). Shop Pay is one of the biggest honeypots of payment revenue available. In a world where growth is tough and there’s a competitive situation, perhaps Stripe goes for that honey pot. It starts by rug pulling (cancelling deal or raising prices) the Shopify deal and starts promoting Link or some variation into the Shopify store network as a cheaper/better/alternative payment mechanism - “Go direct to stripe and save X% per transaction!”

What feels wrong about that thought exercise is neither one feels like a slam dunk win for Shopify and Stripe, there’s less risk and more up-side staying in the symbiotic relationship than breaking it - at least today. The thing about disruption like that being caused by AI is it can create rapid seismic shifts and maybe one of those tilts things towards a break-up, but I don’t see it today.

Observation: Pitches

As part of this exercise, for the first time, we forensically went through all the inboxes and messaging platforms to gather these questions. One unanticipated outcome - we noticed that every week at least 2 founders reach out and describe themselves as building “the decision layer” or “trust layer” or “authentication layer” for agentic commerce and want to come on the pod.

You haven’t seen them on the pod because they are all at the idea level, no customers, maybe a demo and ‘product’, but because these sit in the middle, you need two sides of the equation - supply and demand and maybe three - payment partners. It’s going to be hard being a startup wedged between the billion and trillion dollar behemoths fighting over the Agentic Commerce pot of gold: Shopify, Stripe, Google, SpaceX, ChatGPT, Anthropic, Meta, Amazon, Walmart, Visa, Mastercard, Amex, WorldPay and so on.

That being said, these are smart people that see a need and it’s a reminder that while we’ve come a long way as an industry, there’s still a lot that needs to get solved to reach that transactional tipping point mentioned way back in Question 2 at the top of this article.

Editorial Note

We’re entering the Thanksgiving holiday season here in the US, we’re going to be pausing the Tuesday pieces until after Thanksgiving because I’ll be traveling to some shows and Thanksgiving plus I do expect we’ll have a lot of news before and just after Thanksgiving, so from November 11-December 4 our publishing schedule will be a bit erratic compared to normal. We’ll do our best not to bury your inbox, but we can’t control what’s coming!

This morning at 9am ET, Amazon announced that AFS is now coming on text. First let’s look at the news from Amazon and then look at the strategic implications. Given the Q3 results we just saw, as you can imagine, we have some theories!

Do you like being the first to know breaking Agentic Commerce news? Only one way to do that→

Subscribe now

The Details of AFS via Text

We’re only 20 days from Thanksgiving and as we’ve been predicting, there are a few more Agentic Commerce surprises to drop before it’s “GO TIME” on Holiday 26. Amazon this morning announced AFS via Text. Here’s the details→

Beginning today, a subset of U.S. customers will receive a text message inviting them to try the experience. The option to text Alexa for Shopping will roll out to millions more customers in the coming weeks. To get started, simply accept the terms, link your phone number, and start shopping—all from your text messages. Customers who haven’t yet received an invitation can sign up to get news and offers from Amazon by text at www.amazon.com/text-alexa-shopping and will be among the first to know when more invites are sent.

Of course, once we have access here at Retailgentic, we’ll be doing some live demoing - stay tuned on that.

Amazon continues:

Customers with access can conveniently text Alexa for Shopping with all kinds of shopping requests. Send a text to start a weekly grocery order, search for a birthday gift you need by Saturday, share a photo of the outfit you spotted while scrolling and ask for similar looks, or fire off a quick text that you’re out of dog food and please order more. Alexa for Shopping takes it from there, searching for products, comparing options, and messaging back with suggestions. When you’re ready, it can add items to your cart and seamlessly complete the purchase on your behalf. The more customers use it, the more it learns their personal preferences, shopping habits, and tastes—delivering an increasingly personalized experience over time. Tell it you only want to see vegan products and it remembers. Mention you’re training for a marathon and it factors that into future recommendations. The contact will appear in your phone as “Alexa Shopping.” Just text it to get more of your shopping done quickly and conveniently.

In the press release they go through about 20 AFS features available via text - most of them are the usual suspects (reorder, price history, memory, pics, tracking, auto buy, etc.), but these caught my attention as ‘new’ use cases they must be focusing on/adding:

• Hand off your to-do list. Text “I’m out of dog food, need a birthday gift for my sister, and we’re low on paper towels” and Alexa for Shopping gets to work on all three, messaging back with product suggestions as it finds them.

• Simplify grocery shopping. Set up a weekly grocery order, text a photo of a recipe to add all the ingredients to your cart, or plan a last-minute brunch by texting “deliver eggs, bacon, pancake mix, and maple syrup in 30 minutes.” Make changes on the fly by texting “add broccoli to my last order,” or ask for dinner ideas based on what you’ve recently purchased.

Grocery shopping 🤔 we’ll come back to this one. In fact, the hero image from the release - you can see it above shows this exact use case - texting AFS ‘hey I’m making lasagna tonight, order the ingredients’. Put a 📍 in this one.

Use cases visualized

Here’s an image of a reorder using a picture - what we’d call multi-modal search - I wonder if it will take some voice memos. It’s always been a bit odd to me we can’t talk to AFS - the A is for Alexa afterall.

The final use case they illustrate is a gift buying experience/reco type use case (gift for Mom)→ I can never remember the name of those wax bulb things either, wonder if they pulled data on this ;-)

Multi-device, AFS Meets you Where you are → “One Conversation, Everywhere”

The last tidbit I thought was interesting:

One conversation, everywhere
Amazon is rolling out the ability for a request started over text to carry over to Alexa for Shopping in the Amazon Shopping app, on Amazon.com, and Alexa-enabled devices, and vice versa. Start browsing gift options over text while waiting for an appointment, then open the Amazon Shopping app at home to compare and check out. Alexa for Shopping will remember what you were working on, what you’ve told it, and what you prefer, so every interaction picks up where the last one left off. Over time, your Alexa Shopping contact will also include your Amazon order updates, delivery notifications, and deals—all in the same thread as your shopping conversations.

In today’s World, multi-device, multi-modal is table stakes and it’s good to see them covering the bases here.

Now let’s look at what this mean for merchants in the Amazon ecosystem and then look atwhat Amazon’s strategy is →

AFS via Text: Ramifications for Merchants

This one line from the PR stuck out to me as someone in the day-to-day battles of ACO (on and off Amazon)→

This is a huge one. Ever since we launched our AFS solution, we have found that a surprisingly large % of ASINs have conflicting or flat-out wrong information.

We are now seeing perfect and expanded product data go through this 4-step lifecycle exceedingly fast:

1. Store: People will figure it out from the packaging - low priority

2. Ecommerce: Shopper will look at the image and figure it out - med priority

3. AFS: AFS returns products quickly and remember preferences. If you are vegan and don’t explicitly say vegan, you lose the order. high priority

4. AFS via Text: The consumer never sees a pdp, they trust: CRITICAL

Amazon Strategic Implications

As detailed in this week’s Tuesday piece, I believe in Q3 we hit a consumer behavior shift ‘tipping point’ where consumers now as a majority favor agentic commerce experiences over the trad search experience. Details here→

Also, in Q2 2026 in our big end of Q3 ‘state of Agentic Commerce report here→

one of the things we talked about was that personal shopping agents have two attack vectors on Amazon:

• Out supply Amazon - There are 1000’s of brands that have chosen not to sell 1P or 3P with Amazon. But almost all of them have websites with ecommerce. Therefore if Personal agents add Walmart plus 1000 brands not on Amazon - that’s inventory you can’t get there, that now is finally discoverable, some would argue, in a system even better than Amazon’s discovery system (the OG, not AFS).

• Know the consumer better - If a personal agent can work with a consumer horizontally OUT of shopping - know their schedule, skin situation, health, travel, family and friends, they could jump the pole-position on Amazon and get orders before the consumer even thinks of going to Amazon - I see this happen with Zillennials all the time now.

Let’s look at each of these in detail:

Out Supply Amazon: Amazon’s Reaction: ShopDirect

I’ve long thought that ShopDirect was a hedge agains this selection attack - If you’re Amazon how do defend against this? The brands have already voted with their feet not to do 1P/3P. Hmmm. I know you go and do what the personal agents do, you crawl the sites and make it available as a new third option: ShopDirect.

Amazon is not resting here, I believe they will wire UCP in here eventually and just the last week, the Information found 250m ShopDirect listings (details here) and Amazon launched a merchant center for ShopDirect (details here).

I’m not the only one thinking this, check the Information headline:

Know the Customer Better Amazon’s Reaction: AFS/Alexa has to become a PSA to defend this one.

In that Q2 report we used this graphic to illustrate the horizontal vs. vertical attack vector:

Let’s say you’re talking to Gemini about your skin and your skin regiment and it starts recommending products. Sure you could also go check them on Amazon, but you’re right there and it sure is easy to just hit that add to universal cart and get the cleanser and moisturizer - also what if they aren’t even on Amazon?

One popular use case that we see this showing up in survey data is meal planning (remember the hero image). So it’s another ‘tell’ to me that something is going on that’s enough for Amazon to play a bit of defense. As someone that’s followed them since inception, this is not normal. They are a 100% offense company.

Amazon’s natural defense here is to marry Rufus and Alexa (check), and blend them together so Alexa is the horizontal agent and one if it’s features is ‘shopping’ or AFS - ultimately you drop the AFS and it’s just Alexa.

Conclusion

Even though Amazon doesn’t see any value in personal shopping assistants and Jassy continues to say they are garbage UXs, they sure are moving to be one pretty quickly. Actions speak louder than words. It’s going to be interesting to see how both of these strategies progress into 2027 and if they can slow down the onslaught that of new attackers that Amazon hasn’t seen at this level since their founding.

Stay tuned, we’ll be keeping close track and if you do sell 1P or 3P on Amazon, remember we have a dedicated channel for this, RetailPlaybook that goes way deeper.

Subscribe now

Lisa Yamner is Co-Founder and Chief Brands Officer of Daydream, the AI-powered fashion shopping agent she started with Julie Bornstein. Lisa spent 12 years at Google, where she founded the fashion and luxury vertical in 2011. She then spent about a year at Condé Nast building its fashion vertical before teaming up with Julie at The Yes, which they built from 2018 to 2022 and sold to Pinterest.

Shopping still wasn’t solved, so they went back at it. Daydream raised in the summer of 2023 and started building in earnest in January 2024. Lisa has spent her career at the point where fashion meets technology, often as the person convincing luxury brands to trust digital.

Get your products ready for Meta Muse’s first Black Friday

Shopping is the top use of Meta Muse. Meta says more than 10 million people in the U.S. and Canada set it up in six weeks, and the average purchase is over $200.

ReFiBuy now supports Muse. Find the product data gaps that hold your products back, approve agentic enrichments with the evidence behind them, and monitor how your products show up in Muse.

Black Friday on November 27 will be Muse’s first. Early access is open now for brands and retailers in the U.S. and Canada.

Request early access →

Listen/Watch Our Interview:

When you think about vertical shopping agents, the AI agents built to solve shopping in one category, one name keeps coming up: Daydream. We’d wanted to get the Daydream team on Retailgentic for a while, and this week we finally did. This one is part founder story and part deep dive into what a fashion-native agent learns when shoppers stop typing keywords and start describing their lives.

Highlights

• 𝗧𝗵𝗲 $𝟱𝟬𝗠 𝗿𝗮𝗶𝘀𝗲: Daydream has raised $50 million. The Pinterest exit kept earlier investors on board and brought in new ones who had wanted into The Yes before it was acquired.

• 𝗢𝗻𝗹𝘆 𝟯% 𝗼𝗳 𝘀𝗲𝗮𝗿𝗰𝗵𝗲𝘀 𝗻𝗮𝗺𝗲 𝗮 𝗯𝗿𝗮𝗻𝗱: In Lisa’s Google days, brand search drove everything. On Daydream, people search by occasion, lifestyle and body type.

• 𝗤𝘂𝗲𝗿𝗶𝗲𝘀 𝗴𝗲𝘁 𝗱𝗲𝗲𝗽𝗲𝗿: New users start with Google-style searches like “red dress.” The more often they return, the longer and richer their queries get. One real example: “Going to a Tulum vacation for five days. Help me dress like a Jacquemus ad.”

• 𝗦𝘁𝘆𝗹𝗶𝘀𝘁 𝗶𝗻 𝘆𝗼𝘂𝗿 𝗽𝗼𝗰𝗸𝗲𝘁: “You never ask your personal assistant to actually pick it out for you.” Lisa says that gap is where Daydream fits.

• 𝗟𝗶𝘃𝗲 𝗶𝗻 𝘁𝘄𝗼 𝗱𝗮𝘆𝘀: Daydream works from brands’ existing Google or affiliate feeds, takes its commission through affiliate networks, and gets most partners live within about two days. The catalog covers 350+ partners, 10,000 brands and 3 million products.

• 𝗧𝗵𝗲 𝗳𝗮𝘀𝗵𝗶𝗼𝗻 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗴𝗿𝗮𝗽𝗵: Daydream doesn’t take a feed as finished. It enriches each one using text, images and input from real stylists, so it can tell that a skirt is both A-line and pleated, and that a white dress is the wrong answer for a wedding guest.

• “𝗠𝘆 𝗳𝗲𝗲𝗱 𝗶𝘀 𝗴𝗮𝗿𝗯𝗮𝗴𝗲”: It’s what Lisa hears most often from brands. Daydream’s goal is to level the field so a brand with a 25-person feed team doesn’t automatically beat a one-person Shopify shop.

• 𝗧𝗮𝘀𝘁𝗲 𝘃𝘀. 𝘀𝗽𝗲𝗰: Lisa splits shopping into two kinds. In spec-based buying (“the Samsung Frame TV in 65 inches, just buy me the best price”), agents will act fast. In taste-based categories like fashion and home decor, people “want agency in” the process.

Lisa has spent her whole career bringing fashion along into what’s next, and Daydream may be the clearest picture yet of what a category-native agent can do.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

One of our most popular articles this year was my daughter Rory’s story on how she and her friends did their back-to-school shopping by ‘just asking chat’.

Then during Fall break when she was home, I watched her stand in front of a wall of lip liners at Ulta, point her phone at the display and give ChatGPT three criteria. It told her which product fit and where to find it. Third up from the bottom, fourth from the right. Pretty good directions for something that doesn’t work at Ulta.

Now put yourself in the shoes of the brand two slots over. At what point did you lose the shopper? Was it the register? I’d argue it happened right there in the aisle. By the time somebody taps a card, your chance to win that recommendation has already passed.

In fact, our proprietary ChatGPT user behavior research showed that 84% of the time, the consumer ended up picking one of the products from the product card.

What got my attention this earnings season was seeing a version of that same behavior show up at Google and Amazon, two companies with a very expensive front-row seat to how people decide what to buy with extremely high consumer shopper distribution.

Longtime Retailgentic readers know my three steps of the shopper’s journey. Research → Find → Buy. Research is figuring out what fits your needs. Find is narrowing the field to a specific product and offer. Put those two together and you have Discovery. Buy can be, and frequently is, a separate step in the shopper’s journey. I can ask an agent to find the right product and still be perfectly happy to complete the purchase myself, knowing I’ve found the right product that matches my specific criteria and is the best competitive offer out there.

My thesis is that Q3 was the tipping point for Discovery. Buy will come later, but not too far behind. The handoff will happen fastest where the shopper already trusts the retailer (Amazon/Walmart) with an address, a payment method and the inevitable return. Then, ironically, they will help consumers become comfortable and they’ll move that behavior outside of the retailer’s walls to answer engines and personal agents.

Exhibit 1: Fresh McKinsey Data

McKinsey published its new Agentic Commerce Consumer Survey where they surveyed ~21,000 consumers across 11 markets in June. So this is a look at the shopper heading into Q3.

The headline number was 53%. Across the markets surveyed, just over half of adults had used AI for at least one shopping task in the previous three months. In the US it was 48%. “Shopping task” covers a lot of ground. Turn to Exhibit 4 on page 7 and you can actually see where shoppers put AI to work.

Sidebar: ACO Everywhere

In July we introduced the concept of ACO Everywhere here. The TL;DR is that there is increasing evidence that not only are consumers adopting Agentic Commerce quickly, they are using multiple ‘ACO surfaces’ in their shopping journeys.

McKinsey has fresh new data that supports this and shows a fascinating global picture:

For most countries we are settling in on ChatGPT, Gemini, Amazon but you can see there’s a wide diversity especially in China and Japan. But ChatGPT has the ‘most used’ pole position in 8/11 countries measured.

Compare Products.

Among AI users, 50% had used a stand-alone AI platform for this, versus 29% who had used a retailer’s assistant. Think about what that means if you’re a merchant. Before the shopper ever thinks about your site, there’s a coin-flip chance that an agentic commerce surface has already explained the differences between your product and the one next to it.

Complete Checkout

Now for the early Buy/Checkout data. Across comparable categories, McKinsey puts the average share of US AI users reporting a purchase directly within Agentic Commerce at 17%, up from 10% in the earlier comparison.

We saw the stakes play out in our own ChatGPT clickstream study. When shoppers reached the list of offers for a product, 76% picked the first offer and only 14% picked the second and 84% of the time, one of the product cards won the purchase, regardless of where it completed.

Those are the highlights from that report, but I recommend reading the whole thing, it’s one of the best of the year. That being said it’s stated preference as we love to point out. The next two exhibits are observed behavior through the lens of Google and Amazon’s earnings.

Exhibit 2: Google’s Q3 Results

In Google’s earnings (Wed Oct 28), there was a mystery: Google stated very clearly that the year-over-year growth in Search queries was the fastest in five years. But at the same time, Google Search & other revenue grew 12% to $63.4B, down from roughly 17% growth in Q2. This was well below Wall St. expectations and since then the stock has traded down 14%.

Looking at the absolute dollars highlights how extreme this change was:

Here we see the incremental new revenue:

• 2024: Q2→Q3 - $880m - 1.8% q/q

• 2025: Q2→Q3 - $2.38b - 4.4% Q/Q

• 2026: Q2-Q3 - $80m - 0.1% q/q - basically flat growth on record query growth.

Here’s what’s going on..

The Solution

Unfortunately, Google doesn’t give us anything quantitative on search volume other than the occasional hint (e.g. fastest growing ever), so we can’t show you the lines diverging, but above is an illustrative example. What we have is searches surging, but monetization going down rapidly.

In the Q+A part of the earnings call, Google’s chief business officer, Philipp Schindler, pointed to the answer: AI Mode queries were growing fast, BUT monetizing at a lower rate than traditional Search. There are also some FX elements and y/y World Cup comps, but I’m guessing what’s happening here over the arc of the trend forming is that consumers are moving faster to AI Mode than Google anticipated.

But Why?

We’ve stated on these digital pages that the first principle in our world is consumer behavior, and in both the Google and Amazon results, we see first principles at work - in the form of rapidly changing shopper behavior.

Why is AI Mode surging? Easy: because AI Mode is objectively better than traditional Google. Answers are better than search results. On top of that,iIn the last 3-4 months Google has made several UX changes that make it easier to slide right into AI Mode.

Examples include clicking ‘more’ in AI Overviews, which now goes to AI Mode; Universal Cart and UCP-powered discovery in traditional search mode; and changes to how the ‘AI Mode’ box works and where it’s placed.

What interests me is that shopper behavior can change inside Google’s own network so quickly that it’s out-pacing the best monetization engine ever invented. You can hear that the execs are blown away by the characteristics of the AI Mode searches too. Here’s a sampling of reasons to love this move long-term:

• Longer queries: Pichai said AI Mode queries are, on average, three times longer than traditional searches, reflecting more personalized and complex questions.

• Multimodal usage: One in five AI Mode searches globally uses voice, images, or documents - not just typed text.

• A unified Search experience: Google is building “one seamless experience across AI Overviews, AI Mode, and traditional search,” rather than treating them as separate products.

• Incremental queries and commercial intent: Schindler said the majority of AI Mode queries are incremental, and many are upper-funnel queries that capture “new high-value commercial intent.” Management sees this as creating more monetization opportunities over time. FWIW, I agree, but Wall St probably won’t be too patient waiting for it to materialize.

Sidebar: Google is Working on a Consumer Shopping Agent (My prediction: It drops 11/12/26)

Outside the scope of this article, Sundar was pretty clear on the Google call that ‘later this year’ they will launch a shopping agent. We’re 23 days out from Thanksgiving. Last year on 11/13 they announced ‘agentic checkout’. I suspect we’ll learn more about this agent in the next 3 weeks or sooner. That was the Thursday 2 weeks before Thanksgiving. Mapping that to 2026, I’d put the most likely day at November 12th. Predictions are hard, being super specific is 1000X hard, let’s see how this goes.

Exhibit 3: Amazon’s Q3: AFS and the Amazon Ads Business Slow Down

Alexa for Shopping (AFS, formerly Rufus) starts with an advantage any independent agent would love. Amazon already knows what you ordered and where to ship the next one. Back in Q2 it reported 350M+ customers had used AFS in the trailing year. On Thurs 10/29, Amazon announced Q3 results and revealed that number surged to 400M+, plus interactions were up 570% year over year.

Jassy also said US customers who used AFS spent 50% more per order than customers who didn’t. As I asked last quarter, how much of that is Amazon’s best shoppers being the first to try the thing?

Ready for Deja Vu?

But get this, at the same time Amazon’s AFS Agentic Commerce’s usage is surging, you sitting down, the Ad system has slowed down. Where have we heard this exact same thing before?

Cleaning out the y/y noise…

Right after the earnings dropped and before the earnings call, I posted about the steep drop in y/y Amazon Ad Revenue:

Over 50 LinkedIn commenters were quick to point out that with the pull-forward of Prime Day from Q3 in 2025 to Q2 in 2026, it created a huge comp, but when I did back of the napkin math, it still didn’t add up. There was also a very unclear SEM piece and an NBA piece referenced on the call.

Regardless, my macro point was - you can see Amazon Ads humming along at 22% and sure you see the surge to 26% and then dip to 11, but even moving the $ manually, you have a sudden new normal much lower than the 22-24% y/y growth we’ve gotten used to since Q2 2025.

After earnings, and digesting all the Wall St. reports, here are the details that emerged that point to what’s going on and, I believe, validate my theory.

Clearing out the Prime Day Move

First, Mark Mahaney had a great idea, to get rid of the Prime Day move - why not look at Q2+Q3 in 2024, 2025, and 2026 to get back to ‘apples-to-apples’ comps. Great idea and here’s what that looks like:

Now you can see we went from 23% y/y growth in Q2+Q3 25 to 18% apples-to-apples.

That gets us the story for Q2+Q3, but then Amazon not only guided Q4 and forward to ‘high teens', but Wall St gave their Amazon Ads line in their forecasts a serious haircut into 2027 and beyond:

This happened just days after Google saw a similar situation: a surge in new consumer behavior and a reduction in monetization. I think it’s too much of a coincidence for it not to be the same thing.

Consumers are moving from traditional Google and Amazon search experiences to their new AI-powered experiences (AI Mode and AFS) and it’s putting monetization pressure on the highly monetized mature discovery surface.

Which agent should Amazon be watching?

Could some of that be driven off system by ChatGPT, Muse, etc? Look at that data from McKinsey - 53% and while Google is in the mix, you don’t see Amazon there. We’ll never know because Amazon won’t disclose, but these three things coming out at the same time really made me 🤔.

On the Jason & Scot Show Episode 354 Q3 Amazon deep dive (here) , Jason pointed out that I am looking at Amazon’s ad slowdown and trying to figure out whether ChatGPT or Muse was stealing the traffic. Jason’s candidate for the agent responsible: AFS. I count that as Agentic Commerce, so sure.

But my Spidey sense says that Muse’s 10m extremely active users and ChatGPT’s 1.2b and even Google’s surge in AI mode have to be causing some up-stream disruption. What do you think?

Leave a comment

The Agentic Commerce Naysayers Chime in: Eric, Andrew, Mirages, Hallucinations, Definitions and their Buy Button Obsession

I’ve read with great interest all previous installments of Eric Seufert’s ‘Agentic Commerce is a Mirage’ series - he recently came out with part 5 on the back of Walmart’s new gateway and Amazon’s Q3 surge in AFS supports his position. Eric’s actual position (really a definition imho) is more interesting than the attention grabbing title of the series. Right there in part one, he says retailer-owned agents make economic sense. He doubts an independent agent can get between Amazon and its customer, take a fee and keep the data and advertising opportunity for itself. Amazon blocking Muse and the trouble with OpenAI’s Instant Checkout are pretty good exhibits for the defense.

Where we part ways is what a brand should conclude. Let’s give Eric the whole economic argument. Walmart sets the rules, Sparky handles the transaction and Walmart keeps the payment relationship. Now suppose an outside assistant has already helped the shopper pick three blenders before it asks Sparky which ones Walmart carries. If your brand’s blender doesn’t make the list, how much comfort do you take in knowing Walmart still processed the payment? Walmart’s announced Agent Gateway makes that division of labor possible. We can argue about the eventual fee split while brands worry about losing the game up-stream.

Eric’s original thesis · Part 3 on retailer agents · Part 5 on Walmart · Walmart’s announcement

Hallucinations

Andrew Lipsman’s ‘Agentic Commerce is a Hallucination’ objection starts with the shopper. In his original piece, easy purchases save too little time to bother delegating, while hard purchases are exactly the ones where you want to exercise judgment. His October follow-up catches a genuine sleight of hand. Meta telling us people used Muse to shop and that completed purchases averaged over $200 tells us precisely nothing about how many of those shoppers bought something.

But my daughter at Ulta is a pretty good example of what’s missing from Andrew’s easy/hard split. She had opinions about the product she wanted. She gave the agent three criteria and what she was happy to delegate was discovery for the one item on a crowded shelf that met her criteria. Buying it herself at the counter is entirely consistent with letting AI decide which brand gets the shot. Andrew’s test would record no agent purchase. I would count it as Agentic Commere; Andrew does not.

But, from the brand’s perspective, something pretty important happened - AI played a HUGE role in the zero-sum game.

Andrew’s original case · His October headline rewrites

What Can YOU do for Holiday 26?

If you agree with my definition of Agentic Commerce and that the entire shopper journey from Research→Find→Buy is important, my recommendation is to pick a holiday SKU you care about. Give Google AI Mode, Muse, ChatGPT and AFS the actual constraints a customer will bring to the journey in this AI Era.

Does your product make the shortlist? Is the reason accurate? Can the agent find an in-stock offer that will arrive in time? Is the offer right? Are you happy with the product card(s) that pop? If you wait for autonomous checkout to get big before you care about Agentic Commerce, you may be scratching your head for the next couple of years, wondering where all the shoppers and traffic to your site went, while that checkout button starts to get pretty lonely - at that point it maybe too late to make up lost ACO ground.

Subscribe now

Subscribe now

Kevin Indig is the author of Growth Memo a weekly Substack on growth, SEO, and AI visibility. He comes from the SEO world and has spent more than 15 years in the space. For the last four years he’s advised fast-growing tech companies, including big names like Meta, Reddit, and Ramp as well as smaller startups across commerce and SaaS. Before that, he led a large part of the growth team at Shopify, so his roots in e-commerce go deep.

Growth Memo

Growth Memo delivers sharp, research-backed insights on SEO, organic growth, and AI-driven search for senior marketing leaders.

By Kevin Indig

These days he’s often the person companies call in when growth stalls and they need a turnaround plan. He also runs a series of user-behavior studies on how everyday US adults shop in AI mode and ChatGPT.

Listen/Watch Our Interview:

This week is a research-heavy episode with a conversation of one of the sharpest observers of how real people search and shop now that AI sits between them and the products they want. Kevin’s user-behavior research helped inspire our own product card study at ReFiBuy, so we’ve wanted to have him on for a while.

Highlights

• 𝗧𝗵𝗲 𝘀𝗵𝗼𝗿𝘁𝗹𝗶𝘀𝘁 𝗲𝗳𝗳𝗲𝗰𝘁: Over 88% of participants in Kevin’s user-behavior studies didn’t question the frame from AI. “When AI says, ‘These are the best X,’ people barely question it.”

• 𝗧𝗼𝗽 𝗼𝗳 𝘁𝗵𝗲 𝗹𝗶𝘀𝘁 𝘄𝗶𝗻𝘀: Almost 75% of people picked the item at the top of the shortlist. The top-heaviness of classic Google search carries straight over into AI mode.

• 𝗢𝗻𝗹𝘆 𝟯% 𝘃𝗲𝗿𝗶𝗳𝘆: In a ChatGPT study, only 3% of people checked the shortlist anywhere else on the open web: not publishers, not affiliates, not YouTube.

• 𝗧𝗿𝘂𝘀𝘁 𝗶𝘀 𝘁𝗵𝗲 𝘀𝗵𝗼𝗿𝘁𝗰𝘂𝘁: The one thing that breaks the pattern is a brand the shopper recognizes and trusts. Show up as Nike in position five and people who trust Nike will pick you over the #1 item.

• 𝗧𝗵𝗲 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗯𝗲𝗵𝗶𝗻𝗱 𝗶𝘁: Five user-behavior studies over roughly 24 months, with 100+ participants and 400+ tasks. Every click, hover, and scroll was recorded, and participants talked through what they were thinking.

• 𝟵𝟲% 𝗰𝗮𝗿𝗼𝘂𝘀𝗲𝗹𝘀: Kevin analyzed 4,000+ shopping queries on regular Google (not AI mode). Over 96% showed at least one product carousel, and when one appears, click-through on classic blue links drops by at least 50%.

• 𝗔𝗱𝘀 𝗶𝗻 𝗱𝗶𝘀𝗴𝘂𝗶𝘀𝗲: Google shows more ads, bigger ads, and ads that are harder to tell apart from organic results. “I’ve been in this space for over 15 years. I more and more accidentally click on ads.”

• 𝗤𝘂𝗲𝗿𝘆 𝗳𝗮𝗻-𝗼𝘂𝘁𝘀: AI engines run live searches on anywhere from two to hundreds of keywords before they answer. Ranking better on those fan-out queries has repeatedly increased citation rates and AI visibility.

• 𝗠𝗲𝗿𝗰𝗵𝗮𝗻𝘁 𝗖𝗲𝗻𝘁𝗲𝗿 𝗼𝘃𝗲𝗿 𝗦𝗲𝗮𝗿𝗰𝗵 𝗖𝗼𝗻𝘀𝗼𝗹𝗲: Google’s Merchant Center now says more about how products show up in AI than Search Console does, which is why Kevin calls commerce the beachhead for AI optimization.

• 𝗧𝗵𝗲 𝗼𝘃𝗲𝗿/𝘂𝗻𝗱𝗲𝗿: For AI-assisted product research, Kevin takes the over on 10–20% of shoppers. For fully agentic checkout, he expects under 1% over the next 12 months.

Kevin is one of the best at separating what the data says from what the hype says. Subscribe to Growth Memo after you subscribe to Retailgentic, and enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Just when we thought the Agentic Commerce news peaked in September, October said - hold my Pumpkin Spice Latte! In October at Retailgentic we covered all the breaking Muse data and stories and then META’s Q3 blow-out, pivot to Muse: In this 2 part series called Deep Dive into Meta Muse (Part 1 is here and Part 2 is here). The Earnings shared the stage with the big Claude Duet news and Walmart’s new open-door policy for agents here. Meta’s earnings put us into a new gear, I call: Musemania - welcome to the start of the MuseMania era!

The other breaking news in October was the General Availability of Google’s Univeral Cart - it’s not only in AI mode, but I’ve gotten excited DMs from readers seeing it in their trad Google and I see it there as well - it’s a full-on agentic commerce Google take over for Holiday 26!

The other big October news came from the launch of our new podcast and companion substack, Adgentic featuring the amazing Kiri Masters. The stats for the first episode (snack size at only 13mins) are 🤯 - thanks for the support everyone. But remember, sharing is caring, tell all your colleagues that are interested in ChatGPT Ads, how RMN and AI are going to intersect, or even folks that deal with Google AdWords, Amazon Ads and TikTok paid media, about it because it’s all going to be impacted (hopefully positively, but could also be negatively - the topic of the first ep!) - so they need to track it.

Finally, our most popular post for October came from our in-house research into consumer shopping behavior on ChatGPT (here) . We partnered with the esteemed research firm Clickstream Solutions (led by Erik Van Buskirk). The TL;DR: ChatGPT Product cards drive 84% of the product choice influence. They basically behave very similarly to the Amazon ‘Buy Box’, but with a lot of Agentic Commerce twists along the way.

As if that wasn’t enough:

But Wait There’s More…

Those were the hair-on-fire, we have to alert people breaking news stories and important trends top-of-mind with merchants. One level down there were a ton of interesting stories - in fact we tracked over 40 that met our criteria we discussed internally. We’ve taken that, curated it down to the top stories we think you should track.

This month they fit nicely into three trends:

• October Trend 1: Holiday Agentic Commerce Shopping Gets Real in 2026

• October Trend 2: Are the platforms coming for the shopper AND the economics of the transaction? What’s this mean for retailers and brands?

• October Trend 3: Retailgentic readers have an edge on this one, but the rest of the World is catching on that Product Data and Measurement Move to top priorities

I would say another theme running through those three trends is we see more and more evidence (swirling together stated preference and observed preference datapoints to triangulate as is our practice) that the agentic-savvy shopper is moving faster than the merchant’s ability to measure, merchandise and serve the new journey.

Consumers are asking AI what to buy. Platforms are figuring out how to get paid. Meanwhile, retailers are trying to answer a deceptively simple question: “Are we getting any of these customers?” and “what are they worth?”

Sometimes the answer is yes, but the analytics don’t show it. Sometimes the agent can’t get through the front door. And sometimes the product never makes the consideration set because the catalog can’t answer the shopper’s question.

All three are fixable. Let’s dig in.

Two quick editorial reminders if you’re new or were forwarded this by a colleague:

1. Data: We’ll distinguish consumer surveys, forecasts and observed behavior. Someone saying they would let an agent buy a gift is different from an agent completing the purchase.

2. Vocabulary: We use Agentic Commerce broadly across Research → Find → Buy, including answer engines, personal agents and retailer shopper agents.

Subscribe now

October Trend 1: Holiday AI Shopping Gets Real 🎁

Here’s a summary of Holiday forecasts, reports and survey’s that had an Agentic Commerce angle.

The holiday research flood has arrived

Apparently everyone’s corporate calendar had the same reminder: “Release holiday AI research.”

Shopify’s US holiday survey give us a useful view of the delegation curve:

• 60% of consumers are likely to use AI for at least one shopping task.

• 48% are comfortable letting it recommend the right product.

• 40% are comfortable letting it build a cart, provided they complete the purchase.

• 31% trust AI to make a purchase on their behalf.

That US consumer sample was 2,000 adults surveyed in July. The broader research covered 18,000 consumers across nine countries, with a separate merchant survey conducted in August. These are stated behaviors and intentions, not checkout records. (Shopify research PDF, Shopify holiday research).

As personal agents gain more adoption, it will be interesting to see if the 60% ‘usage’ to the 30% ‘let it buy for me’ rise rapidly.

The wider holiday outlook adds several different perspectives:

• Adobe: Forecasts AI-referred traffic to US retail sites rising 130% year over year during the holiday season. This measures shoppers clicking through from AI tools. (Adobe)

• Accenture: Finds shoppers expect generative AI to influence 45% of category spending, with roughly four in ten planning to use it for holiday shopping. Its survey also puts the average amount shoppers would allow an agent to spend without approval at $102. (Accenture)

• Salesforce: Forecasts 20% of holiday ecommerce traffic originating from AI chat agents, but its definition includes consumer-facing bots, autonomous tasks and competitive scraping. Don’t read that as “20% of human shoppers arrive from ChatGPT.” (Salesforce)

• Practical Ecommerce: Predicts AI-referred shoppers will convert at least 25% better than shoppers arriving through other channels during peak season. That is a forecast, not the holiday result. (Practical Ecommerce)

Triangulate these datapoints together: AI has a meaningful role in holiday discovery, comparison and purchase consideration. The interesting question is how much of that influence makes it all the way through to ‘Buy’. We’ll only know the answer when the dust settles - stay tuned.

CivicScience: 37% adoption, and a BIG attribution clue

CivicScience launched an ongoing Agentic Retail Report, and its Q3 headline numbers are worth watching:

• 37% AI shopping adoption, rising for an eighth consecutive month.

• 61% of AI shoppers research on a platform and then purchase elsewhere.

• 48% name ChatGPT as their primary AI product-research destination, versus 21% for Gemini.

• Grocery and electronics each register 32% for AI-assisted shopping.

These findings come from ongoing surveys of opt-in US adults. They describe reported behavior rather than transactions observed directly. (CivicScience)

Keep that 61% in your head.

A shopper can discover a product through AI, compare it there, and then buy through a retailer’s app or website. The influence happened before the transaction, potentially before any measurable referral.

My takeaway: Merchant’s looking only at traffic labeled “ChatGPT” are missing a substantial part of the story.

Amex: merchants see the opportunity, but need help

Amex released a merchant playbook and new research that captures the readiness gap:

• 83% of surveyed businesses view agentic commerce as an opportunity.

• 41% report actively experimenting with or implementing it.

• 86% expect to need at least moderate support from payment providers and partners.

On the consumer side, 66% say they are likely to use agents for price research in the next year and 61% for recommendations. Comfort with delegated purchasing varies sharply: 52% for everyday, low-cost items versus 15% for high-value purchases.

The August research surveyed 502 US business leaders and 2,005 US adults who shop online. (Research announcement)

That last comparison is interesting. A shopper may welcome an agent’s help researching a refrigerator while keeping a much tighter grip on the Buy button.

Amex’s five-part playbook covers discovery, transactions, trust and standing out, alongside a new pilot merchant advisory council. That’s useful groundwork for moving merchants from “we should do something” to an actual operating plan.

→ Amex’s business playbook announcement

EMARKETER: don’t lose sight of Google’s distribution

With all the personal-agent excitement, here’s a chart that caught my eye.

EMARKETER’s August forecast projects Google AI Mode reaching 41.5% of US internet users in 2027, moving ahead of the other generative-AI platforms shown, and reaching 50.1% by 2030.

This is a forecast of monthly platform use, not shopping share or GMV (but platform use and shopping/GMV share have historically been very correlated).

My take: Google will work to bring Agentic Commerce into a behavior consumers already repeat constantly. That distribution advantage deserves attention even when another company has the more exciting demo that week.

For brands, the practical implication is to always prepare product information for multiple destinations. A strong presence in one answer engine leaves plenty of shoppers elsewhere.

October Trend 2: The Platforms Want the Shopper AND the Economics

Now let’s pivot to platform economics and the discussion there. Anchoring this point was Zuck on Cheeky Pint and earnings talking about Muse taking an un-specified % of GMV and the announcement of 10

TikTok adds a shopper assistant and UCP-powered Buy Direct

TikTok announced two connected pieces:

• Shopping Assistant: Answers product questions using merchant-provided information inside TikTok’s in-app browser.

• Buy Direct: Provides native checkout using saved payment and address information, while the brand remains the merchant of record.

Buy Direct requires a UCP integration. TikTok names commerce and payment partners including Salesforce, Shopify, Shoplazza and Stripe. (TikTok) Here’s a demo:

The interesting part is where the conversation starts.

A shopper sees a product, gets interested, then asks about fit, materials or availability. TikTok wants to answer those questions while the shopper is still engaged and give them a shorter route to purchase.

That puts merchant product information directly into the conversion path.

If your data can explain why those boots work for a particular hike, the assistant has something useful to say. If all it gets is a title, price and a vague paragraph about adventure…well, good luck with that.

Seven shopping agents, compared from the retailer’s side - Roger Dunn

Retailgentic friend, Roger Dunn, published a useful comparison of seven shopping experiences: Muse, Instinct, Dots, Gemini Spark, Cue, GrokBot and Copilot.

I liked that he approached it from the retailer’s side of the transaction. How does the agent pay? What arrives in the order system? Does the purchase stay connected to the shopper’s account?

His comparison shows different payment approaches, from virtual cards to merchant-stored credentials and checkout integrations. It is a snapshot of a fast-moving market, but a useful one. (Roger Dunn)

We tend to put all of these products in the same “agentic commerce” bucket. For an operations team handling orders, loyalty and returns, the differences can be material.

This one is worth forwarding to whoever owns ecommerce operations, alongside the person tracking the shiny new Personal Shopping Agents.

ChatGPT adds visual ads and keeps expanding

OpenAI introduced a visual ad format for testing during image generation on the Free and Go plans, with ads separate from the image being created. It also expanded conversion, attribution and incrementality measurement partnerships. (OpenAI)

Then came another geographic expansion: 26 additional countries, taking ChatGPT Ads to 89 countries. OpenAI says more than 100,000 advertisers have run campaigns since launch. (OpenAI)

The visual format is interesting, but the measurement work may matter more to the people approving budgets.

An advertiser needs a way to understand whether a channel helped acquire customers, what those customers cost, and whether the result was incremental. Connecting into familiar measurement systems makes that conversation easier.

I also noticed the grocery/CPG example. It’s another reminder that the commercial opportunity extends well beyond someone explicitly typing “buy running shoes.”

→ New formats and measurement
→ Expansion to 26 additional countries

Amazon’s sponsored questions bring ads into the shopping conversation

Our team also spotted examples of sponsored questions in Amazon’s shopping experience, including a live-search example shared internally.

The format is interesting: a question can invite the shopper into a product conversation at the moment they are evaluating alternatives.

For advertisers, that introduces another creative question. Which product question is most useful to a shopper considering this item?

For catalog teams, it introduces a familiar obligation: make sure the answer is accurate, specific and supported.

I’ll be watching how these placements develop and how Amazon measures them. A sponsored question could occupy a different role in the journey from a conventional sponsored product tile. Feels more top-of-funnel, but we’ll see.

The Wallet Wars: how do these agents get paid?

Two pieces are worth reading together: Tanay Jaipuria’s How Personal Agents Get Paid and Terry Angelos’ Wallet Wars.

Tanay lays out four possible revenue streams: subscriptions, convenience fees, merchant or partner fees, and advertising. He then explores how the economics might change as an agent moves from executing a purchase to generating demand or taking on more of an aggregator’s responsibilities. His take-rate scenarios are illustrations, not established industry pricing. (Tanay Jaipuria)

Terry focuses on the payment mechanics and makes a useful distinction: completing a payment does not necessarily preserve the customer’s original card benefits or merchant relationship.

For example, a wallet that charges the shopper’s card and pays the merchant with a different virtual card may complete the order while changing how card-linked benefits work. (Wallet Wars)

Follow the incentives.

An agent that helps a merchant acquire a customer has a different argument for compensation from one that simply executes an order the shopper already specified. And a shopper may consider rewards, returns and account history part of the task, even if the instruction was only “buy it.”

There is a lot of commerce detail and plumbing to get right hiding inside those two words.

October Trend 3: Product Data and Measurement Move to the Front

Forrester: product data needs to answer real shopper questions

Forrester’s Emily Pfeiffer and Tony Plec published a piece on the missing dimension in product data: helping shoppers understand real-world fit, use and outcomes, beyond conventional attributes. (Forrester)

This connects directly to what we’ve been discussing here as conversational attributes.

Consider a hypothetical carry-on bag. Dimensions are necessary, but a shopper may ask whether it works for a three-day business trip, fits a laptop and a second pair of shoes, or meets a particular airline’s limits.

Some answers can be derived from specifications. Others require additional evidence or product knowledge. The job is to make those answers available and accurate, without inventing capabilities the product doesn’t have.

For retailers and brands, I’d start with the questions customers already ask before purchasing. Which can your current catalog answer? Which require someone to consult reviews, manuals or customer-service history?

That gap is a useful enrichment backlog.

ChatGPT traffic goes dark, and the product shelf may be changing too

We’ve been seeing discussion in the time line about ChatGPT referral traffic declining or becoming harder to identify. We also were early on this calling it Dark Agentic Commerce Traffic (DACT) here back in April.

In our own team’s testing, Link noticed missing tracking parameters on some desktop links. He also reported fewer prompts returning product cards and fewer cards in some responses.

Those are observations from testing, not a quantified market-wide study.

There are two things to investigate here.

First, a tracking change can reduce reported referral traffic without an equivalent decline in shoppers. Second, a change in the interface or product-card frequency can genuinely change the opportunity to be discovered and clicked.

Both could happen at once.

My advice: before concluding that AI shopping has stalled, test the journey. Look at the recommendation, the link, the landing session and the transaction. Then compare that with what the analytics report says happened. Be sure to test all modalities - a lot of the ‘traffic tracking stripping’ comes from mobile app users and desktop users.

CivicScience’s (covered earlier) research-elsewhere, buy-elsewhere finding makes this even more relevant. AI influence and identifiable AI referrals are different measurements.

Block the agent, and someone else will get the order (aka Agentic Commerce is a zero-sum game)

Olivia Moore at a16z shared a real-world shopping example that makes the access issue very concrete.

In her account of a winter-coat search, Patagonia blocked Muse. The agent considered other retailers, and REI emerged as the preferred option, with accessible inventory, pickup and pricing contributing to the recommendation.

It’s one example, not a detailed conversion study. But it is an excellent prompt for a conversation between ecommerce and security teams.

A brand can be desirable enough to make the initial list and still lose the direct purchase opportunity if the agent cannot complete the journey there. The merchant won discovery, but lost the transaction (sad trombone sound 🪊).

Retailers need to make intentional decisions about agent access. They also need to know what their current systems are doing, because a default bot rule may already be making that decision for them.

Try a few real shopping tasks against your own site. Watch where the agent succeeds, where it gets confused and where it gives up.

That’s a much more useful starting point than another slide declaring the company “AI ready.”

That’s a Wrap

Across this month’s news, the same operational questions keep coming back:

• Can the agent find your products?

• Can it determine whether they fit the shopper’s needs?

• Can it verify price, inventory and delivery?

• Can the shopper or agent complete the purchase?

• Can you tell what happened afterward?

The holiday surveys suggest plenty of consumers are ready to try. TikTok, ChatGPT and Amazon are creating more places for those interactions to happen. The wallet companies are working on the transaction, and the advertising teams are working on the economics.

For retailers and brands, there is already useful work to do.

My Holiday 26 priority list: test your agent shopping journey, fix the product-information gaps, and check that your measurement can recognize the results.

We’ll learn a lot when the holiday data arrives. In the meantime, your next customer may already be asking an agent which product to buy.

Let’s make sure yours is not only considered, but wins the transaction! 🎁🏆

Subscribe now

Late Wed Oct 21 was one of the newsiest days in the young life of Agentic Commerce, and were digesting all of it. Today we’re going to walk you through the big picture impact of yesterday’s news. We’ll start with two appetizers - the Walmart and Claude news and then go deep on the big Muse news. Let’s dig in!

Subscribe now

Walmart Professes Their Love for Agents!

One theme that came out yesterday is a lot of energy suddenly forming around the idea of agents transacting with agents. Long-time readers will remember way back to April 2025 when Google launched an A2A. It started with Walmart’s big news.

Yesterday, Walmart announced here that they were making Walmart even MORE agent friendly.

There’s two parts to this Agent Profiles and the Agent Gateway.

Agent Profiles

Above you can see a screenshot of what this looks like - your agent can have it’s own profile at Walmart with the card you want used, the permissions(purchase levels you allow), the ship-to/bill-to address, etc. I imagine this will hang off your user profile.

Here, as an example of that duality, we see a purchase history shot→

The first item was ordered by a human directly, the second one (the shot is fuzzy, apologies) says “ordered placed by YOUR AGENT”.

Two entities, sharing one account - very cool!

Agent Gateway

Here’s how Walmart described this on the corp blog→

As news came out, it became clear that the vision here is ‘consumer’s agent talks to Walmarts agent for A2A type communication. This is a level above UCP/ACP because it includes more around your order history and you can imagine that the agent may want to ask some questions to a customer service agent. There’s a whole new protocol called PAP - Personal Agent Protocol, that was introduced recently by customer-service agent company Sierra. Details on PAP here. In Sierra’s launch of this protocol, both Shopify and Walmart are in the mix.

Noticeably missing from these logos are Google, ChatGPT, Grok and Claude. It’s still early, let’s see where this one goes.

Claude Enters The Agentic Commerce Arena with Claude Duet, But There’s An Important Caveat…

Unfortunately Duet is in a waitlist preview mode right now so we can’t show it to you live yet, but enough details have dribbled out we can put together how this is going to work and walk you through it.

The first hint we were going to see Agentic Commerce is from the blog post here→

You can imagine, that caught my attention. We dug through and found this post by Stripe that shows a video of the experience→

Then Shopify posted two videos -one was showing Duet setting up a new Shopify store and adding 4 products, then they showed one where Duet buys from the store and uses Shop Pay→

But wait, there’s a catch: Duet is B2B vs. B2C?

Looking at all the early Duet reviews, one thing becomes crystal clear, Duet is not a consumer-oriented personal agent like Muse or Instinct. It doesn’t even seem to be designed to have one foot in both consumer and business worlds like GrokBot and ChatGPT Dot. Instead, it feels squarely focused on the SMB mainly, but business-user generally.

The starting user-interface is extremely clever - it’s a TO-DO list that your Duet works with you to knock things out, add things from your various communication channels, and connectors.

Even in the stripe example, you see an SMB use case, where Shopify went straight for someone buying a handbag.

We’ll see if there’s some leakage over to to consumer, but for now, this seems a bit late to the game and not consumer at all, so we’ll see.

META’s Q3 Earnings: Welcome to MUSEMANIA!

Meta reported Q3 2026 earnings last night and the headlines were all about the beat-and-raise quarter ($64.6B in revenue, +26% year over year).

But on the conference call and in the Wall St notes published later, the real story was the amazing start and enthusiasm that Meta and Zuckerberg have for Muse only 43 days since launch.

MUSE by the Numbers Revealed in Q3

Here’s everything Meta disclosed on the call:

• 10M+ people in the US and Canada set up MUSE personal agents in the first ~6 weeks, with the app at #1 in the App Store “almost every day.”

• ~25% use it on any given day - and retention improves the longer people use it. Zuck said that’s more than twice the level Threads started at.

• Shopping is the #1 use case. More than a quarter of weekly actives shop through MUSE, with an average purchase size above $200 (and yes, that includes travel - Mahaney asked, and Zuckerberg confirmed travel is itself a top use case).

• The commerce plumbing strained under the load: Zuckerberg said Plaid told Meta it had never seen connector activity like this on any other service, and in week one Meta had to work with Stripe to raise credit limits because volumes exceeded everyone’s expectations.

• Beyond shopping, the top connectors are personal productivity and linking MUSE to email and calendar, and businesses are wiring in too - connector applications have doubled from ~1,500 since Meta Connect (per Morgan Stanley), with >30% of MUSE users having connected at least one service.

• The business model, in Zuckerberg’s own words: MUSE stays free for most people, and Meta will “build commerce tools so we get paid a fee based on the transactions that we help facilitate. We expect this will be paid for by businesses.” A take rate, paid by merchants.

• Scale ambition: capacity to serve 1B+ people within a few years, en route to “billions of Muses in a network.” MUSE is compute-intensive today, with a >10x cost-efficiency target over the next year before monetization scales.

Early SWAG Math on Meta’s GMV:

In take-rate based businesses, like marketplaces, we have this formula:

GMV * Take-rate = Revenue

We now have three things to track - the amount of transactions going through the engines, for payments that’s called TPV or Transaction Processing Value. In marketplaces GMV stand for Gross Merchandise Value - the value of the overall purchases, and then a smaller number (e.g. about 15% for Amazon 3P) is the take rate so for every $100 of GMV, the Amazon revenue is $15, the seller, or merchant keeps the other $85 as their revenue.

We now have enough tea leaves in the Q3 disclosures to do a very rough swag on GMV:

• Fact: 10m MAU * This is 6wks but it was scaling up from 0, it’s sticky so feels high trailing, low forward - we’ll use it as a middle-point

• Fact: 2.5m DAU

• Unknown: WAUs - We know it’s got to be between 2.5m and 10m - let’s call it 7.5m on the high side of mid-point (again Zuck said it’s sticky)

  • Best guess: 7.5m WAUs

• Fact: Zuck said that 25% WAU are shopping (w travel) for $200 sales each user

That gives us 7m WAUs 25% $200 GMV (over that week). 7m * 25% * $200 yields $350m/GMV/week.

That doesn’t seem very high, but the deceptive thing, and it’s not clear if Zuck meant $200 ‘in the life of the user’ or ‘over that week’ - I’m assuming that Zuck is hyper-aware of the differences between DAU/WAU/MAU and how to align metrics to those things.

If it’s what I’m assuming that’s $1.4b/month or $18b/yr GMV run-rate.

To put that in perspective, from 2025 year-end data:

• Amazon is at ~$575b/yr (marketplacepulse estimate) for their marketplace

• Shopify is at $378b/yr

• eBay is at $80b/yr

• TikTok is at ~$65b/yr

Could Meta have created a marketplace opportunity that is already 20% of eBay’s GMV in 43 days?

We know that Instinct said they are at $1B/yr run-rate, so that’s out there too.

We’re hearing industry whispers that payment providers are shocked by the amount of transactions coming through these agents now and the ‘stripe had to raise credit limits’ bit and the ‘Plaid hasn’t seen anything like it’ disclosure are other indications.

Shopify reports November 3rd, maybe we’ll get some clarity from them on what they are seeing.

Regardless, the details provided by Meta on the call and the body language started what I’m going to call Wall St’s MuseMania fever…

Wall St. Catches MUSEMANIA!

I pulled the MUSE highlights from four analyst reports (Baird’s Colin Sebastian, Evercore ISI’s Mark Mahaney, Morgan Stanley’s Brian Nowak, and Citi’s Ronald Josey). Three of them raised their price target, and every one of them anchored the bull case on the same thing we obsess about at Retailgentic HQ: agentic commerce. One of them even coined a term for it that I suspect will stick.

Most importantly, all four put MUSE at the center of the thesis. Here’s the highlights from their research:

Mahaney (Evercore) coined the phrase of the quarter: “Agentic Intent.” His point: Meta built one of the great ad platforms on reach, frequency, and data - but it never had true commercial intent the way Google has (1P data either). With shopping as MUSE’s #1 use case, that changes. He also flagged management’s belief that MUSE US ARPU can eventually surpass Meta’s US ARPU - effectively a doubling-plus from his estimated $350-400 today - and connected the dots to Meta’s newly disclosed 500M self-identified small businesses: who answers all those MUSE agent requests? Business agents.

Josey (Citi) called MUSE “a search tool” - loaded words for anyone tracking the Google comparison. Citi’s Sensor Tower work shows 11M+ US app downloads in six weeks, US DAUs above ~2.7M and climbing, and a download curve that outpaced the launches of ChatGPT, Gemini, and Meta AI. His sharpest commerce stat: ~20% of MUSE users generate $1K+ in annual Family of Apps ARPU. Citi sees monetization sequencing from subscriptions to transaction fees within the next year, then enterprise, and ultimately advertising.

Nowak (Morgan Stanley) focused on the ecosystem wiring: >30% of MUSE users have connected a service, and connector applications from businesses have doubled from ~1,500 since Meta Connect. He reads the high commercial intent as encouraging for long-term monetization given Meta’s existing relationships with ~500M businesses - and his $1,000 bull case kicks in if MUSE users scale faster than modeled. (He’s literally built a MUSE cost-to-serve model for clients that is really amazing.)

Sebastian (Baird) laid out the monetization stack: subscriptions are the only MUSE revenue stream today, with management planning to add a transaction take rate - timed, he notes, to coincide with the strong shopping momentum - then business agents, and over time, advertising. Baird expects initial contributions from Meta’s new AI businesses in 2027.

The skeptical thread running through all four: compute economics. MUSE is expensive to serve, Meta deferred its 2027 capex guidance, and everyone is watching whether the targeted >10x cost-efficiency improvement arrives before the take rate does.

Agent-to-Agent Commerce Gets Real

What Meta’s Q3 means for retailers and brands:

The intent layer is moving. When 25%+ of a 10M-user agent’s weekly actives are shopping at $200/week six weeks after launch, product discovery is happening inside an agent, not on your site or a search box. Mahaney’s “Agentic Intent” framing is the right one: the scarce asset in the agentic era is knowing what the consumer wants right now, and MUSE is capturing it at a pace that surprised even Meta’s payment partners.

The Muse business model is a take rate on transactions, paid by businesses (aka the marketplace business model, but with the merchant-of-record twist!) Zuckerberg was unusually explicit: free for consumers, with Meta getting “a percent of the value that we are helping to create.” And he said monetization will come sooner than Meta’s usual build-audience-first playbook, both because of compute costs and because “such a large part of what people are doing is commerce.” If you sell online, a new marketplace-style fee line is forming - attached to an agent, not a website.

The protocol fight is on. On the call, CFO Susan Li discussed the PAP (there it is again) “attempting to navigate every digital experience like a human.” Li expects the market to evolve away from screen-scraping agents toward authenticated APIs and approved agent partnerships.

Agent-to-agent is the stated end state. Zuckerberg’s “Muse network” vision: buying something isn’t navigating a retailer’s website, it’s your agent talking to the business’s agent. Meta disclosed 500M self-identified small businesses on its apps and is already shipping MUSE skills and connectors for SMBs (Shopify, QuickBooks, Canva, ad accounts). The supply side of agentic commerce is being recruited right now. I suspect we’ll get some kind of a merchant agent out of them soon.

What this means practically: the questions we ask in these digital pages every week about AI shopping readiness - can agents access your catalog, is your data structured for them, are you present across engines - now extend to a new one: is your business ready to be transacted with by an agent, under a protocol, with a take rate attached? That’s not a 2030 theoretical question any more. Stripe raised credit limits in week one.

Bottom line: There’s Agentic Commerce Substance to MUSEMANIA!

Six weeks in, MUSE is the fastest-adopted consumer AI launch on record by Citi’s download data, shopping is its #1 use case, and the four analysts covering the quarter all moved their targets up with MUSE as the reason. The monetization sequence is now public - subscriptions today, a merchant-paid transaction take rate next (Citi says within a year), enterprise and advertising after that - and the Personal Agent Protocol puts Meta, Walmart, Shopify, Stripe, and Sierra at the same standards table.

What we’re watching for going forward:

• Whether the take rate gets a number (best guess somewhere between 5-10%)

• When/how fast MUSE goes international (Zuckerberg called the US/Canada-only footprint the biggest constraint on growth)

• Which retailers show up next on the PAP partner list.

• When do the retailers with connectors launch and what does that look like - is it PAP or UCP or yet another protocol in there?

Six weeks ago Meta was flirting with agentic commerce was our thesis. As of this earnings call, it’s Meta’s business model going forward for the company.

Subscribe now

Subscribe now

Oscar Rundqvist is the CEO and founder of eComID, a Stockholm-based company that builds AI modules for brands’ own sites: sizing, AI search, personalization, an on-site agent and, soon, virtual try-on. All of them run on one portable “e-commerce ID.”

Before eComID, Oscar led digital customer experience at H&M. Working there, he saw up close how broken online shopping is, especially when it comes to returns. He pitched the idea internally, and H&M’s startup studio funded eComID from day one.

Listen/Watch Our Interview:

This week is a product deep dive with a fresh funding story attached. eComID just closed Europe’s largest fashion tech seed round, and the company is building something agentic commerce needs: shopper context that belongs to the shopper and goes wherever they shop. The ID was born inside H&M, it already runs on about 70 brand sites, and it’s now headed for the US.

Highlights

• 𝗧𝗵𝗲 $𝟭𝟳.𝟰𝗠 𝘀𝗲𝗲𝗱 𝗿𝗼𝘂𝗻𝗱: Europe's largest fashion tech seed round. Oscar says it's the second-biggest in the world, led by London's Systemiq Capital, and Nordic sports retailer Stadium (one of eComID's biggest customers) joined as a strategic investor. Scot rang the gong 🔔

• 𝗕𝗼𝗿𝗻 𝗶𝗻𝘀𝗶𝗱𝗲 𝗛&𝗠: Oscar pitched eComID internally in a Dragon's Den–style process and H&M's startup arm invested from day one.

• 𝗔 𝘃𝗲𝗿𝘆 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝘀𝘁𝗿𝗮𝗻𝗴𝗲𝗿: Models are very capable, catalogs are open to agents and payments are getting solved. But without shopper context, AI "still starts from zero." It doesn't know your size, your taste or what you sent back.

• 𝗧𝗵𝗲 𝘀𝗵𝗼𝗽𝗽𝗶𝗻𝗴 𝗽𝗮𝘀𝘀𝗽𝗼𝗿𝘁: One source of truth for your size, taste and preferences, tied to your email. It goes with you across eComID brands like H&M, Björn Borg, Peak Performance and Haglöfs. "We want to democratize personalization."

• 𝟵𝟭% 𝗯𝘂𝘆 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝘀𝗶𝘇𝗲𝘀: 91% of shoppers buy different sizes from different brands. Oscar says that shows how broken sizing is.

• 𝗧𝗵𝗲 𝘁𝗿𝗮𝗰𝘁𝗶𝗼𝗻: About 70 merchants, about 20 million shoppers reached each month, and 15% of users already in the US. Shoppers who use eComID have about 30% lower return rates and about 10% higher conversion.

• +𝟮𝟲% 𝗿𝗲𝘃𝗲𝗻𝘂𝗲 𝗽𝗲𝗿 𝘃𝗶𝘀𝗶𝘁𝗼𝗿: In a head-to-head test, eComID's AI search beat traditional on-site search by 26% on revenue per visitor.

• 𝗞𝗶𝗹𝗹 𝗮𝘃𝗲𝗿𝗮𝗴𝗲 𝗼𝗿𝗱𝗲𝗿 𝘃𝗮𝗹𝘂𝗲: "Average order value is a completely useless metric." What matters is average kept order value, because each handled return costs brands $21–46.

• 𝗧𝗿𝘂𝘀𝘁 𝗶𝘀 𝘁𝗵𝗲 𝗽𝗿𝗼𝗱𝘂𝗰𝘁: If an agent spends $1,000 for you, you need to know whose agent it is, what you authorized, what context it used and what happens if something goes wrong. Oscar argues Europe's GDPR-first approach suits agentic commerce.

• 𝗦𝗵𝗼𝘄 𝘀𝗵𝗼𝗽𝗽𝗲𝗿𝘀 𝘁𝗵𝗲𝗶𝗿 𝗿𝗲𝘁𝘂𝗿𝗻 𝗿𝗮𝘁𝗲: 92% of shoppers want to see their own return rate, and 75% say seeing it makes them want to return less. The live demo on Houdini Sportswear shows the search, sizing and nudge modules working together.

In the last few minutes, Scot shares his screen and goes to Houdini Sportswear, an outdoor and ski brand he says he’d never visited before. Oscar walks him through eComID running live on the site, in this order:

1. AI search / on-site agent: a chat window at the bottom of the page. It offers recommendations before Scot has typed anything, and it has clickable suggestion buttons to keep the conversation going. Oscar says this is the search that beat traditional search by 26% on revenue per visitor.

2. Sizing module: sits under the size selector on the product page.

3. Nudge module: sits below the product description and tells shoppers about returns. Oscar uses it to explain that 92% of shoppers want to see their own return rate and 75% then want to return less.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Guide to this Meta Muse Deep Dive Series

This is Part 2 of a 2 part series:

• Part 1 (here) We reveal the 5 Muse Optimization Best Practices you can implement today to optimize your products to maximize discovery in this new personal agent

• Part 2: (YOU ARE HERE!) What’s all the buzz about? In Part 2 we look at the various signals that Muse is catching on with consumers and look at detailed usage data.

Before We Jump In: Announcing our new Podcast: AdGentic!

One area of Agentic Commerce we find very interesting that cuts across all the ACO surfaces (Answer engines, Retailer shopping agents, Personal Agents and on-site agents) is the intersection of pre-AI ad models, AI and Agentic Commerce. The opportunity is huge, the disruption is even bigger. What’s the latest news, technological advances and how are brands, retailers and publishers thinking about this.

We’ve partnered with Kiri Masters who will be our primary launch contributor. Hear directly from Kiri her vision for AdGentic→

If you are interested in tracking the intersection of Advertising, AI and Agentic Commerce, subscribe to the newsletter and podcast (audio and video available) today:

• Substack Newsletter: here

• LinkedIn Newsletter: here

• Apple Pod: here

• Spotify: here

• All other podcast apps: here

• YouTube: here

Keynote Video and Presentation

Our Q3 ‘State of Agentic Commerce’ presentation was quite popular, so when time allows we’ll be supplementing these more strategic series with a companion video and PDF. This allows you to consumer the content in audio/video and many readers ask for high-res copies of the images for integration in their board/management team decks. We’re happy to provide both and had the opportunity to put both parts 1 and 2 together into one ~30min presentation here:

Slides are here→

Muse Deep Dive Keynote 10_20_26

4.36MB ∙ PDF file

Download

Download

Detailed Muse Data Deep Analysis

Even though we’re super early in the life of Muse, if you’re obsessed about data, like we are here at Retailgentic HQ, there’s a goldmine of early indications of what’s going on at a Macro and Micro level with Muse. As is our tradition at Retailgentic, let’s go deep! 🪏

Muse Audience Trends

We have two sources of data for Muse’s audience trends. First up we have downloads (this is worldwide, probably iOS/Android and mobile app only - no mobile web, no desktop app)→

Here you see an early surge, followed by consistent 150k/day downloads through launch to Oct 6th.

SensorTower also has download data, but compared to the data above, tracks cumulative downloads and aligns that trend to the launch of ChatGPT for an apples-to-apples comparison:

About 9 days in, Meta must have decided they had a hit on their hands and started pushing paid advertising and also flexing the distribution of their massive network as illustrated in the next datapoint. The SensorTower data shows that as of Day 22, Muse was at ~5m downloads which was more than 2X ChatGPT at the same number of days after launch.

Meta Flexes its Distribution Muscles

Meta has 5 properties with over 1B users: Facebook (>3b), Instagram (>3b), WhatsApp (>3b), Messenger (~1.3b) and Meta.ai (~1b). On top of THAT they have a bunch of tier-2 properties like Threads (500m) and FB Marketplaces (not disclosed separately but widely known to have over 1b shoppers). Swirl all that together and Meta has an extremely large footprint to launch a new service into.

The SensorTower data below illustrates the in-house property being promoted as a % of impressions across the network:

Here you can see that around 9/17, Meta started turning up the heat on Muse, pointing the distribution firehose at it to drive adoption.

The Information has reported citing anonymous internal sources that Mark Zuckerberg rushed the launch of Muse out of concern that Instinct was rapidly gaining traction, inferring a desire to get out both offensively to get in the game and compete and defensively to slow it’s exponential adoption curve.

Note: Because Instinct’s primary interface is iMessages, it is not in any clickstream data, we’ll have to rely on stated preference surveys for anything on that.

Early Retention/Churn Curves

On X, there’s a popular VC/data analyst, Deedy Das that focuses in on the ‘smile curves’ also known as user retention/churn curves of Answer engines like ChatGPT and Claude.

He analyzed early SensorTower data→

And came to these conclusions→

That’s shocking, especially when you consider that it’s a v1 battling it out against much more mature ChatGPT….and ahead early, but behind slightly at D14! More on how people are using Muse that is driving such high levels of activation and engagement already coming later in the post. Nobody, and I mean nobody, knows how to growth hack like Facebook, they created the term. I can assure you they know exactly how to get these curves to ‘smile’ - the holy grail of growth.

Wall St Analyst Perspectives

First, ‘friend of Retailgentic, Ron Josey, internet analyst at Citi, released an epic 54 page super deep analysis that we highly recommend spending time with. Some highlights:

• TAM - They calculate that Muse has a $4T TAM

• Muse Revenue Potential - If you take that $4T and assume Muse gets 4% quickly and can achieve a 5% take-rate, that’s $8B potential revenue.

Retailgentic Wall St friend Mark Mahaney, the internet analyst at Evercore ISI. On 10/12, Mark put out a an epic 24 pager all about Muse. In there they have a very detailed analysis of the cost of Muse (~$100/yr) and note that across the Meta network, they generate $380/user. They then math out how with a % take-rate type model, Meta could get Muse to break cover the $380+100=$480 through driving GMV * take-rate. For example at a 5% take rate it would take $10k/yr in GMV to clear that hurdle, not crazy when you look at the metrics coming out of Instinct and remember that includes travel, tickets and other non-physical good transactions.

Both Citi and Evercore are extremely enthusiastic about the user experience, the incremental revenue opportunities and where Muse is going even though it’s early.

HUMAN: Muse Agent-traffic Level Trends

September 28th, infrastructure-level security company HUMAN, released an analysis of AI agent traffic focused on what they are seeing from Muse. This is all based on traffic observed by HUMAN.

Even though Muse is starting with a small audience, they are already generating 72% of agent requests and Muse’s volume is 5x the bot request volume of ChatGPT’s agent.

The implication that either Meta is a ‘chatty’ agent (lots of bot traffic per comparable task) or consumers are using it for a LOT of tasks compared to ChatGPT (or both!). Put a pin that question.

The other interesting analysis HUMAN provided is tracking how Meta’s bot(s) are moving through sites. 84% of the requests are shopping-oriented hitting product and search pages. It’s also 25% more likely to make it to checkouts compared to other agents. The implication is a bit of reverse engineering that Muse users are doing a lot of shopping and going through to completion (checkout).

If you want to learn more about Muse, I strongly recommend watching Cheeky Pint, it features John Collison (President, Stripe), Mark Zuckerberg (CEO, Meta) and Alexandr Wang (CAIO, Meta Superintelligence Labs). I bring it up here, because in it, there are a couple of revelations that make sense to cover at this point:

• Zuckerberg talks about the business model combining some subscription/token usage with % of sales. Note he doesn’t mention ads which is interesting and if you’ve used Muse, there’s not really an easy place to inject an ad without it being disruptive to the ux. Collison casually drops this in there:

If we tie this together with the HUMAN information about checkout activity, we’re getting our first inclination that Personal Agents could be driving some real transactional volume.

Note that ‘reaching checkout’ does not mean 100% of consumers checked out - personally when I have my Muse heading into checkout, I’m all the way down the funnel and personally have completed 100% of those situations. As more data comes out, this is where we’re really focused.

Gener8’s Muse Panel Data

Gener8 is another ‘friend of Retailgentic’ - they have a huge panel and gather detailed clickstream. Note for this dataset it is an Android user data set.

On October 9th, they released a very detailed Muse report, you can download here.

Some of the highlights and key learnings follow.

Gener8: Adoption Rate

According to Gener8’s analysis 11% of the US adult population used Muse in the first month. There’s approximately 270m US adults, and 11% of that is 29.7m which is way off from the download data. My best guess on this misalilgnment:

• This is an android audience

• It’s well known that the android audience has an older and value-oriented bias

• Muse has two value props: value-orientation and convenience. The value-orientation messaging has really resonated with the android audience.

Gener8 estimates that conversely ChatGPT is used by 26% of US adults, putting Muse’s usage already at 42% of ChatGPTs after 30 days!

Gener8: Muse vs. ChatGPT Prompt Volume

That’s uses. Gener8 then looks at the prompt per day for Muse which is at 9.2 and ChatGPT at 5.9 - Muse users are prompting 55% more than ChatGPT users.

Gener8: Muse Demographics

This was the most surprising analysis, the demographics:

Usually the demographic curves for Answer Engines skew younger - what we call Zillennials - GenZ and Millennials, but here (table on the left) you see the older generations of Gen X and Boomers are leading the charge. Perhaps it’s the blue Facebook audience getting there before the Instagram’ers (or the Zillennials are on TikTok?)

Gener8: Muse Usage Types

Gener8 analyzed the clickstream data and put the various Muse activities into buckets. Shopping is the highest life-admin activity at 8.6% (Note ChatGPT has disclosed that 20% of their prompts are shopping related)

Gener8: Muse’s Top Mentioned Brands

Within those shopping prompts, Gener8 saw Consumer Electronics brands as a group with the highest share at 2.7% and CPG came in at 1.5%. The brands that came up in prompt results the most (kind of a Muse visibility metric) are detailed.

Conclusion

When we swirl all this together we start to see a picture:

• Muse has had one of the fastest consumer launches ever, based on a short first 42 days.

• While they have leveraged the massive Meta distribution advantage and layered ads on top, there’s a ‘there there’.

• The engagement/stickiness curves are as good/better as ChatGPT has now

• The prompt volume is through the roof

• Muse’s agents are going crazy across the internet swamping previous bot sources

• The Muse bots are hammering shopping category and PDP pages and then finding their way to checkout

• Stripe saw payments go Vertical in September (probably a double whammy from Muse and Instinct both hitting their growth curves simultaneously). Reminder: Instinct’s founder disclosed they are at $1B GMV annualized and growing 10% m/m.

• The demographics skew older than we would have anticipated

We’ll be watching closely for more information and probably be back in 15-30 days with an update. We do expect Meta to discuss Muse on their 10/21 Q3 earnings call and I can virtually guarantee given Mark and Ron’s enthusiasm and what I’ve heard from others, that a large % of Wall Street questions are going to be around Muse.

If anything important drops, we’ll let you know - stay tuned!

Subscribe now

Subscribe now

Justin Bomberowitz is Senior Director of E-commerce at Wilde Brands, the company behind chips and crackers made from chicken breast. Justin has made a career of joining challenger snack brands right as they’re ready to break out, with stints at SmartSweets, PopCorners, and Ocean Spray, plus consulting work for giants like P&G and Mondelez. These days his job reaches well beyond e-commerce, into operations, media, and the full digital path to purchase. He’s also the person at Wilde building the AI agents.

This episode was recorded live at the RetailClub AI Festival in Huntington Beach (Sept 22–24), where Jason Goldberg and I set up shop to record with retail and e-commerce leaders between sessions. You’ll hear a few nods to the festival in the conversation, and if there’s a little extra crunch in the audio, blame the snacks.

Listen/Watch Our Interview:

We start with a quick tour of every CPG brand Justin has worked for, and the very scientific method he uses to pick the next one. From there it turns into a crash course in how challenger brands actually grow: the brutal economics of the physical shelf, why big brand houses are reshuffling their portfolios, and what it really takes to get into Costco and stay there.

There’s also a snack reveal and a crunch our audio engineer will never forget. Hit play.

Highlights

• 𝗧𝗵𝗲 𝘀𝗵𝗲𝗹𝗳 𝗶𝘀 𝘇𝗲𝗿𝗼-𝘀𝘂𝗺: “We haven’t added any square footage into a shelf.” When an emerging brand earns its way in from the innovation aisle, something else has to come out.

• “𝗜 𝘄𝗮𝘀 𝘄𝗿𝗼𝗻𝗴”: A year ago Justin would have said TikTok Shop would never be a main part of his job. Then came “a tsunami of interest,” and today Wilde runs live selling on TikTok Shop and Whatnot.

• 𝗧𝗵𝗲 𝗖𝗼𝘀𝘁𝗰𝗼 𝘁𝗵𝗿𝗲𝘀𝗵𝗼𝗹𝗱: Landing a Costco rotation is the moment brands post the hype video. If you miss the sales threshold, you don’t get a repeat, and the regional buyers talk to each other.

• 𝗦𝗽𝗲𝗲𝗱 𝘀𝗲𝗹𝗹𝘀: “Conversions go up the faster you get delivery.” Justin says ship-from-store is the fastest-growing part of Walmart digital, and if your product isn’t in that store, the conversion you paid Walmart Connect for disappears.

• 𝗘𝘃𝗲𝗿𝘆 𝘀𝗮𝗹𝗲, 𝗲𝘃𝗲𝗿𝘆 𝗭𝗜𝗣 𝗰𝗼𝗱𝗲: With two years of Amazon sales mapped by ZIP code, Justin can see that barbecue over-indexes in the Southeast and sea salt and vinegar in the Northeast. With AI, building creative for those regions is now quick.

• 𝗢𝗻𝗲 𝗰𝗲𝗻𝘁𝗿𝗮𝗹 𝗻𝗲𝗿𝘃𝗼𝘂𝘀 𝘀𝘆𝘀𝘁𝗲𝗺: Instead of building agents on top of separate data silos, Wilde built a data lake. Target store data, Roundel media, Shipt, Instacart, and DoorDash data all sit in one place, so a sales lead can ask a small language model what to do next.

  𝗧𝗵𝗲 𝗽𝗿𝗼𝗺𝗼-𝘄𝗮𝘁𝗰𝗵𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁: When a retailer’s promotion drops a price on Instacart, one of Justin’s agents spots it through the API and flags it. He then decides whether to push ads to new-to-brand shoppers or pause them for 30 days and let the promo do the work.

• 𝗨𝘀𝗲 𝘁𝗵𝗲𝗺 𝗮𝗹𝗹: Justin builds mostly in Claude but stays LLM-agnostic, trying Grok, Muse, Kimi, and NVIDIA’s open models. “Don’t let your frame just get put into one.”

• 𝗗𝗼𝘇𝗲𝗻𝘀 𝗼𝗳 𝗮𝗴𝗲𝗻𝘁𝘀: “I have dozens of agents that I’m working with. That is the new norm.”

I learned more about how CPG actually works in 30 minutes with Justin than I have in years of nodding along when Jason explains it. Grab a bag of something crunchy and enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Background

As a reminder of Google’s Agentic Commerce timeline, here’s the timeline

• January 11, 2026- On stage at the NRF Big Show, Sundar announced UCP (our coverage here)

• May 19, 2026 - At Google I/O, they announced Universal Cart (our coverage here), and says will roll out in ‘Summer’.

• June 19, 2026 - Retailgentic was the first to find Universal Cart live in the wild in an early user experience test (coverage here with video)

• October 15, 2026 - Universal Cart is Live!

The Launch of Google’s Universal Cart

This morning in a blog post (here), Google’s Senior Director, Product Management, Consumer Shopping, Suresh Ganapathy announced that Universal Cart is officially available to all users in the US. Since the early experience I discovered in June there have been many enhancements to the user experience, let’s dig in.

Feature 1: Multi-Item, Multi-Merchant Cart

This video demonstrates the capability..

by adding a Steve Madden hand bag and a Quince sweater - two very different merchants, two different items - one Universal Cart.

Feature 2: Price Drop and Inventory Alerts

Not only does Universal Cart allow you to add items to the cart and checkout (we’ll get there soon), it also acts as a type of ‘wishlist/product tracker’. You are alerted whenever the price changes on something in your cart:

This BFCM, this could be the killer consumer agentic commerce feature. The shopper can add all their gift purchases in one cart and track the deals across all the Black Friday→ Cyber Monday promotions and pull the trigger when they get to the price they want.

Not pictured, let’s say that you want a Nintendo Switch 2 as a gift, but it’s sold out - you can add to Universal Cart and get an alert when it’s in stock.

Unfortunately, Cart stops there today, requiring human interaction. Conversely, Amazon’s AFS is ahead on autonomy - the Auto Buy feature allows you to both set an inventory or price feature and when your ‘trigger’ is met, automatically buy the item. Should they choose to go this far, Google has all the pieces now to provide similar functionality.

Feature 3: Loyalty and Credit Card Features

Here we have some Ulta products in the Universal Cart - the cart conveniently tells us about any loyalty offers, promo codes and credit-card benefits. In the Retailgentic Five Levels of Agentic Commerce, we talk about Level’s 4 and 5 giving the consumer a 360-degree view of their options across all the dimensions they may consider as part of the purchase: product features, price, merchant, loyalty, payment benefits, shipping times, etc. Google’s Universal Cart is ahead of the entire pack of answer engines, personal agents and even retailer agents.

Feature 4: Universal Cart Checkout

Finally, at the time of checkout, Universal Cart allows the consumer to checkout the items they now want to purchase and they have two options: checkout on the merchant’s site, or checkout ‘off site’ via the Universal Cart and Google Pay:

Here we see the consumer purchase the Steve Madden bag right in the cart, then checkout 2 Ulta items by ‘transferring’ the cart over to Ulta.com for completion.

Accessing Universal Cart

Universal Cart is available in the US in AI Mode. When I was in the early experience group and did Traditional Google product searches that invoke the Google Shopping widget, I was given the Universal Cart in that mode too. Later today we’ll post a live demo of the ‘official’ version of Universal Cart and look for this behavior in the production version.

Subscribe now

Google’s When to Buy - Holiday Gifting Guide - Data-Backed Tips for Optimizing Holiday Shopping

Also this morning, in a blog post (here) Google announced their first ‘When to buy guide’. The guide researched the bazillions of historic datapoints in the Google Shopping Graph and has tips surfaced from that data for the big gift categories:

My personal favorite insights:

• Pet supplies - lock down your dog and cat items on Black Friday, but speciality pets, wait for late December for the best deals you snake, turtle and ferret lovers.

• TVs - Black Friday is go time - load that 65” OLED beast in the Universal Cart and get ready!

Sadly Cyber Monday is losing some of it’s juice, as it doesn’t get a call out that I saw.

That’s a Wrap 🎁

Only 42 days left until Thanksgiving, last year in the 2-3 days leading up to Thanksgiving we had ChatGPT launch a Shopping Research (pieces worked their way into GPT-6 Dynamic UI for products recently), Target launched their ChatGPT app and Perplexity&PayPal launched their short-lived ‘Instant Buy’ functionality (on Thanksgiving Eve) - so we have lots of time for more big news to break. The best way to follow it all? Yep you guessed it 👇

Subscribe now

Happy 1 Month Birthday Muse!

Meta Muse has only been out 35 days, having launched on September 8th. In that time, it’s become top of mind with brands and retailers as the buzz has surged and we’ve got a whole new ‘agentic channel’ to think about this close to BFCM.

Since then, we’ve been poking and prodding at Muse to figure out how it works. Good news, we have some conclusive proprietary findings that we’re sharing today that provide actionable and clear best practices for optimizing your products to show up and be recommended by Muse.

Subscribe now

Guide to this Series

This is Part 1 of a 2 part series:

• Part 1: How to optimize your products for Muse - We reveal the 5 Muse Optimization Best Practices you can implement today.

• Part 2: What’s all the buzz about? In Part 2 we look at the various signals that Muse is catching on with consumers and look at detailed usage data.

Usually in Retailgentic, we build up to a conclusion. Given the importance of this one and the fact we’re only 44 days out for Thanksgiving, we decided to flip the script and start with the conclusion first and then show you how we got there, then why it matters in reverse.

Introducing the Five Best Practices for Muse Optimization

Our deep analysis of how Muse works produced five clear best practices for optimizing your products to be discovered by Muse. Here are the five best practices, the rest of the post shares our findings that led to and support each of the five best practices:

1. Catalog - Retailgentic proprietary research discovered that Muse pulls its catalog primarily from the Meta.ai product catalog. There are three other sources of catalog data: Shopify Catalog API, UCP and crawling. The bulk of SKUs are sourced through the Meta.ai catalog. Conclusion - if you aren’t using Shopify* you should make sure your products available in and optimized for Meta.ai.

2. Enrich - Both Meta.ai and Muse, being part of the Facebook family, know a lot about their users, the ‘intent gap’ between shopper intent and product intent is quite large. To make sure Muse can recommend your product for the high-intent shopper, like with other LLM-based Agentic Channels, the biggest gap is going to be in product data - who are your products designed for, the use case, etc.

3. Monitor - You can’t improve your Muse sales without monitoring. Because Muse discovery is primarily driven by the Meta.ai catalog, you can get a two-for-one by measuring your product card level visibility and quality there vs. Muse which is much harder to monitor.

4. Checkout - Beyond discovery, to capture the transaction, unless you are on Shopify, you need to verify that your infrastructure layer and guest checkout are Muse friendly. We’ll give you a step-by-step example of how to do this.

5. UCP - We see evidence that Muse /Meta are using UCP for discovery, and checkout - but primarily for Shopify. If you aren’t on Shopify, we believe this is coming soon, maybe exclusive tot he Muse Connector program.

That’s your ‘cheat sheet’, each section below digs into the ‘how we know’, the ‘why’ and the ‘how’ - how to implement each of these best practices.

Muse Best Practice 1: Catalog Discoverability

I want to first, thank Vedant Gupta on our team, he did the leg work on the CSI work for Muse. If you like this more technical side of things, he’ll be posting more details on LinkedIn shortly.

In a nutshell - it turns out that Muse is a derivative of the OpenClaw system, so you can ask it to see the system files. If you ask your Muse, it will share the system files, that are basically Markdown files that all together define the Muse ‘harness’ (we covered harnesses in part 1 of our personal agent series here). In the Muse operating system there is a very specific shopping skill that lives in a human readable markdown file (shopping-SKILL.md).

These ‘primary tools’ are where our CSI work ends, but the knowledge in this skill file gives us enough to throw the right shopping-related prompts at Muse and figure out what’s going on inside the ‘meta catalog search’. Before we jump into that, note these parts of the shopping-SKILL.md file:

• The skill explicitly tells the agent to focus on low prices, but also to consider the ‘constraints’ (another word for context - e.g. did the user ask about features, color, size, location)

• There are three top-level pools of inventory:

  • Meta catalog search - we are opaque to the details inside this tool.

  • Browser product search - you’ll note the ‘universal product coverage’ verbiage here and it also suggests the meta catalog search is particularly weak for ‘home goods’. Besides this contextual logic, we can not see inside the browser product search tasks.

  • FB Marketplace - This internal tool provides the rapid search of facebook marketplace.

For our purposes we can ignore FB Marketplace (FBM) - it only pops up for collectible, furniture, auto and other requests where FBM has a large local inventory set. Note that FB blocks crawls of FBM, so this is proprietary inventory only available to Muse. Like Amazon, Meta is simultaneously, relying on crawling the internet, but blocking its own assets from crawl.

Hypothesis and Results from our Meta.ai and Muse Prompt Test

The name of the first skill, meta catalog search, gives us a clue of the source of a large part of Muse’s internal catalog. The word ‘Meta’ is right there. Waaay back in March, we were the first to report that their chatbot, Meta.ai, was live with product cards (here)

Our hypothesis: Why would Meta build TWO product catalogs - why not use the one they revealed in March and have been working on and making improvements to all that time. Logically you would leverage the Meta.ai catalog. To test the hypothesis, we built a testing framework to run product-level prompts against Meta.ai and Muse and compared the results. The results:

• 89% of the time when a test product appeared in Meta.ai’s top three results, Muse also showed it. The conclusion from this data is that Muse’s top source for product information is Meta.ai (it says as much in the skill, we wanted verification). The crawl is a catch all for any missed great deals/products and updating/verifying what’s in the Meta.ai catalog.

• 73% of the time, prompts returned at least one listing witht he same product title and seller on both platforms - another indication there is a lot of overlap in Meta.ai’s product catalog and Muse.

Where the differences come in is by its nature and with connectors (email/calendar/etc), the agent has more context and that seems to be part of where the 11-27% delta in there comes from. There’s also the ‘universal product coverage’ mention in there. We do know that Muse and Shopify announced a deal and our tests also indicate that inside of the skill, (not clear/known if it’s from the Meta.ai side or the Muse skill), Muse is utilizing the UCP protocol to browse products as well - this is definitely the case for Shopify stores, and we believe that it’s generally available for anyone.

Discovery: Putting it all together

Here’s what happens:

• The user sends the Meta agent a prompt that’s determined to be product-based

• The shopping skill swirls together:

  • the prompt

  • context from agentic user history

  • low price focus as per the skill

  • any other criteria

• In parallel, the Muse agent’s skill pulls products from:

  • The meta.ai catalog

  • UCP

  • Web crawler

• The products from the three are combined, canonicalized and ranked and delivered to the user

• Finally, a (usually low price at top) short list of products is produced

Discovery Conclusion: Meta.ai

Based on our research, if you want to do one thing to increase your Muse discoverability, it boils down to: Make sure your products are available in Meta.ai. Since 2015, Facebook has taken datafeeds from their version of a merchant center called Meta Commerce Manager. That gives your products a direct line into Meta.ai and, as we now know - right into the Muse agent.

Meta.ai will get you 90% of the way - for bonus points if you want to implement UCP and make sure your products are crawlable it will help your odds. But having a product available for selection is only part of the discovery puzzle, what we’ve learned from this exploration is that you have to bridge the gap between the massive shopper context and the typically under-contexted product catalog.

This brings us to Best Practice 2.

Subscribe now

Muse Best Practice 2: Enrich

As a reminder, back in July in this post→

we introduced the concept that not all ‘enrichments’ are the same, in fact we have identified 5 levels. Unfortunately many brands work with PIMs that have an enrichment feature. That’s what we call Level 1 enrichment. That’s the basics for pre-2022 era optimization. In the Agentic era it’s now a ‘must have’, but it’s not enough. For Muse type operation you need at least what we call ‘Level 2 - Conversational Attributes’. Ideally, we recommend ‘Level 3 - Shopper Context’.

Why?

Remember this box from our discovery diagram above:

As people are using agents, the agent typically has access to their email, which pulls in shopping history, their past conversations, their travel, their health information. They know if the outfit is for an event, they know the trip the customer is going on, they know the preference of the customer to buy beauty products with a jasmine scent. But without YOU telling it, it will NOT know:

• Answers to FAQs about your products

• Why people love your product

• The occasion your product is great for

• Your product’s scent profile

• Your product’s target use case

• What do you say about negative reviews? Is there new product packaging, a new formulation, directions not always followed, fit details?

The list can be long, but as we’ve pointed out previously, you can use agentic commerce to surface contextual clues about what all ACO surfaces don’t understand about your product (we call this context capture) and use that to prioritize this list.

In fact, long-time readers of Retailgentic are in a great position for Muse optimizations, because, like all other Agentic Commerce surfaces, our 11 step framework is 100% applicable.

Leveraging Meta.ai to get your products into Muse is a shortcut that let’s you skip steps 1 and 2 and go straight to 3+.

A Note for Shopify Merchants

It’s our point of view that what Shopify’s catalog provides enrichment-wise and distribution wise is fine for smaller < $20m/yr merchants which is 95% of their customer base, but for larger merchants, as detailed in this post from August, the enrichments are Level 1-2 and the ‘clustering’ and ‘compression’ may not result in the representation and agentic optimization of your products that advanced merchants would want. In that post we detailed our free utility - Shop Cat Explorer (here) that allows you to see exactly how an agent like Muse sees your catalog via the Shopify Catalog UCP endpoint.

Muse Best Practice 3: Monitor

Another point we’ve been discussing since inception is the importance of a closed-loop optimization system. For Muse Optimization Best Practice 3 - Monitor, the nuance is with Muse it is extremely hard to repeatedly monitor a set of products because of the pesky power of its memory. There is no ‘temporary chat’, but since we now know that the bulk of the Muse catalog is coming from Meta.ai, you can loop on that and get 90% there, saving prompts on Muse for the last 10%.

What you’re monitoring here is:

• Are your products showing up on Meta.ai for product-level prompts, product-constrained prompts, feature-level prompts, category prompts, and broader AEO/GEO prompts.

• Are your product cards correct? Do they have the right price, features, variants, attributes, etc.? Are you happy with the hero image? Do they land on the right page?

• Verify the checkout isn’t blocked, which brings us to ‘Best Practice 4’…

Muse Best Practice 4: Checkout

In the shopping skill, there are details for us about how Muse purchases products:

Parts 1,2 and 3 are implementation details. The meat is in instructions 4 and 5. It basically says - if this is a Shopify product, checkout via UCP, otherwise, go through the guest checkout and use Link.

Therefore, there are two steps here:

• If you are on Shopify, verify that everything works and is wired correctly for both Shop Pay and Link.

• If you are not on Shopify, on a regular basis have Muse run through your checkout to make sure there are no blocks on enabling the agent to checkout on the consumer’s behalf

Muse Best Practice 5: UCP

We put this last, because it’s a bit of speculation mixed with what we know:

1. We know that Muse ‘speaks’ the UCP protocol because it’s crystal clear from the shopping skill, it is making a tool call that uses both the UCP powered catalog discovery and the checkout.

2. We’re speculating that the connectors unlock a Muse/Meta implementation of UCP that you would imagine would be as close to Google Gemini’s as possible to minimize merchant work and maximize time to Muse availability.

Therefore, our last best practice, while not a requirement is an investment in the Muse future and you get Google Gemini - think of it as a BOGO.

What about the Muse Connectors?

On 9/23/26 at Meta Connect, Meta announced 10 retailers as part of an inaugural class of Retailers with ‘Muse connectors’.

The retailer partners announced (in addition to the launch partner of Shopify) are:Walmart, Best Buy, American Eagle Outfitters, DICK’S Sporting Goods, Fanatics, Gap, Michael Kors, Sephora, Ulta, and Wayfair.

None of these are live yet, so we can only speculate on what this means. Based on past connector programs for ChatGPT, etc. Most likely it includes:

• A discovery component - I would imagine via UCP or datafeed

• A checkout component - almost certainly a UCP-powered checkout

• Wall St is speculating a revenue share type agreement

• Potentially advanced features such as multi-cart, loyalty program integration, participation in a BFCM deals program, etc.

We will report back when these are live.

Muse Shopify Shop Pay Integration is Live!

We have been able to checkout using Shopify Shop Pay - offered as an option next to Link, this is live now. For example, on the Glossier store. Here’s what it looks like:

Clicking Shop Pay leads to a typical Shop Pay login flow→

Summary

Between now and BFCM, if you had to do 3 tactical things to optimize for Meta Muse

1. Distribute your products to the Meta Commerce Manager and verify they are part of the Meta.ai catalog

2. Level-3 Enrich your top 5 or 10% of SKUs

3. Verify your checkout isn’t blocked

After that, assuming Muse continues to surge, you can revisit Monitor and implementing UCP.

What’s Next?

In Part 2, we’ll highlight the latest news on Muse - the latest data on scale, usage and just yesterday, Meta dropped a new model, “Muse Wonder” that they have explicitly said will be powering the Muse agent in the near future. Details here on that if you’re interested. The current Muse agent runs on the model called ‘Muse Spark’.

It appears that Meta’s huge bet in their AI reset has worked and they have jumped onto the frontier:

I’ve spent a LOT of time watching Muse Spark do computer use (on the Muse agent harness) vs. Astra, Fable and Opus - it’s very slow compared to those models. I haven’t seen the computer use benchmarks yet, but to jump this high on AAII, I imagine we’re going to see the Spark→Wonder jump in capabilities be a doubling or tripling. Said another way Muse today is very good, but it’s also the worst it’s ever going to be - it is only going to get smarter, faster and more proactive from here.

Next week, we’ll dig into more of that.

Subscribe now

Subscribe now

Ekta Chopra is the Chief Technology and AI Officer at e.l.f. Beauty, where she oversees all of technology: infrastructure (”the plumbing”), data, marketing technology, AI, governance and security. She has been at e.l.f. for more than a decade (10 years and nine months, by her count) and was Chief Digital Officer when we last talked to her a year ago. Ekta also writes a Substack called AI Chef.

We recorded this one live on the first day of RetailClub in Huntington Beach, with a live audience and a very sneezy Jason Goldberg as co-host. Ekta walked over straight from her main-stage talk with an entourage of about 30 people in tow, so we knew we were in for a good one.

Listen/Watch Our Interview:

We pick up where we left off a year ago and get into what has changed since. Ekta walks us through how e.l.f. is structuring its AI work, what kind of person they’re hiring to do it, and how a public company makes the case for investing before the ROI is fully provable. The live audience jumps in with some sharp questions along the way.

We close by looking ahead to what Ekta thinks brands need to get ready for in the next 12 months. Hit play.

Highlights

• 𝗟𝗲𝘁 𝗶𝘁 𝗿𝗶𝗽 𝗼𝗿 𝗹𝗼𝗰𝗸 𝗶𝘁 𝗱𝗼𝘄𝗻: Ekta says companies tend to pick one of two AI philosophies, giving it to everyone or locking it down. e.l.f. has tried to do both so it doesn't kill innovation.

• 𝗧𝗵𝗿𝗲𝗲 𝗽𝗼𝗱𝘀: e.l.f. runs three AI pods covering agentic commerce/growth, intelligent operations, and the human-agent workforce. Each pod owns data, change management, integrations, and training. The agentic growth roles are now being hired as dedicated full-time positions across e.l.f.'s brands.

• 𝗡𝗼 𝗼𝗻𝗲 𝗵𝗮𝘀 𝗺𝗲𝗮𝘀𝘂𝗿𝗲𝗺𝗲𝗻𝘁 𝗳𝗶𝗴𝘂𝗿𝗲𝗱 𝗼𝘂𝘁: "Anybody that tells you they have figured it out haven't figured it out." Ekta measures readiness instead, and focuses on the products where e.l.f. has the right to win.

• 𝗥𝗲𝗮𝗹 𝗽𝗿𝗼𝗺𝗽𝘁𝘀, 𝗳𝗶𝗻𝗮𝗹𝗹𝘆: Jason points out that most visibility tools rely on synthetic prompts. Google's new Agentic Insights tab in Merchant Center is the first look at what real people actually type, at least on Google surfaces.

• 𝗕𝗮𝗱 𝗱𝗮𝘁𝗮 = 𝗯𝗮𝗱 𝗔𝗜: Ekta's case for fixing the foundation first. If you don't solve the plumbing, you can't scale.

• 𝗘𝘅𝘁𝗲𝗿𝗻𝗮𝗹 𝘃𝗼𝗶𝗰𝗲𝘀 𝗹𝗶𝗴𝗵𝘁 𝘁𝗵𝗲 𝗳𝗶𝗿𝗲: About 14 months ago e.l.f. brought in an outside search expert, and that put "fire in the belly."

• 𝗦𝘂𝗽𝗽𝗹𝘆 𝗰𝗵𝗮𝗶𝗻 𝗶𝘀 𝘁𝗵𝗲 𝘀𝗹𝗲𝗲𝗽𝗲𝗿: e.l.f. has invested in Lyric for forecasting and demand planning. Its planners "are actually now engineers," and AP, AR, payroll and inventory teams are building their own AI workflows.

• 𝗔𝗴𝗲𝗻𝘁 𝗿𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝘀𝗵𝗶𝗽 𝗺𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁: Ekta's prediction is that Meta wins the agent shopping race. She says brands need to put agent identity in place now, so they know which agent is arriving and which customer it represents. It's the next CDP/CRM-style plumbing project.

Huge thanks to Ekta for running over from the main stage, and to our live audience at RetailClub for the great questions. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Earlier today, ReFiBuy announced the availability of the first detailed user behavior study into how consumers shop on ChatGPT from ‘AI Visibility’ (AEO/GEO) through to the transaction. The data in this newsletter and accompanying podcast are the tip of the ‘interesting and actionable data’ iceberg we discovered in this research. You can download the complete 39 page detailed PDF with many more insights, recommendations plus all the data and answers to questions you may have here.

Subscribe now

Backstory: Why We Did This and What is It?

We’re big fans of a SEO/GEO/Growth Substack called ‘Growth Memo’ by Kevin Indig. About 9 months ago they had a foundational report that provided details for how people interact with Google’s AIOs and some of the elements in AI Mode. Where that study stopped short was going beyond the super top-level of discovery - today called “AI visibility” and previously we used to call this, in the Google Era, sentiment analysis.

We researched the company behind this wizardry and found Eric Van Buskirk at Clickstream Solutions. They do User Behavior Studies. What’s a User Behavior study? Good question….

Consumer Behavior: Surveys (Stated Preference) vs. Panel Data (Observed Behavior) vs. User Behavior Study

If you’ve been reading Retailgentic for more than a couple days, you’ve seen us publish a TON of data in two buckets:

1. Consumer Surveys - These are Jason’s favorites. Customer Surveys have an n=X where X is usually 500-10,000 and represents the number of respondents in the survey pool. Statistically you can take these ‘stated preference’ observations and scale them up to the US population. The core problem with surveys is that what people say they would do and what they actually do are often different.

2. Panel data -Similarweb, Sensor Tower, Profound, Gener8, collect millions to tens of millions of consumer clickstreams that typically come from Chrome extensions and Android apps that when installed collect consumer clickstream data (prompts and all!). There are several issues with this data:

   1. Given the sources (android/chrome extension users) it can be hard to get demographic representation that matches the US consumer ‘target’.

   2. Millions of streams may be from hundreds of thousands of consumers and hard to scale up statistically to match the US demographic/trends.

   3. Some people question the provenance of this data and privacy practices behind it.

Turns out Eric and team have taken the previously out-of-reach ($100-$300k+) third option, a scientific user behavior study, made it virtual and focused to be more approachable. User behavior studies have fewer foundational datapoints than surveys and panels, but give us strong directional behavior because they are an observed behavior experiment with real user interfaces and demographically chosen participants that meet a statistical target. If you want more information on this, we go deep in the topic with Eric on the podcast.

Our ChatGPT Hypothesis (derived from Amazon Buy Box behavior)

At ChannelAdvisor, 2 years after launching the Buy Box concept on Amazon, where third parties, plus first parties could ‘battle’ for the buy box, we had enough closed-loop transactional data to make a startling observation. The seller that ‘owned’ the buy-box grabbed 85-95% of the sales depending on the category.

Owning the buy-box was the solution to the zero-sum game. If you owned it, you got the vast majority of sales, if you didn’t, you lost all but 5-15% of the sales for that SKU.

After realizing that, we were the first to offer Buy-box analytics, buy-box dynamic bidding, detailed FBA analysis and more. To the best of my knowledge, the focus on buy-box ownership still drives the bulk of Amazon success, or failure.

What about ChatGPT?

ChatGPT is similar, but unlike Amazon, you now have product cards (in a variety of display options), the written/textual recommendations and the offer card.

We had several questions that are not answerable today without a user behavior study:

• What’s the value of the first product card vs. third vs. fourth?

• What % of people click the offer card and are referred to the retailer vs. influenced?

• Once the product card is selected, what do consumers click on next? The first/top offer? The second offer?

• What influences consumers - the price, the AI recommendation, or the brand?

• Both Google AI Mode (powered by Gemini) and ChatGPT have similar user-experiences, is what we learn from ChatGPT similar to Google AI Mode?

• What other interesting observations can we tease out of watching real people shop on ChatGPT?

• Finally, this is all intellectually interesting, but most importantly: how do we apply it (like the Amazon Buy Box observation) to actionable strategies and tactics that drive sales for retailers and brands?

Working with Eric, in July/August we executed the study and have been working on processing all the data and results. Today we’re excited to reveal the findings. There are three places you can find the information - this Retailgentic newsletter, the detailed report and the podcast

Retailgentic Deep Dive Podcast Available Now…

This newsletter offers my take on the data from the user behavior study. The accompanying podcast is an interview with the behavioral scientist behind the project, Eric Van Buskirk, with Clickstream solutions.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or you can listen right here→

Or, we have a video version on YouTube (make sure to subscribe!) →

Brief Anatomy of the ChatGPT Sentiment→Research→Find→Buy Consumer Flow

Before we jump into my takes on the data, here’s a quick reminder of the three core parts of both the ChatGPT and Google AI mode:

1. Product Card(s) - This is the consumer’s ‘view’ into the underlying Answer Engine’s catalog and the recommendations for the given prompt. Sometimes a single card is shown, sometimes a carousel, sometimes both, sometimes 2-3 cards in a vertical arrangement. To keep the Amazon analogy going, these are like ASINs, though very basic in their development phase compared to what we’re all familiar with on Amazon. Long-time readers will recognize that making sure your product catalog provided via feed or crawler maps cleanly and 100% correctly to the answer engine’s product cards is called canonicalization.

2. Written Recommendation - LLMs are natural language engines and great at generating narrative. Sometimes this is below the product cards, sometimes above, sometimes lower on the page.

3. Offer Card - When a consumer clicks on the product card, they are shown a list of offers that match that product. The sorting of this, just like the Amazon Buy Box is built off of an opaque algorithm. We know the inputs, but we don’t know the underlying algorithm (the output). The inputs are enough for us to decode how to optimize the Offer Card position and give the merchant plenty of ability to make changes to show up higher on the list of offers.

Top 5 ChatGPT Findings and Actionable Strategies from the User Behavior Study

With all that background let’s jump right into the findings. Here are my top five findings along with the actionable strategy from each.

Insight One: Product Card Hypothesis Confirmed, Kind Of…

As mentioned, we had a thesis that being the first product card was important. Turns out that product cards in aggregate on the prompt page draw 75.3% of the consumer’s engagement. On top of that when you look at the influence of the product card - 84% of final product choices in ChatGPT came from a product card

The caveat is our hypothesis guessed that the first product card would be the lion’s share of transactions, but the data tells a different story:

When multiple cards were presented, 43.4% of the time, the consumer picked Position 1, then 17.1% position 2 and, surprisingly, there’s a material bump in position 3 and then it declines to sub 10% for positions 4-6.

Obviously when there is one product card, this surges to as close to 100% as statistically possible.

Actionable Strategy One: Product Cards

The User Study highlights how important it is that your product catalog maps to a product card (canonicalization) and the correct card is chosen. If you don’t do this, you will miss out on 75.3% of the engagement of the page and 84% of the influence.

Long-term readers will know that the key to mapping correctly on the product card is optimizing your product data.

Insight Two: The ChatGPT Offer Card is the New Amazon Buy Box

Interestingly, where our hypothesis proved true was one step further in the shopper’s journey - the Offer Card.

In this graphic, we see that 76% of the consumers picked the first offer on the offer card, with position two plummeting to 14%.

Actionable Strategy Two: Offer Cards

After making sure your products are mapped to the product card, the second priority is offer card optimization. The one input into the algorithm that most merchants won’t change is price, but there are up to 10 other factors, all in the merchant’s control.

At ReFiBuy, we’re actively monitoring over 1m SKUs and in that aggregated dataset, we observe that incorrect offer details drive lower offer card placement for the majority of SKUs.

Why? Because crawler-derived offer data is particularly fragile: shipping details, site-wide offers, and return policies are frequently not reflected correctly. Fix these basics and you can move up as many as 2-3 positions on the offer card.

From there a long-term investment in the five levels of enrichment (covered here) is the next step to improving your inputs into the ChatGPT offer card ranking algorithm.

Insight Three: Product Card Influence > Clicks

The first two insights were what we wanted to know more about with the user behavior study and where our hypothesis was focused.

The biggest surprise from the report came from observing what shoppers did after they moved through the ChatGPT results:

The detailed pdf has all the details, but basically, we discovered that 61% of consumers clicked a product card (when multiple were presented). The users, as part of the task, were asked to pick a product. Even after only hovering on the card (not clicking) they would go to the merchant’s site and pick the product seen on the product card. That’s where the 84% number comes from - any engagement with a product card.

Actionable Strategy Three: Product Cards are the new AI Shelf

In Retailgentic, we’ve talked about Dark Agentic Commerce Traffic. This is additive to that - a consumer very likely will see your product on ChatGPT and be influenced to purchase it by going directly to your website. This will appear as ‘direct’ /organic traffic.

Interestingly, we saw this phenomenon when Amazon launched their first ad products. A popular baby product maker would run an Ad on Amazon and sales across physical stores, their website and other retailers would rise due to the influence of the product’s prominence on the Amazon SERP. How does Amazon compare to ChatGPT?

ChatGPT vs. Amazon’s Digital Shelf influence

ChatGPT announced last Tuesday they have crossed 1.2b weekly active users. Best guesstimate that equates to 1.4b monthly active users. ChatGPT has said that 20% of those users are shopping for products, that gives us 280m shopper MAUs.

Amazon does not disclose global shopping MAUs; it reported approximately 193.9 million average monthly active recipients in the EU for the first half of 2026. The EU is about 1/3 of their revenue so let’s guess 600m MAUs? This is a guess on a guess, but illustrates that ChatGPT’s product influence is likely much larger (about one half the size of Amazon’s) than most merchants realize and certainly is not showing up in click data.

Insight Four: Race to Zero Narrative Violation: Consumers Trust Specs and Brands over Price!

The most common concern that both brands and retailers have about Agentic Commerce is that in a world where agents find the best price, brand doesn’t matter and it’s going to be a race to the bottom.

In the study, we asked participants to weight the trust factors in their product decisions:

• 34.8% said the product’s specifications drove their choice

• 29% chose a brand they were familiar with

• 18.3% chose price/value

The study wasn’t built this way, but my guess is this is a waterfall that goes like this:

If the product meets what I’m looking for, then I look for a familiar brand. If I find more than one, I look at price or perhaps take a risk on a new brand at a compelling price difference, all things equal.

Actionable Strategy Four: There is no Agentic Commerce Race to the Bottom.

Focus on making sure Answer Engines know your detailed product specifications and your brand is properly represented. If a consumer is a value-oriented shopper, Agentic Commerce won’t magnify/minimize that, just like the broader e-commerce and even retail, they represent about 20%. Brand and ‘fit’ still matter the majority of times.

Insight Five: ChatGPT Ads Are Working

The real unintended insight from our study came from ChatGPT ads. By design, our participants were on the ChatGPT free tier, which means they saw ads. The study goes into much more detail, but here are four datapoints that caught my attention:

• 61.7% of our shoppers saw ChatGPT ads

• 71.2% rated the ads at least ‘as useful’ as traditional Google ads

• 48.6% said the ads helped them make a choice at least a little

• 35.1% said they were influenced in their final decision by the ad.

Sometimes insights leave you wanting more. In retrospect, I wish we had measured more about the impact of ads in the design of the study, and, of course, one datapoint is interesting, but not a trend. It would be interesting to see what these datapoints were closer to April/May to see the trajectory. Ad systems are closed-loop learning systems, particularly in the hands of an AI frontier lab. I imagine, the datapoints are increasing from that April/May and by this time next year, the influence will have at least doubled.

Actionable Strategy Five: Don’t Sleep on ChatGPT Ads

It’s early and we need much more data, but directionally this is promising. ChatGPT appears to be executing towards their goal of ads that simultaneously don’t erode trust, add to the overall experience and give the advertiser high ROI for the impact. They may not be there yet, but if ChatGPT can optimize the ad system at even half the logarithmic curve the models are growing on, these ads are on track to give Google, Meta and Amazon a run for their money.

Remember, ChatGPT’s ad system crossed $1b in 200 days, beating Google, Meta and Amazon to the same milestone.

Insight Six: BONUS Insight: Data Quality Problems Drive Consumers Crazy

In the podcast, you’ll hear Eric’s description of how each consumer study was conducted. One important element is the shoppers are asked to verbalize the experience as they go. These three comments from participants show the importance of making sure your product data is picked up 100% correctly by ChatGPT and updated regularly via a datafeed:

Actionable Strategy Six: Details Matter

Invest in making sure ChatGPT and other Agentic Commerce surfaces have the big details right and measure it to make sure you are not eroding consumer confidence in your products and brand.

Interestingly, product data glitches were the biggest negative consumer experience we discovered from the study.

Summary

I hope you enjoyed these six highlights from our agentic commerce User Behavior Study. Here’s a summary of the action items:

1. Product cards drive 84% of the product choice influence - focus on making sure your catalog is mapped correctly. Correct canonicalization (product card mapping) is your friend, from there make sure the ‘hero’ image is right, the price is dialed in and updated correctly as reflected on the product card. Be sure to measure the % of your catalog that is mapped correctly as a KPI.

2. The offer card is the new Amazon buy box - driving 76% of the ‘wins’. Once mapped correctly, make sure your offer is reflected and enrich your product data to further enhance your position. Measure your relative offer card position as another KPI.

3. The product card drives significantly more purchase influence than measurable clicks. Work on tuning your analytics to measure the known DACT and most likely a % of your direct organic traffic is being influenced by ChatGPT product cards. Think of ChatGPT product cards as part of the larger AI Shelf.

4. There is no evidence that Agentic Commerce shoppers are using the new technology for a ‘race to the bottom’. Instead, Agentic Commerce empowers them to get to the right features and brand they want faster.

5. ChatGPT ads are surprisingly impactful and viewed positively or neutrally by shoppers. Running a test in the next year may be prudent.

6. Shoppers hate data quality problems, make sure this holiday if you run promotions they are correctly reflected across all agentic commerce surfaces.

The detailed user study presents 8 different action items based on the broader dataset.

You can download the detailed data here to get the complete study in a ~40 page PDF that also includes the detail of all the shopper demographics, the tasks they were given, the categories they shopped and much much more.

Subscribe now

Subscribe now

Danny Smith is the Global Agentic Commerce Solutions Lead at Stripe, where he runs technical go-to-market (think pre-sales architect, not engineering) for Stripe’s agentic commerce products.

One of his first projects was adding the payments layer to Amazon’s Rufus shopping assistant, and three years ago he got on stage at Stripe Sessions to talk about agentic commerce.

We recorded this one live from day two of RetailClub in Huntington Beach. It was a little overcast, which after a day of baking in the sun, we didn’t mind at all. Jason Goldberg joined me as co-host, and we grabbed Danny between meetings. It was his first time at the show.

Listen/Watch Our Interview:

Danny’s a first-timer, both on the pod and at RetailClub. This week’s episode is a product deep dive: Stripe has been shipping agentic commerce products at a pace that makes three months ago feel like ancient history. Danny walks us through what’s live, what launched last Monday, and what’s coming soon, including how the checkout Scot kept seeing inside Meta’s new Muse app actually works.

Jason brings the payments-industry history. It goes back to when card-not-present transactions were illegal and runs up to Stripe’s jump from developer darling to incumbent. We also get into the thing that’s still breaking agentic commerce today, and it isn’t the agents.

Along the way there’s a hallucinating agent, a leather jacket, and a Cheeky Pint swag request. Hit play.

Highlights

• 𝗟𝗶𝗻𝗸 𝗪𝗮𝗹𝗹𝗲𝘁 𝗳𝗼𝗿 𝗔𝗴𝗲𝗻𝘁𝘀: A skill you teach any agent (Claude, GrokBot, your homebrew Hermes agent) so it can pay “in a guarded and secure manner.” Details at link.com/agents. Learn more about Link Wallet here.

• “𝗬𝗼𝘂𝗿 𝗔𝗣𝗜 𝗰𝗵𝗮𝗻𝗴𝗲𝗱 𝗺𝘆 𝗹𝗶𝗳𝗲”: Developers really do stop Danny at conferences to ask for a photo. Stripe’s founders’ mantra: “Let’s win all startups, and then let’s go win them again.”

• 𝗢𝗽𝗲𝗻 𝗵𝗲𝗮𝗿𝘁 𝘀𝘂𝗿𝗴𝗲𝗿𝘆: Changing payments at an enterprise is “like open heart surgery.” Jason adds that it’s open heart surgery while you’re running a marathon.

• 𝗪𝗼𝗿𝗹𝗱’𝘀 𝗳𝗶𝗿𝘀𝘁 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝘁𝗼𝗸𝗲𝗻: Stripe’s shared payment tokens launched with OpenAI, and Etsy was the first merchant. Link wraps them as a single-use virtual card capped at the amount you budgeted, so the merchant doesn’t have to be on Stripe.

• 𝗦𝘁𝗿𝗶𝗽𝗲 𝗖𝗵𝗲𝗰𝗸𝗼𝘂𝘁 𝗳𝗼𝗿 𝗔𝗴𝗲𝗻𝘁𝘀: Launched last Monday and built on WebMCP, it makes human checkouts machine-readable at the same time. Stripe merchants on hosted checkout or invoices get it with no extra work.

• 𝗧𝗵𝗲 𝗪𝗲𝗯𝗠𝗖𝗣 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: 42% fewer tokens burned, 38% fewer tool calls, and 39% faster checkout. “Three seconds is annoying to a human, 800 milliseconds is beyond annoying to an agent.”

• 𝗗𝗼𝘂𝗯𝗹𝗲𝗱 𝘄𝗲𝗯 𝘁𝗿𝗮𝗳𝗳𝗶𝗰: Through the Agentic Commerce Suite (ACS), some merchants have doubled their web traffic in three or four months just by getting their catalogs in front of agents. Instant checkout on Gemini is expected around late October.

• 𝗗𝗼𝗼𝗿𝗺𝗮𝗻 𝘃𝘀. 𝗯𝗼𝘂𝗻𝗰𝗲𝗿: Most merchants are bouncers that block all bot traffic, and bots were already 53% of internet traffic in 2024. Stripe is working on protocols to tell good bots from bad ones.

Stripe is building both sides of the agentic transaction, and Danny is the one out there getting retailers across the line. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Subscribe now

New Agentic Commerce Surface: Personal Agents

In our State of Agentic Commerce update yesterday, we introduced this concept of the four Agentic Commerce Surfaces. ChatGPT dots is a significant new entry into the ‘Personal Shopping Agents’ category with META Muse, GrokBot, Instinct, Microsoft CoPilot Autopilot, Gemini Spark and Manus cue.

While these agents do a little bit of everything, our focus is on their agentic commerce capabilities.

In this video:

In this < 20min video, we:

• Setup my first dot!

• Search for a complex dog treat

• See if Amazon, Chewy, Petco, Petsmart, Crocs.com or Dick’s Sporting goods block

• Use the ‘call your dot’ function to look for an Emergency Croc

• Look at local inventory availability

If you prefer to watch it on YouTube, here it is→

Subscribe now

Yesterday on Retailgentic, we wrapped up our two-part series on the State of Agentic Commerce as of the start of Q4 2026. We had an executive brief today and I used the opportunity to take the themes from the substack, integrate them in with a bunch of other new material shot a video walking through it all.

If you didn’t have time to read the long-form two parter, this presentation should get you up to speed with all the innovations happening in Agentic Commerce as of 9/30/26.

A PDF of the presentation follows and some interviews and media I mention in the presentation, followed by a quick Dots overview.

Note: if you prefer the Youtube Viewer you can watch it here.

State of Agentic Commerce Presentation Slides

Here are the slides→

State Of Agentic Commerce End Of Q326 Sm

2.17MB ∙ PDF file

Download

Download

You are free to use these in your own presentations, but please reference Retailgentic

References from the Presentation

• Instinct website: www.instinct.com

• Instinct founder and ceo interview:

ChatGPT Dots

Yesterday at OpenAI DevDays, they announced their entry into the Personal Agent battle, dots. On first blush, it’s more pro than consumer (like GrokBot, but not like Muse). We don’t have access yet, once we do, we’ll post some live demos.

Until then, here’s the launch video:

Welcome to part 2 of this 2 parter. In Part 1 (here), “Personal Agents enter the arena”, we looked at the newest category of Agentic Shopping that we have classified as ‘Personal Shopping Agents’. These horizontal Personal Agents, such as Muse, GrokBot and Instinct, all feature shopping as a key component of their value proposition to consumers. In that post, we gave some background that answered the ‘why now’ question about the surge and talked about these important concepts:

• Exponential growth in capabilities of underlying models

• The frontier labs focus on training for long-running agentic tasks

• They are also training for ‘computer use’ across hundreds of domains including ‘shopping’

• What is a harness?

• A brief history of personal agents

• Browser use: server vs. local and why it matters

• Detailed demos of Muse, Instinct, Grokbot and Gemini Spark

Subscribe now

Rumor: OpenAI ‘o’ or ‘Dots’ on the way later today?

Updated Agentic Commerce Datapoints

When we started Retailgentic in May of 2025, there were 2-3 firms doing surveys about consumer AI adoption and they updated the datapoints every 6-12 months. We were starving for data!

Today, there are at least 2 and sometimes as many as 6-10 surveys to pick from each WEEK! For example, if you look at the Retailgentic August Agentic News Roundup, you’ll see we cherry picked 7 great data points from BCG, Microsoft, Narvar, Morgan Stanley, FinanceBuzz, Koddi and Croud.

That was 30 days ago and we could flood you with more survey data here. Instead we’re going to highlight 3 new datapoints.

NielsonIQ: Majority of US Consumers Now Use AI to Shop

NIQ, one of the most trusted measurers of consumer behavior announced a big milestone in their tracking - more than half of US consumers actively use AI to shop.

SimilarWeb: AI in the path-to-purchase is up >2X since 2024

This chart shows:

• Only visits that end in a transaction

• The % of movement goes 4.5% in July of 2024, to ~9% then now at 11.4%

• Because they have panel data, this most likely includes AFS/Sparky and shopper agents as well as answer engines and they should be able to also track mobile apps. With the proliferation of dark agentic commerce traffic, this is the most reliable datapoint and it aligns with what we see at ReFiBuy when we’ve worked with clients to detect and track DACT.

Similarweb: Answer Engine Share Trends (August 2026 Update)

Similarweb used to publish an every 2-week AI Traffic tracker and now only publishes this view monthly. What we see here is:

• Web only (mobile apps, desktop apps not captured which is a material part of how Answer Engines are now used).

• That being said, you can see ChatGPT bottoming out at 50% and starting to claw back some share. You can also see Claude moving into a third slot behind Gemini and Gemini shrinking a fair amount.

The deceptive thing about this chart is it does not show the growth of the whole industry because it’s basically a percentage chart always capped at 100% to show share.

Fortunately, they just published an Anthropic deepdive that gives us this tidbit: (again web only)

Here we can see Claude closing in on 1b visits per month and their tremendous 540% y/y growth followed by Gemini at 262%. While ChatGPT is showing flat to slightly down y/y, that’s probably seasonality because they are in the stratosphere at 5.64b visits/month.

Those are the three biggest datapoints - Agentic Commerce has crossed 50% usage of US consumers, it’s driving transactions and it’s branching out into new areas. Let’s dig into those details and that will inform what Retailers and Brands can do to optimize Agentic Commerce for Q4.

Subscribe now

The Four Agentic Commerce Selling Surfaces

In the early days of Retailgentic (2025-Q1 26) while Rufus and Sparky were out there, they were generations behind the Answer Engines, so we focused on the Answer Engines. Then around Q2 26, the Retailer Agents took a significant step up. Now every large retailer is either re-doing their strategy to level-up (Target announced this strategy) or launching a very advanced retailer agent such as Instacart’s Clementine (which just launched a national TV campaign, here’s my favorite one).

OpenClaw came out in November of 2025 and everyone saw a glimpse of the future. Now we have our third ‘Agentic Commerce Selling Surface’ - Personal Agents.

Finally, both retailers and brands seeing the success of AFS/Rufus and Sparky and also realizing they need to elevate the customer experience to match the wave of Agentic Commerce are investing in next-generation on-site agents (Gemini, AWS, Claude, Salesforce and hundreds of other options are available), AI/LLM powered on-site search engines, next-generation personalization and recommendation systems, etc.

Today the result is we have four Agentic Commerce selling surfaces:

1. Answer Engines

2. Retailer Agents

3. Personal Agents

4. Brand.com/Retailer.com

In the next section let’s look at a snapshot of where each is using our Retailgentic Agentic Commerce 5 Levels of Autonomy framework.

Retailgentic Agentic Commerce Levels of Autonomy

I met many long-time Retailgentic readers at RetailClub and the Agentic Commerce Levels of Autonomy framework always comes up. We created this to help ‘level-set’ the various answer engines. Today we’re expanding it to cover Retailer Agents and Personal Agents as well.

One clarification we’ve added - to get to Advanced Buy, a pre-requisite is a cart, or more technically the ability to buy more than one thing. For example, for a single unit, there are now Personal Agents that can achieve Level-5 full autonomy. But the single-item limitation keeps them below level 4. The good news is I think we’re months away from some clever cart implementations, so stay tuned.

Before we get into that in detail, let’s revisit Answer Engines.

AC Autonomy: Answer Engines

There hasn’t been much movement here, of note:

• This year ChatGPT decided to focus on discovery (Research/Find) and not checkout, so they moved back to Level 2

• Claude did come out with product cards

• Apple’s new Siri with iOS 27 has some agentic commerce capabilities (more on that later) so we moved it to Level-2

• Gemini’s UCP and Multi-cart announcement have it ahead of the pack at Level-4

Anatomy of a Product Card

In answer engines, there is an underlying ‘global catalog’ that the answer engines have built. The consumer’s view of that catalog is through Product cards and Offer Cards.

In the above example, you can see the 7 different components of a product card. Whether you are a brand or a retailer (or both), making sure your products show up, are correctly represented and correctly linked to your site for purchase is absolutely essential.

Now let’s look at Retailer Agents:

AC Autonomy: Retail Agents

There are six Retail Agents we are tracking closely:

1. AFS - Amazon moved Rufus to AFS this year and recently implemented the ‘auto-buy’ feature. AFS works with the Amazon cart so it gets multi-item capabilities ‘for free’. With cart and auto-buy we have moved AFS to the highest level recorded thus far: Level 4.5. What keeps it out of Full Autonomy is no predictive element and the multi-cart and auto-pay support is still rudimentary. Amazon has reported that 350m consumers use AFS, interactions are up 5x y/y and AFS users spend 40% more per order.

2. Sparky - Walmart has moved Sparky along dramatically in 2026. It went from mobile only to web and now has a more dedicated panel. Along the way the experience has improved dramatically. Like AFS, Sparky gets cart ‘for free’, but does not yet have an auto-buy type feature. Walmart’s CEO Furner said in their Q2 that Sparky usage is up 70% y/y, and Sparky users spend approximately 40% more than non-Sparky (sounds familiar!).

3. Target - Target’s agent used to be called TSA - Target Shopping Assistant. In their quarterly earnings they announced they have something new coming out later this year. We’re keeping an eye out for that.

4. Instacart Clementine - If you haven’t tried Clementine, you should give it a shot. It’s got meal planning, multi-item, advanced cart capabilities and is a very solid offering that sets the bar high for CPG/Grocery type agents.

5. Home Depot and Lowes - Both companies have increased their investments and their agents are improving, but still several steps behind the state-of-the-art.

Now let’s look at the exciting new OpenClaw-inspired category: Personal Agents:

AC Autonomy: Personal Agents

We went from zero to 7 offerings here and more are coming. You can see we’ve clustered Muse, GrokBot, Instinct, Cue (from Muse), and made a guess that ‘Dots’ will be at the 3.5 level. As mentioned, we are placing them there due to the lack of a multi-item/cart capability. Microsoft’s Autopilot and Gemini’s Spark are very far behind, but we expect Google to catch up by end of October.

Instinct’s CEO yesterday announced a massive funding round and that the company is driving over $1B of transactions. We suspect the bulk of that is Travel, but there’s a good bit of commerce in there - best guess is 70/30 split travel/ecomm. Instinct is growing 10% w/w so that will be $2B plus very quickly.

Agentic Commerce: Brand.com/Retailer.com

Because it’s so unique to every retailer, we can’t place all of these on the Levels of Autonomy framework, but you’re seeing every retailer and brand invest heavily in this space. We are evaluating adding a measure of where the top 1000 retailers are on this dimension to our popular AI1000 index.

Personal Agents and Answer Engines vs. Retail Agents

For me, one the most interesting arguments in Agentic Commerce is ‘who will win’:

• Horizontal play: Personal Agents and Answer Engines vs.

• Vertical play: Retail agents

Here’s a simple visualization of the argument:

The argument FOR Answer Engines and Personal Agents goes:

• If the consumer trusts the agent with multiple-aspects of their lives, the agent will have a ‘first shot’ at any shopping that happens. For example, a project like ‘I’m planning a trip to Europe next Summer’ could lead to trip planning, travel booking, and purchases of luggage and apparel for the trip (and sunscreen - Jason Goldberg’s favorite example).

The argument FOR the retailer agent goes:

• The retailer agent (AFS/Sparky) is so focused on doing this well, that consumers will leave the personal agent for the fast shipping, the store experience, the returns capabilities, etc.

Who do you think will win?

Agentic Commerce Optimization: What is it and Updated Strategies

When it comes to what you can do to optimize for a strong holiday 2026, we coined the phrase Agentic Commerce Optimization:

AEO/GEO focuses on AI visibility - does the Agentic Commerce surface (AEO/GEO is really only Answer Engines) know your brand, does it mention you - this is called AI Visibility. Most of these companies are focused on a very vertical set of customers: Fintech, travel, b2b, etc. and for those types of companies being strong at the sentiment/visibility layer is important.

In our world of commerce, our register doesn’t ring on AI Visibility, it rings on a purchase - the BUY part of the funnel. ACO picks up where GEO leaves off:

• Making sure your products are on the product card

• Are all your product’s features and offer components correctly stated?

• Are you optimized for each surface?

Underneath ACO it always comes back to the product catalog - it’s the foundation for everything:

Here we’re illustrating a framework to think about:

• First, your product catalog used to be largely static - today it needs to be updated regularly to keep up with the optimizations and changing ways consumers are discovering products.

• Previously we used to primarily focus on the on-site PDP for your product catalog and them some data feeds to Google Shopping. Today you need to surface your catalog via meta data for crawlers, enhanced feeds for answer engines, UCP, if you are on Amazon/Walmart, you need to apply ACO principles for AFS/Sparky, TikTok Shops, your on-site agents, etc.

The good news is an investment in your product catalog pays off across all 4 Agentic Commerce surfaces and we believe it’s in the sweet spot of ‘low effort’ / ‘very high impact’.

ACO Everywhere

We call this strategy of applying ACO-oriented optimizations to your catalog on a recurring basis: ACO Everywhere:

Thankfully despite all this new complexity we’re dealing with and the 30-40 new destinations for consumers, your catalog can be simultaneously optimized for Answer Engines, Personal Agents, Retailer Agents AND your on-site Agentic Commerce investments. It’s like a 4X BOGO!

ACO 11-Step Strategy

Because of what we’ve learned from the power of loops and context capture to pull in the learnings from all these surfaces back into the catalog in a closed-learning-loop, we added steps 10 and 11 to the ACO 9 step strategy sequence:

Conclusion and Happy Agentic Commercing…

In Summary:

• Consumers continue rapid adoption of Agentic Commerce

• There are now four Agentic Commerce Surfaces to consider: Answer Engines, Personal Agents, Retail Agents and on-site agents.

• Agentic Commerce Optimization (ACO) provides actionable and measurable steps to increase sales on these surfaces

• ACO focuses on optimizing your product catalog

• Advanced ACO is a closed loop system that captures product-level context, measures the impact at each Agentic Commerce surface, grabs context clues and feeds all of that back into the system for continuous improvement.

• Anyone can (and should) implement ACO, small marketplace sellers, brands, grocers, CPG, auto.

• An investment in your catalog today, amplifies to all four agentic commerce selling surfaces and will future-proof your company through these exponentially growing changes and explosion of consumer options.

Happy Agentic Commercing and we hope everyone has a strong Q4 - we’ll be back in Q1 with another State of Agentic Commerce Update and, of course, we’ll be reporting all the news heading up to and through Holiday 26.

Subscribe now

Subscribe now

Bryan House is the CEO of Elastic Path, the composable commerce platform for complex B2B and enterprise commerce. He joined Elastic Path in 2020 to run product and became CEO in 2024.

ReFiBuy and Elastic Path announced a partnership in May, bringing ReFiBuy's agentic commerce optimization to Elastic Path customers. So this conversation is equal parts old friends catching up and partners comparing notes on where commerce is headed.

Listen/Watch Our Interview:

This week’s episode is a platform deep dive with one of the longest-tenured companies in e-commerce. Elastic Path has been building commerce software for 26 years and Bryan is now steering the company straight into the agentic era.

It’s part founder-journey (craft brewing to CEO is not a typical path), part product deep dive, and part industry-shift conversation about what AI does to the economics of running a commerce platform.

Highlights

• 𝗙𝗿𝗼𝗺 𝗯𝗿𝗲𝘄𝗲𝗿 𝘁𝗼 𝗖𝗘𝗢: Bryan spent eight years as a professional brewer, with an actual brewing degree, at a brewpub across from Fenway Park. "Everything I learned in my career that really moved the needle for me, I learned as a brewer."

• 𝟮𝟲 𝘆𝗲𝗮𝗿𝘀 𝗼𝗳 𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗲: Elastic Path's claim to fame was building the store for the Vancouver 2010 Winter Olympics, the birth of its self-managed commerce product.

• 𝗧𝗵𝗲 $𝟱𝟬𝟬𝗞 𝘂𝗽𝗴𝗿𝗮𝗱𝗲, 𝗸𝗶𝗹𝗹𝗲𝗱: Elastic Path built an AI updater so one developer can upgrade a self-managed platform in about two weeks, replacing what used to be a half-million-dollar annual project.

• 𝟭𝟬,𝟬𝟬𝟬 𝘃𝗮𝗿𝗶𝗮𝘁𝗶𝗼𝗻𝘀: where typical platforms cap out at 250 variations per item, Elastic Path supports 10,000, plus bundles of up to 1,500 items.

• 𝗦𝗶𝘅 𝘄𝗲𝗲𝗸𝘀 𝗮𝗻𝗱 𝟭𝟬𝟬 𝗴𝗿𝗮𝗻𝗱: Bryan's prediction… AI takes re-platforming from a 6-month-to-2-year, $500K–$3M digital transformation down to a setup fee. "That's the place we're gonna see the biggest impact of AI in commerce."

• 𝗧𝗵𝗲 𝟴𝟬-𝗽𝗮𝗴𝗲 𝗣𝗗𝗙: in a world of AI discovery, every product needs the equivalent of an 80-page PDF of data to train the LLM on when to recommend it. (Scot's line, which Bryan says he "steals with credit" as often as he can.)

• 𝗣𝗿𝗼𝗼𝗳 𝗶𝗻 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: a UK custom greeting-card seller saw conversion jump 40% almost immediately after adopting Elastic Path's checkout

• 𝗡𝗼𝗯𝗼𝗱𝘆 𝘄𝗮𝗻𝘁𝘀 𝗰𝗵𝗮𝘁𝗯𝗼𝘁𝘀: customers are asking about UCP and ACP, discovery, and the long tail, not chatbots.

• 𝗪𝗵𝗲𝗿𝗲 𝘁𝗵𝗲 𝗿𝗼𝗮𝗱𝘀 𝗮𝗿𝗲 𝘄𝗲𝗹𝗹-𝗽𝗮𝘃𝗲𝗱: Bryan's framework for where AI lands first… well-understood patterns like checkout, front ends, and reorders, not judgment-heavy edge cases.

Bryan is a product guy after my own heart, and this one gets satisfyingly deep, from CPQ to protocols to the future of re-platforming. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

This feels like the busiest week of the year, we’re in that frantic ‘post labor day’ heading into conference season with Holiday 2026 bearing down on us - Thanksgiving is 72 days way - basically we’re in that critical '2 months left to get all our ducks in a row’ period.

Going to RetailClub or GroceryTalk or ShopTalk?

At ReFiBuy we have folks at all three shows - Jason and I are at RetailClub where we’ll be doing live podcasting of both Retailgentic and the Jason and Scot show. If you’re not going, don’t worry, we’ll be streaming many of these pods live from the show floor (well sand, it’s on the beach!) so be sure to subscribe to our YouTube channel and turn on notifications.

Subscribe now

Heading into Holiday 26…

This is a great time to update some of our big picture Agentic Commerce Optimization thoughts so everyone can focus on getting the best bang for the buck in this short time. The Good news? → The easiest thing to change and also the highest impact is how you describe your products to agents.

In this part 1 of the 2 parter we focus on: "Personal Shopping Agents enter the Arena”

In part 2 coming after RetailClub on 9/29 we’ll cover the rest of the update and have a corresponding podcast/video presentation that puts both parts together and gives you all the data and slides/graphics so you can use them in your Q4/27 planning.

• New: Latest datapoints on Agentic Commerce Scale and Growth

• Updated: Anatomy of a Product Card

• Updated: 5 Levels of Autonomy

• Updated: ACO Everywhere strategy map

• Updated: ACO 11 Steps Sequential Strategy

Personal Shopping Agents Enter the Arena

In the last two weeks we have seen an entirely new category of Agentic Commerce come seemingly out of nowhere that we call Personal Agents - these are horizontal agents meaning they do many more activities, such as email, calendaring, event planning, travel booking, but they do all anchor on and tout their Agentic Commerce capabilities. These are different than the vertical shopping agents:

Top Five Brand and Retailer Questions about Personal Agents used for Agentic Commerce

These agents are gaining traction rapidly and generating a lot of question for retailers and brands. I’ve rolled them into these 5 questions:

Q1: Should I block these?

Q2: How do I track personal agent transactions?

Q3: How do I optimize for personal agents?

Q4: How do these personal agents work and where are they going?

Q5: How is this different than Answer Engine-driven Agentic Commerce?

In Part 1, of this series we’re going to dig into and answer these questions, then in part 2, we’ll zoom back out and answer question 5 by updating what’s going on in the rest of the world of Agentic Commerce which is not holding still at all either.

To answer these questions, we need to dig into background, talk a bit about Instinct and GrokBot as those are new here on Retailgentic (We have detailed Muse walk through with purchase flow here).

Personal Agent Backgrounder

To answer the big five questions, we need to look at:

• The underlying exponential model improvements - it’s the rising tide lifting all the capabilities we cover here.

• History of Personal Agents - The history here will help us see where this is going and how fast.

• What’s a ‘Harness’? - You’ve probably seen this floating around, Harnesses and Loops are all the rage in the AI builder community. We stay focused on 'commerce’’ here at Retailgentic, but to answer the five questions, it will help to have a working knowledge of what a Harness is, what it does and what it doesn’t do.

• Two Modes of Agentic Browser Use - The most important part of these harnesses that impacts us and leads to the answers for our five questions is Agentic Browser Use.

Underlying it all: Exponential LLM Improvements

At this point, you probably gloss over the charts showing exponential growth of LLM intelligence. Some look like this:

Or this→

For example, Claude Haiku came out 10/25 and scores an indexed 638 on this meta-benchmark aggregator and now Fable 5.1 is at 7,798. The models are improving at a compounding 10x rate and everytime we think there’s a slow down, a new innovation, busts through that. This time next year, it will be at 80,000. 🤯

But, just like humans, the model’s knowledge is based on what it learns, or is trained on and this creates what AI researchers call ‘jagged edges’. LLMs like Fable 4.1, Sol 5.6 can write code for hours if not days now that’s better than humans, but some simpler tasks, they completely fail at. If you want the best examples of these follow ‘Husk’ on YouTube or X.

Here’s a fun example: (remember this model can also code for days on end)→

How can something so smart also be so dumb? Jagged edges.

One of the edges the model companies have been working hard on is browser/computer use - training the models to understand human interfaces so they can better do work on our behalf. In our world we’re most interested in the Research→Find→Buy cycle, but computer use includes things like: “Create a ticket in my crm system”, “open up SAP and create a new widget”, “Load up my PIM and translate this SKU into 93 languages.”.

In fact, The Information recently reported:

The root cause of the emergence of these super smart personal agents is due to the new (last 6-12 months) focus from the frontier labs on models trained for computer use.

There are ~50 different benchmarks for browser/computer use and this one aggregates them:

To put this in perspective, Gemini Pro 3.1 Preview (the one on the far right that scored 457) came out in 2/26. Today the models are scoring 1720 on an indexed scale, that’s a 4-5x improvement in 10 months. Many of the benchmarks have scores that have a % of tasks completed. These models are scoring 93%, 96%+ at tasks that are much much more complex than buying something online.

Part 1 of the story is in the last 12 months, the underlying agents have gotten 10x smarter at general intelligence and 4X smarter at computer use. Researching→Finding→Buying a product on a consumer website is child’s play**

**with the right context, memory, tools, etc.

What’s a “Harness”?

That ** above is important. Think of the models as raw intelligence - computer brains, but they kind of all over the place and wild, they need direction. A harness, borrowed from a horse harness metaphor, gives the LLM direction for a specific use-case.

The picture above illustrates how the harness ‘surrounds’ the agent and wraps it with the pieces it needs to guide the raw intelligence, keep it focused, keep it grounded, help it remember where it went right and wrong (writing things downs in a scratchpad turned out to be a huge breakthrough - who knew!).

Harness are increasingly adding advanced capabilities:

• Agentic Loop - On a timed or event-based basis, the agent wakes up and takes pre-determined actions. For example, when I get a new calendar invite, do x, y, z or closer to home: if I get an email about a product I’m looking for that’s less than the 30-day average, go buy it.

• Multi-player capabilities - On the consumer side, this is usually inviting friends and/or family to book with you via a Resy/OpenTable or flight/hotel/airbnb

• Compounding self-improvement - The harnesses understand their goals (“buy the widget!”, “book the hotel” - the models execute and the harness learns what context to feed in and what overwhelms the engine.

• Browser use - This is where the magic happens in our World, stay tuned for more details.

If you’ve used chatgpt, claude - those are light harnesses. If you’ve used Claude cowork or ChatGPT Codex→Work, those are heavier more meaty b2b harnesses.

To understand the consumer agent harness progression, we need to go way way back to the earliest days of consumer - way back to January of 2025 (or Oct 25 depending on how we’re counting).

Retailgentic’s History of Personal Agents

The pre-cursors to these agents is OpenAI Operator, in fact ChatGPT’s co-founder, Greg Brockman has been saying since early 2025 that ‘computer use’ is a top priority for ChatGPT→

Jan and July you can see OpenAI working on computer use with the introduction of Operator which became Agent. Then they folded that into Apollo - their browser Atlas (here’s our first Atlas LIVE Retailgentic demo if you want to jump in the time machine for a spin.)

IMHO, what really broke this open, was the introduction of OpenClaw in 11/25 which by March 26 had a thriving community of people extending the harness to Agentic Commerce.

After OpenClaw got progress, some ‘clones’ came out that simplified the install, added some multi-player improvements and beefed up the security. Examples in this category are: Hermes, Perplexity Computer and Microsoft Scout is transparently based on OpenClaw. At Google I/O they announced and launched Gemini Spark.

I’ve tried all of these it’s very hard to get the right mix of proactivity/safety/pedantically asking permissions.

On one side you have OpenClaw that’s scary aggressive about access to your computer and all your files. On the other end of the spectrum you have Google Spark which asks for permissions so much and for every task and sub-task that it’s basically unusable (you’ll see in the demo below).

The Q3_26 ‘class’ of personal agents broke through this. They have a great balance of getting up fast, not bothering you for permissions every five seconds for permissions, but asking permissions at the right break points (e.g. hey you’re about to spend money, review this transaction for me).

Two Modes of Browser Use: Server and Local

Back in the Agentic Browser wars days, we spent some time walking through how the first ‘computer-use’ agents used servers to do their browsing. The challenge with this model is it’s easy to block - the destination (e.g. Amazon) or the infrastructure layer (e.g. cloudflare) can detect the bank of IPs via fingerprinting and say: “Oh this is a Meta muse bot, I will shut it down”. What seems to always happens is these systems start with server-based (or a virtual machine, or VM) and then add the ability to use the local machine. When you’re on the consumer’s local machine, you are effectively unblockable - the agent’s traffic looks exactly like a consumer. As a countermeasure, companies are developing technology that looks at how fast you’re using the browser, mouse movements and what-not to try and decide: “human or agent”. The problem with this is ‘false positives’ - I frequently, especially when using some specific sites and retailers get put through Captcha sequence (pain) or the slidey-puzzle thing, etc.

Of note - all of these ‘prove you are a human’ tests have fallen with the growth in computer use, so they are placebos at this point.

This all ties to the ‘track-ability’ of personal agents which also ties to ‘optimization’ so It’s worth going through this diagram in more detail:

1. The user is on their local computer - they have what we call a consumer IP

2. They use a personal agent which, even if it’s a desktop or mobile app, is running in a server/vm.

3. That agent needs to do something with a browser, so spins one up in it’s VM (usually chromium-based so it “looks like” chrome).

4. For Agentic Commerce the browser tool decides if it wants to crawl the merchant’s website, use UCP (or some other protocol like ACP, WebMCP, etc.) for discovery and transaction or some mix of those.

5. Alternatively, the agent may decide to connect back to the consumer’s computer and use a local browser, a browser extension, or IP routing.

In fact, literally as I was working on this post Elon conveniently posted this→

They very clearly say the intent is to get around server-side IP blocking - nobody is hiding the ball here.

Demos: Gemini Spark, GrokBot, Instinct and Muse

To demonstrate the agentic commerce capabilities of all these systems, I’ll briefly walk through Spark, GrokBot and Instinct to show you the experience.

The Veggiedent test

Long time readers have probably recognized that one of my favorite use cases is the dog treat →

On the surface this seems easy, but the lowly veggiedent has broken many an agent because, just like all of e-commerce, there is a lot of complexity hidden in this image:

• These are not widely available

• They have these variations:

  • FRESH and 2 other ‘styles’ of chew

  • Dog size - this is small 11-22lb

  • Pack size - this is the 30 chew

Agents struggle to get this right - spoiler alert, at least one of our agent demos will not survive the Veggiedent test!

Also, I happen to know that Chewy and Amazon battle it out for this SKU and they are two of the best blockers of agents.

Gemini Spark Walk Through With Transaction

After finding 10mins how to get to it (mobile only, personal account not business, pro or something, hidden in a menu, accept/yes/yes/yes/yes/yes/yes/yes/ok/upload pick/accept/ok/ok/ok/phew). I was able to get to shopping:

GrokBot Walk Through With Transaction

I’m in very deep in Grok Bot for work usage - i have a fleet of 15 bots doing things for me every day and collabing on many projects - it’s amazing for that use case. But let’s try it for ecommerce. I go in and setup a bot, I called mine cleverly: “Scot’s Shopping Bot”

I’m including the setup here because it’s pretty cool and gets a lot of the nagging questions out of the way in one big chunk. The most interesting one→

You can see I freed my shopping bot to buy anything under $50 without bugging me. Let’s hope that doesn’t backfire.

Only Chewy was able to block GrokBot - kudos to the team for figuring out how to get Amazon to work - not easy. (Note: I’m not using the local IP, not available to me at the time of this writing, that would get around the Chewy block though for sure).

From here let’s look at the Amazon buying flow:

To finish, I have to login to my amazon account - I can give this info to Grok to hand over to Amazon or take over. Both Muse and GrokBot have secure VMs that the LLM can ‘write to’, but not read from so their claims are that your credentials are secure. Personally I’m increasingly becoming a fan as a consumer of the Stripe Link System - it’s one click that has a complete review and you’re done - pretty amazing. Speaking of which, let’s look at Instinct now.

Instinct Walk Through With Transaction

First of all, Instinct is in private beta and you need an invite code, they are pretty hard to come by, but if you’ve made it this far and need one - DM me on LinkedIn - we have a pool of Instinct Invites here at ReFiBuy HQ for our BFFs.

Instinct runs right in your iMessages and it’s best characterized as “insanely proactive”. You have to be very clear with it or it will go and do things you may not have intended. But it always has the best intentions and that proactively has really been a game changer for me - it’s saved my 🥓 a couple of times and I've been using it 2wks.

Instinct famously is dealing with a crushing load of new users and can’t get servers fast enough, so there was a glitch with the image upload that was easy to fix→

Notice it got through to chewy, but not Amazon in this use case.

Then I clicked on the link to see the product before buying and…

Womp womp - Instinct has a bit of a ready/fire/aim thing going on and it got caught in the Veggiedent trap - (these are the extra small vs. small) - third variant got it.

Lesson - always check the transaction before approving. Speaking of which, I ordered a cooler using Instinct for tailgate season and, like Muse, it uses the Stripe Link connection for payment which I really am enjoying→

When I click on that link (I won’t show that as it has tons of PII), it lays out the whole transaction and once you authorize it and TFA, it magically works.

After you order something, Instinct checks your email and pings the order site proactively so you know where it is in it’s journey to you.

Some people have even wired Instinct up to a phone calling service and have it call merchants and do things IRL for it - wild stuff, on my weekend fun experiment list.

Muse Product Walk Through

Ok, I realized that I didn’t show the Veggiedent example in my Muse walkthrough so let’s do that real quick to be fair→

From there, Muse found a $20 off and processed my order with a stripe link!

Recap: Muse vs. Grok Bot vs. Spark vs. Instinct

Spark came in last for me: while it didn’t trip up on Veggiedent, it is so far behind the others, capability-wise it has a lot of room to make up. GrokBot is more a business agent for me, but it did well.

Muse is the real standout. I’m not sure why Chewy/Amazon aren’t blocking it yet, perhaps it’s early, or, if you’re META, you have one infrastructure for FB, WhatsApp, Instagram, Threads, etc. all that traffic, including Muse looks like ‘meta family traffic’ -perhaps it’s hard for the blockers to separate that out?

The real winner: Distribution

The hardest thing to conquer with a consumer business is getting consumer attention. In this field of 4 Gemini and Muse have a huge advantage - existing billions of consumers. All you have to do inside one on of the big existing platforms with massive distribution is put a little ‘try Muse now button’ and BOOM, off to the races. In fact, there’s early evidence they are doing exactly that→

Back to the Five Big Merchant Questions Personal Agents and Agentic Commerce

Q1: Should I block these?

A1: For me there are 3 angles on this one: retail/dtc brand/non-direct brand

Retail Answer: This is the toughest question you face right now and comes down to theology and where you are as a company. You see Amazon, Chewy and many of the top retailers in a category increasingly actively blocking. The argument here, is “I don’t want to be a logistics company and give away the front door.”

DTC Brand Answer: For DTC Brands, they are always looking for more distribution, so they tend to open the door wide. Because many of them both DTC and wholesale. they are already used to ‘giving away the front door’, so theologically they are ok - the argument here goes: “meet the customer where they are” and where they are is increasingly using Agentic Commerce across all these new surfaces.

Non-DTC Brand Answer: This is the easiest one - you should make sure your brand site is open so all agents can get all the product info they can consume and work with your retailers to optimize for their on-site agents and if they are embracing other outside agents (Walmart is prime example), collab with them on that.

At the end of the day, the theology question seems to come down to: How badly do you want new customers? Can you afford to lose the sale to a site that doesn’t block?

Q2: How do I track personal agent transactions?

A2: In their current configuration you can see with GrokBot, where i’m using my credentials, all you see that’s different is a new login from a new device. Eventually GrokBot’s IP range will become known, but again, they are already implementing local which gets rid of any hope of tracking that model.

I think the good news is once you start using these super efficient, fast and easy agents, you open up your mind to getting out of the middle of the checkout - the Stripe Link thing got me over that very quickly.

We’re looking into specifics, but there’s definitely a way to track that so at least you can bucketize “agents that paid with Stripe Link” and put a number on this category and see how big it gets, bounce it against your customer file and see the existings vs. new and what-not.

Q3: How do I optimize for personal agents?

These bots all see the meta data on your site. There’s speculation Muse is already or soon will start looking at UCP, they use it in other parts of the FB family. The WebMCP people see a world where everyone switches to that.

To be honest, the computer/browser use is so good, I’ve been using this very heavily for 30 days and it’s not getting stumped with anything. The discovery process does break down sometimes and you can see some of the agents are using intermediate product databases (Muse and Spark do this as best I can CSI) and in there they maybe using the UCP part of discovery.

We’ve said this here maybe 1000 times, but it works here too - the single thing you can do regardless of retailer/brand/etc. is work to make your catalog agent friendly - keep chipping away at those 5 levels of enrichment (details here).

The harnesses need context and content to chew on - feed them well!

Q4: How do these personal agents work and where are they going?

You can see in the timeline here that in just 12 months these agents have gotten very sophisticated. Using the Retail 5 levels of Agentic Commerce Autonomy as a guide→

These consumer agents are squarely straddling 3/4/5 with different elements. Where they need to go next:

• A ‘wallet’ for your favorite preferred types, loyalty programs, etc.

• Multi-cart/wish-list and deal watching (like Gemini multi-cart)

• Extracting your purchase history from email (they can do it, but requires prompting and they are forgetful compared to Answer Engines)

• Prediction of needs - Instinct is getting there, very cool and helpful.

Q5: How is this different than Answer Engine-driven Agentic Commerce?

That is the main topic of Post 2, we’ll be back after RetailClub!