Adobe is out this am with an update to their AI Traffic data and we have a complete analysis. then, for the first time they have released AI Visibility data that is an urgent reminder of the importance of ACO. Let’s dig into the AI traffic/retail traffic trends first…

Adobe AI Traffic Data Update

In Q1 2026, Adobe reports traffic from AI sources to U.S. retail sites jumped 393% year over year. In March alone, it was up 269%. As a reminder, during the November-December 2025 holiday period, AI traffic surged 693%.

That’s the top-line number, when you dig deeper, it gets better.

The Great AI Conversion Flip

As of March 2026, AI traffic is converting 42% better than non-AI traffic (paid search, email marketing, affiliates, etc.). That’s a new all-time record in the Adobe data!

To understand the historical implications of this datapoint, if we look back at March 2025, AI graffic converted 38% worse than traditional channels. That’s an 80-percentage-point swing in conversion in 12 months 🤯.

What’s driving this surge in AI conversion?

Another trend we’ve been tracking for a year is the trust relationship between Answer Engines and shoppers.

Adobe’s companion shopper survey (5,000+ U.S. respondents) found that 66% of consumers believe AI tools provide accurate results. And 39% say they’ve already used AI for online shopping, with 85% saying it improved their experience. When shoppers trust the AI recommendation, they buy.

Only one thing can explain this 80% swing in conversion rate - an Agentic Commerce flywheel -

• Answer engines are getting better at recommending products

• Because they have better shopper data (memory) and product data (ACO)

• Better shopper/product data is increasing trust

• More trust increases conversion rates, that closed ‘re-enforcement learning with AI Feedback (RLAIF)’ loop gets smarter and smarter

The result: an unprecedented 80% swing in conversion rate in one year.

Agentic Coding (Codex/Claude Code) use a similar RLAIF closed-loop flywheel system, but they’ve been at it longer and we’re in the phase where each ‘spin’ of the flywheel is creating huge gains in capabilities. New models (Mythos+SPUD) will even further elevate this. In 6 months we could see another 40-60% improvement in conversion.

Sidebar: Life on the Exponential Growth Curve

Like many of you, I’ve been in ecommerce for over 20yrs. We’ve experiences periods of growth like this in the early days of ecommerce and then it slowed dramatically - the standard ‘rule of large numbers’. Even Amazon has succumbed to this rule. That’s what our brain tells us is going to happen.

But it’s different this time. What’s happening now is we have just seen clear evidence two key factors of ecommerce are simultaneously growing (traffic and conversion) and one of them, conversion-rate is accelerating. We’ve never seen a RLAIF flywheel before. This is new territory.

The improvement we are seeing across AI happens when not only does one factor improve rapidly, but 2,3 even 4 factors all improve at the same time. When you multiple these you get an exponential curve.

Exponentials are hard to predict because they accelerate when we’re used to the ‘law of large numbers’ kicking in, they create surprising, even shocking results. Failure to realize you are on an exponential and you don’t invest in front of it can have faster and more severe ramifications than we’ve ever seen.

I know, I know, that experience we all have makes it hard to imagine something like this. Here’s a real-world example happening right now - using the exact same RLAIF flywheel→

PLUS: AI Shoppers Are Better Shoppers

Not only are traffic and conversion rates growing at a frenetic pace, but the Adobe data reveals 3 more factors that are more evidence we’re on the exponential":

It’s not just conversion. Adobe’s March 2026 data shows AI-referred visitors outperform across the board:

• +42% higher conversion rate vs. non-AI channels

• +48% more time spent on site

• +13% more pages viewed per visit

• +12% higher engagement rate

Now let’s look at the new data in this report - US Retail AI Visibility:

Adobe US Retail AI Visibility Data

The scorecard above shows the average score across the entire retail sector using Adobe’s proprietary new AI Visibility data. We assume they are looking for the basics on each of the pages of a website: bot blocking, page readability, metadata for javascript-hidden content, etc and then rolling them up, by category, into this report-card style scoring.

Here you can see the top performing pages are Returns/Exchanges, Contact Us and FAQ all above 80% - these are typically text-heavy pages so this makes sense.

In the 70-80% zone we have: customer service, loyalty/membership, homepage and cateogry pages.

Finally, the most plentiful pagesof any ecommerce site, the PDP, scored a terrible 66%!

Adobe Data Reveals: The Invisible Shelf

In the visibility data, Adobe doesn’t reveal the range of outcomes for the PDPs, but for homepages at 75%, they do show the lowest performing retailers are as low as 54.2% and the highest at 82.5% - that’s a ~28% swing. Homepages are pretty basic and other than product carousels they tend not to be javascript heavy.

I’m sure in future reports we’ll see it, but my educated guess here is the range of outcomes is at least 2X from homapages - that’s a 56.6% swing with a midpoint of 29.3% - that would mean PDPs are as low as 37.7% with the midpoint at 66%.

Critical Action Item for Q2/Q3: ACO!

Since 9/25 here in the virtual pages of Retailgentic, we’ve been advocating for something we coined: Agentic Commerce Optimization. We released our 9-step framework:

Back in September of 25, this was a ‘prepare for the future’ type scenario. Sitting here in April 2026 staring at Holiday 26 bearing down at us with AI traffic surging 393% y/y and conversion rates on that traffic swinging +80%. In 8 months, ACO has moved to a ‘must have’.

Note, we also have this updated and expanded ACO guide here.

Conclusion

In real time, we’re watching the internet get re-intermediated. Google built a $200B+ business as the intermediary between shoppers and brands. Now Answer Engines are becoming the new intermediary — and unlike Google, they don’t just give you a list of blue links. They make recommendations. They pre-qualify. They hyper-compress the funnel.

Adobe’s new data makes the economic case undeniable: AI traffic converts better, engages more, and is growing nearly 4x YoY. But here’s the twist — only the retailers who invest in their product catalog, optimize their product cards and start providing extra context to engines, and get on their own RLAIF curve will ride this wave vs. missing out on the biggest opportunity since the dawn of ecommerce. 🚀

Joseph Lee is the Co-Founder and CEO of Delectable AI, a vertical agentic commerce platform built specifically for grocery. He brings nearly three decades of experience in enterprise commerce, having held leadership roles at companies like ATG (acquired by Oracle), Hybris (acquired by SAP), VTEX (IPO), and Mirakl.

Joseph has spent his career at the center of e-commerce infrastructure, and is now applying that experience to one of the most complex and competitive categories in retail: grocery.

Listen/Watch Our Interview:

In this episode, we go deep into what agentic commerce actually looks like in a real-world vertical, and why grocery may be one of the most important battlegrounds.

We walk through the shift from “tools that help you shop” to systems that shop for you, and how that fundamentally changes everything from discovery to monetization.

We also unpack the tension between grocers and platforms like Instacart, the rise of retail media networks, and why owning your data, and your customer relationship, may matter more than ever in an AI-driven world.

Highlights

• Regional grocers are losing both data and margin to platforms like Instacart

• Agentic commerce shifts grocery from manual planning → “do it for me” automation

• Delectable combines food intelligence + shopper intelligence + real-time inventory

• AI closes the gap between inspiration (TikTok/recipes) → instant cart creation

• Vertical AI beats horizontal models with better accuracy, speed, and cost efficiency

• Grocery e-commerce hasn’t saved time, until now

• Retail media networks are massive, but broken by middlemen and inefficiency

• Agentic advertising creates unlimited, contextual ad inventory

• The future of grocery is about owning discovery, not renting it

If you want to understand where agentic commerce gets real, beyond theory and into infrastructure, this is one to pay attention to.

🎧 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!

Arjun Bhargava is the CEO and Co-founder of Rye, a company building the infrastructure layer that enables AI agents to actually complete purchases online.

With a background in software engineering at Goldman Sachs and Reddit, Arjun is now focused on solving one of the hardest problems in agentic commerce: turning intent into a real, fulfilled order.

Listen/Watch Our Interview:

Everyone is focused on how AI helps people discover products. But what happens after that?

In this episode of Retailgentic, we go deep on the missing layer of agentic commerce: checkout, payments, and fulfillment. Because it turns out, the hardest part isn’t finding the product. It’s buying it.

Rye is building what they describe as the “Stripe for commerce.” It connects AI agents, apps, and platforms to merchants handling:

• Checkout and order placement

• Payment orchestration

• Fulfillment across retailers like Amazon and Shopify

In short: if an AI agent decides what to buy, Rye helps it actually buy it.

Highlights

• Checkout is the real bottleneck: Discovery is improving fast, but actually placing an order across fragmented systems is still incredibly hard.

• Rye enables real AI purchasing: They connect agents to merchants, allowing products to be bought directly from platforms like Amazon.

• Amazon is a critical unlock: Most bots are blocked, but Rye’s API-based approach allows real transactions to go through.

• Protocols won’t solve everything: Even if UCP/ACP take off, long-tail merchants and major players won’t fully standardize.

• B2B is the sleeper opportunity: Complex procurement decisions are a perfect fit for agent-driven commerce.

• AI reduces returns: More context leads to better decisions, driving return rates extremely low.

• We’re still early, but close: One breakout use case could flip massive volume into agentic commerce quickly.

Agentic commerce isn’t just about what gets recommended, it’s about what actually gets bought.

🎧 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!

Editorial note: To keep this article’s vibe flowing, we’ve detailed the eight referenced resources at the end of the article if you want to dig into this topic more. Also, we’ve done this enough times with GA4 that we are happy to help anyone that would like to do an analysis with their own data. Join our Retailgentic Slack community here and let us know in the #dark-agentic-commerce-traffic channel.

Retailer’s Unexplainable 2026 Traffic Trends

We had a scenario recently where we were working with a large fashion retailer and they were using ReFiBuy’s Commerce Intelligence Engine to implement our 9-step Agentic Commerce Optimization framework. Their product-card visibility and offer-card positions were all showing very strong improvements at the product-level. But they weren’t seeing an increase in AI referrals 🤔.

Simultaneously, organic/paid Google traffic across all merchants continues to decrease as we accelerate towards a zero-click future. It’s outside the scope of this article, but some good current reads on this are here and here (we believe publishers/b2b are a canary in the coal mine for merchants).

At the same time, traffic identified as ‘direct’ traffic is inexplicably growing faster than ever before. Engagement on ‘direct traffic’ is unusually high resulting in super-sized conversions

This is not a tagging error. That’s what we call dark agentic commerce traffic. Let’s dive deep into this one, given that we are are already 233 days away from Thanskgiving, we strongly recommend getting a handle on quantifying and tracking your DACT to create a baseline now enabling you to create a before/after for Holiday 2026.

What is Dark Agentic Commerce Traffic (DACT)?

When a user opens the ChatGPT mobile app, asks “what’s the best trail running shoe under $150 for wide feet,” gets a recommendation with a product link, and clicks through — that visit arrives at your PDP with no referrer. The referrer metadata and any UTM tags are stripped before the request is sent.

GA4 (Google Analytics 4) sees a new user landing on the /products/[shoe-name] PDP page with no referrer. It then puts them in Direct and you never know ChatGPT sent them.

🤯 The kicker

In controlled testing, when the Gemini iOS app sends traffic to a website, only 5 out of 56 visits (under 9%) are correctly identified as AI referrals in GA4. The other 91% land in Direct. (Source: Topify)

Which Traffic Sources Are The Culprit?

We’ve reviewed the research and done some of our own tests and the above matrix is what we believe to be the current state. Basically summarized - if someone is using an AI ‘app’ (both mobile and desktop versions) - that traffic is nearly 100% ‘dark’. For browser based traffic, that traffic is mostly trackable. The Answer engines want you to use their app experiences so they tend to put exclusive features in there. There’s no great reference that looks at which ‘surface’ shoppers are using for Gemini or ChatGPT, but I can guarantee your GA4 analytics is telling you that your AI referrals are 95%+ desktop. But we know that across e-commerce in most categories we’ve passed 50% mobile traffic and corresponding sales.

Is This Bot Traffic?

Note that this article is 100% relating to humans and not bots. 99% of bot traffic is easily identified and pre-filtered out by analytics programs, it’s identified by the ‘user-agent’ - bots/crawlers have to identify themselves as a specific user agent (we went into this back in this article if you want to learn more). These are most likely humans, unless someone is using an agent to manipulate a browser on their local machine which, outside of the Openclaw community, is pretty rare for your average shopper and would be in the browser side anyway, not on the app side. Because of this we’re certain these are real living breathing shoppers we’re talking about in this analysis.

Subscribe now

What’s the Scale of Traffic We Are Missing?

Loamly ran the most rigorous controlled analysis to date — 446,405 tracked visits — and found that only 29.4% of AI-influenced traffic carries any referrer at all. The other 70.6% lands as Direct. Apoorv Sharma at DerivateX validated this independently. Topify confirmed it with the Gemini iOS testing already mentioned.

Three independent datasets, same number. That’s not noise. That’s signal.

Now here’s the part that should make your head explode. That dark AI traffic — the stuff GA4 can’t see — converts at 10.21%. Average ecommerce traffic converts at 2.46%.

Therefore, What is essentially your best-performing traffic source is completely invisible and not attributed to anything. Note to our friends on Agentic Commerce teams - you should work on establishing credit for this ASAP.

What’s the Growth Trend of DACT?

Visibility Labs (via search engine land report) spent all of 2025 analyzing 94 ecommerce brands — 9.46 million non-branded organic sessions vs. 135,000 ChatGPT referral sessions. What they found:

• January 2025: 1,544 ChatGPT-attributed sessions/month across the dataset

• April 2025: ChatGPT launches shopping carousel → hockey stick begins

• December 2025: 18,202 ChatGPT-attributed sessions/month — +1,079% YoY

• CVR advantage: ChatGPT referral converts 31% higher than non-branded organic (1.81% vs. 1.39%)

• Revenue per session: $3.65 from ChatGPT vs. $3.30 from organic — despite lower AOV

Which Agentic Commerce Engine Causes the Most DACT?

Not all answer engines are created equal here. ChatGPT dominates referral share but is a mixed bag on stripping referrers.

ChatGPT is the dominant channel — Lantern’s data shows 87.4% of all AI referral traffic comes from ChatGPT.

Gemini on mobile is basically invisible in analytics. With 91% of its traffic landing as Direct, every Gemini AI Mode product recommendation that converts is almost certainly being credited to “Direct” or “Branded Organic” — robbing your agentic commerce optimization efforts of any signal.

Perplexity is actually the best citizen here. With only ~30% landing as DACT, Perplexity sends cleaner referrer data than its peers. This is one reason Perplexity’s referrer data shows up more clearly in GA4 reports, even though its absolute share is smaller with shoppers.

How to Track Your DACT In GA4

There’s no silver bullet here, we can’t make the various answer engines stop ‘not’ sending referrer data (perhaps when they have more ad revenue/focus this will change, but we can’t control that).

What we’ve helped customers do is:

• Understand and track the distinct ‘fingerprint’ for DACT

• Create a GA4 pre-Agentic Commerce Baseline and then measure from that baseline.

The DACT Fingerprint

The traffic that you can safely assume is DACT has these X characteristics:

• It is categorized as Direct or Brand (where traffic is categorized when referrers there are no referrers).

• It lands directly on a PDP page (not home/about/category pages - deep into the site).

• It has a low bounce rate and converts relatively high.

The Results

Based on this finger print, we created a baseline back to January of 2023 - days after the birth of ChatGPT, it didn’t have web search yet, Let’s call this X - and you will find it’s a very low number expressed either as a % of your traffic or as sessions.

Then we track this DACT traffic through March 2026. The results across many customers reveal that “Known AI sessions” (flagged by GA4 because they have referral tracking) are off by 100x of our estimates from the DACT traffic. For example, on a sessions % share basis if Known AI sessions are tiny at .12%, when you consider DACT fingerprinted traffic increased from baseline using GA4, that number grows to 12%.

I can also share that this traffic has an mind bogglingly high conversion rate and a very low (sub 20s) duration. Shoppers are hitting that PDP, adding to cart and moving on to checkout or to shop more of the assortment. As theorized, DACT is most likely your highest quality traffic and you’re not tracking it outside of Direct/Branded correctly.

As discussed from third party research, we can also confirm that the overwhelming majority (90%+) of traffic in the known AI referral bucket is attributed by GA4 to ChatGPT.

As mentioned at the top - we’ve done this enough times with GA4 that we are happy to help anyone that would like us (or help) to do an analysis with their own data. Join our Retailgentic Slack community here and let us know in the #dark-agentic-commerce-traffic channel.

One More Bucket of DACT: Influence Attribution 🤦‍♂️

Here’s where it gets even more mind boggling. There’s a second layer of dark influence that’s impossible to measure even with this fingerprint/baseline trick.

For example, what if a shopper asks ChatGPT for trail running shoe recommendations. ChatGPT names your brand. The user does all their research on ChatGPT, then opens Google. She searches “[your brand] + [product name]” — a branded query. They click your organic result and buy.

GA4 attributes that sale to Branded Organic Search. ChatGPT gets zero credit. Forever.

Interestingly, the Visibility Labs study found this pattern to be extremely common. They call it “intent compression” — Agentic Commerce Engines pre-qualify the shopper so thoroughly that by the time they hit the product page (via whatever path), they’re at the very bottom of the funnel. Your traditional channels are getting credit for closes that Agentic Commerce drove 90%+ of the influence on.

I do believe eventually, for this very reason, Agentic Commerce Engines will ‘pull up’ the transaction into the engine. In other words, I don’t think we’ve seen the last of ChatGPT Instant Checkout, especially if Google’s UCP checkout implementation has a strong Q4 in 2026.

Conclusions

DACT isn’t an analytics nuisance. It’s a signal that the ecommerce funnel has moved — and it’s moved to Agentic Commerce.

Shoppers are asking ChatGPT what to buy. They’re asking Gemini to compare products. They’re asking Perplexity to research specs. And then they’re landing on your PDPs — pre-qualified, high-intent, and ready to convert at 10x the rate of your average visitor.

The merchants winning this transition are making sure their products are showing up and giving the answer engines as much product-level content and context they can so that their products are recommended as that shopper works down the funnel.

Shopify’s Harley Finkelstein recently put a 7x traffic number on DACT.

Agentic Commerce isn’t coming. It’s here, hiding in plain site.

Subscribe now

Dark Agentic Commerce Traffic Resource Resources

This article references these nine recent and well researched sources:

1. ReFiBuy Internal Proprietary Anonymized Client-level Research, performed by Link Walls (inspiration and referenced here.

2. Armando Roggio — The Truth about ‘Direct’ Traffic , Practical Ecommerce — Feb 15, 2026

3. Apoorv Sharma (DerivateX) — AI Search Traffic: The Invisible 70.6%, LinkedIn — March 3, 2026

4. Loamly — State of AI Traffic 2026: Industry Benchmark Report, Loamly.ai — Dec 13, 2025

5. Dan Taylor — Why Direct Traffic in GA4 Isn’t What It Looks Like, MarTech — March 9, 2026

6. Yotpo — Track AI Referral Traffic: 9 Expert Tips (2026), Yotpo.com — March 25, 2026

7. Danny Goodwin — ChatGPT Ecommerce Traffic Converts 31% Higher Than Non-Branded Organic Search, Search Engine Land — Feb 26, 2026

8. Topify.ai — The Traffic Analytics Tools Every Marketer Relies On in 2026, Topify.ai — April 4, 2026

9. Laurence Nixon - Agentic Commerce is Here, fabric.com blog, Dec 2025.

Debra Aho Williamson is the Founder and Chief Analyst at Sonata Insights and one of the most respected voices in digital advertising.

In her 19+ years at eMarketer, Debra helped define how the industry understands social media’s impact on marketing. Today, she’s focused on the next wave: how consumer AI behavior is reshaping advertising, media, and commerce.

Her Substack, The AI Ad Economy, tracks the fast-moving intersection of AI platforms and advertising strategy and we recommend it here→

The AI Ad Economy

The AI Ad Economy follows the money as AI transforms how brands and retailers reach consumers.

By Debra Aho Williamson

Listen/Watch Our Interview:

In this episode, Scot sits down with Debra to unpack what might be the most important shift in digital advertising since social media.

This isn’t theoretical. It’s happening right now, and brands that don’t understand it risk getting left out of the new chain of commerce.

Highlights

• Why ChatGPT’s ad rollout is just the beginning, not an experiment

• The tension between trust vs. monetization in AI interfaces

• How retailers are both excited and terrified of AI shopping agents

• The emergence of “AI Media” as a new category (like social + retail media before it)

• Why current ad formats are “boring”… and what comes next

• How companies should actually prepare (hint: it’s not just media buying)

• The shift from advertising to agent influence, and what that means for the future

We’re still early, but not that early. The rules of digital advertising are being rewritten in real time, and for the first time in decades, the winners aren’t obvious.

The question isn’t whether AI will reshape commerce. It’s whether your products, and your data, are ready to compete when agents decide what gets bought.

🎧 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

Adweek Covers First ShopTalk Debate

Adweek had a great piece summarizing the debate. Note if you missed it, we released the audio last week here. Here’s a good summary of Ekta’s take:

Gemini Updates

Google Gemini “Sponsored Store” Ad Unit Spotted In the Wild

Back on Feb 11, Google’s VP/GM of Ads and Commerce, Vidhya Srinivasan, posted this blog post and announced a new ad unit called “sponsored store”.

At the time, we questioned how this will work, is it basically going only be sponsored stores on this listing or will there now be ‘tiers of stores’? A sharp-eyed X user found a live example in the wild.

In this example, Gap is the only merchant with the offer, so they are effectively listed twice. If we pretend that StoreX also had this item, but wasn’t sponsored, they would be listed either above or under the ‘All stores’ designation. This is a pretty valuable ad unit, it’s way up top, very delineated and basically doubles your shelf space. It will be interesting to see how it’s priced and how retailers are going to be thinking about the ROAS on this one.

We tried in a variety of accounts to try to replicate this one, with no success, so it maybe a A/B test only exposed to a small flagged audience. If you want to try it:

• AI Mode

• “I am looking for gap men’s chinos”

GAP and Google Partner for UCP and Agentic Commerce

Tuesday, on-stage at Shoptalk, the GAP’s EVP and CTO, Sven Gerjets was on stage with CNBC’s Gabrielle Fonrouge and they announced that they are the first major fashion brand to jump in feet-first with Gemini/UCP.

I think this is spot-on and really echo’s what Ekta said in the debate on the same day - consumers have a new way for researching and eventually finding and buying products, the forward-looking retailers and brands should be leaning into these experiences, testing and trying everything and keep an open mindset. I think Sven’s mindset here is spot-on and bodes well for GAP’s future in Agentic Commerce.

The article does say the checkout isn’t live yet, but ‘imminent’ - we’ll be keeping an eye on this one and see if Google and the GAP use this big launch as an opportunity to show off the four new UCP features announced March 19th:

• Multi-item cart capability

• Enhanced catalog endpoint

• Loyalty program integration (via a capability called ‘Identity Linking’)

• The launch of a self-service merchant center

In true Retailgentic style, we’ll let you know when it launches!

They did release a screen shot which is here. Note there’s a partnership with a sizing company called bold metrics.

Here you see a typical variation picker (color/size) and then the size picker. The size thingy pops in there and tells the user that it’s a high confidence match which is a clever addition to the checkout flow and illustrates how flexible and customizable this can be with the right underlying investment in infrastructure.

META Updates

While on the surface, this group of announcements covered by TechCrunch are really AI additions to traditional ad formats, I thought this ‘one click checkout’ shoppable ad unit was interesting→

A couple of things stand out about this and I think we’re going to see this ‘treatment’ showing up in Meta’s Agentic Commerce Offering. Here’s some observations:

• It let’s you pick the variations (beige/medium) in the ad. Therefore, this is an ad unit tied into variation-level inventory - that is no easy feat.

• This is a merchant-of-record purchase - notice the clear messaging that the shopper is buying from “Wind & Wool” both on the button and under in the Ts+Cs

• This ad unit keeps the user in the instagram app. In the old format, where you may ‘click out’ from Instagram to a Shopify checkout and not come back into the app.

• Techcrunch credits this on: “built in partnership with payment providers Stripe and PayPal, which allows consumers to complete the purchase in just a tap”. Notice that Shopify isn’t mentioned, their “checkout register” did not ring on this one. 🤔

• This would be a great way to further compress the Research/Find/Buy cycle - imagine this in a product card, where you click and for certain items, you can checkout right in the product card.

More Evidence of Meta Joining the ACP Alliance?

Last week, we highlighted speculation that Meta would pick ACP over UCP. This week, Joe our Agentic Commerce Chief Sleuther, found that on github, this user ‘ihwa’ works a facebook and has some fresh pull requests (indicating they have resources working on the protocol and now actively making code contributions).

Will see an official announcement soon? Q2 ends tomorrow, therefore we’re < 45 days from earnings and META has a history of dropping this type of news on that cycle.

Subscribe now

Big ChatGPT Agentic Commerce Updates

To recap the big picture, early March saw a diversion of paths between ChatGPT/ACP and Gemini/UCP:

• March 4th - Information drops piece that ChatGPT is focusing on discovery over checkout and then there was a bit of a dramatic Walmart breakup around the checkout

• March 11th - Conversely, Gemini stays the course with in-line checkout via UCP and announces the 4 things that address the obvious shortcomings with everyone’s first attempt at a hosted checkout: loyalty, multi-cart, self-service merchant center and improved discovering capabilities.

I think this is a good thing for the industry because we now have one of the big players really focusing on this checkout to see if it can be solved and we have another one really experimenting a lot with discovery. In the Google updates above, you see them refining this with the ad unit and picking off one of the big fashion brands to be an early ‘anchor tenant’. In this section we’ll update you on all the things ChatGPT has been doing since that March 4th announcement. There’s so much going on here we’re going to lightening round this one with fast short updates

ChatGPT Apps Refresher

The ChatGPT Apps UI is, to be kind, a bit hard to navigate. We receive tons of messages from readers that get lost in the UX. Here’s how it works on desktop:

Steps to Add an App:

1. You may have to start by Selecting the ‘+’, then go to “… More” and then click on ‘Add Sources”

2. From there (as pictured) in step 1, you choose apps and then ‘Connect more’

3. that takes you to the apps section of your settings (you can also enter this through settings BTW) - in there you see your existing apps and click the ‘Add more’ button on the top.

4. From there now you are in the ChatGPT apps ‘discovery’ experience - there’s a browse experience and a search experience (box in upper right).

ChatGPT and the Sephora App

First up, at Shoptalk, Sephora took the stage and announced the Sephora App. It’s unfortunately not live yet, but here’s the features:

• The Sephora ‘Beauty insider’ loyalty program will be completely integrated

• You’ll do discovery on ChatGPT and then ‘finish’ in Sephora. You’ll see an example of this in a second.

• You’ll be able to ‘at or ‘@’ the Sephora app and ask beauty questions

While the press release and coverage states the app is live, there’s no indication of it anywhere I can find and no screenshots -we’ll keep an eye out for this one.

I thought these quotes were interesting→

ChatGPT and the Walmart App

Walmart is on the record as saying that conversions from ChatGPT were 1/3 their on-site conversion rate and the app will solve that. In this video I take it for a spin and illustrate how it works→ (~7mins)

To summarize the two use cases:

• Pecan pie ingredients - Not a great experience, unlike the integration with Instacart’s app, this is cumbersome and the walmart app doesn’t “listen” to the instructions I’m giving it through ChatGPT.

• Big Screen TV - Same feedback, the whole point of having Walmart is everyone (in my neck of the woods) is < 10mi from a Walmart so it’s perfect for the urgent in-store stuff. How come that basic use case isn’t supported? Gemini and ChatGPT checkout support/supported that so we’ve taken a step back IMHO.

ChatGPT’s Shopping Discovery Becomes More Opinionated

Joe the Agentic Commerce Sleuth put together this great ‘before and after shot→

Three Big ChatGPT Discovery Changes:

• New layout - First, a much cleaner, more visual, cleaner result peppered with product cards (3x3 carousel in the OG and now 8 in the new)

• Best Overall - At the top, ChatGPT has actually given a more firm recommendation vs. just showing you the pros and cons- we’re calling this the Best Overall Reco. This is, obviously, where you want your products to be (how? ACO people! Product-level Content and Context are the keys - search our archives for ACO if you are new).

• Top Four Grid - Finally towards the bottom, you will see a top four comparison matrix that’s very handy and completely personalized. If you told ChatGPT about your stride, and other things important to you as you look for shoes, those will be the rows in the matrix. Welcome to the Matrix!

ChatGPT’s Pinterest-ey Wall (or Browse All)

Joe doesn’t show it in the above screen shot, but now frequently at the bottom of either the conversation or after you click on a product card and see the offer card and scroll down, you’ll see either a 5 product card carousel with one marked “Browse all” or an “Explore more” section like this→

Once you click on this you are taken to a ‘Pinterest-esque’ view or an ‘Endless Aisle type display. What’s happening here is we now have an ability to look directly into ChatGPT’s core catalog and poke around. This is super helpful for, you guessed it, canonicalization! But let’s not go there today.

Sometimes you even get what we’re calling the “Browse All View’s Filtering System” at the top that lets you filter what shows up in the product wall based on various attributes provided (I’ve popped down the color menu in this Dyson Airwrap example

Here’s a short video that shows you just how ‘endless’ the endless aisle is.

ChatGPT’s New Product Offer Card Content and Context Expansion

Tons to go over here, some of it we knew was coming because we saw it being tested, but it’s all live now. First, there’s a ton more information on the product card, it’s so long now I had to chop it up into 4 pieces to show you:

The new sections of the Offer Card are:

• Offers - The Offers are sorted by ChatGPT’s proprietary algorithm (but easy to decode 😁)

• What to know - An ai-generated tight paragraph about the product and who it was built for (product-level storytelling - see, we predicted this!)

• What people are saying - Here you start with a summary of the reviews starting with positives and then any concerns. After that, somewhere between 2 and 10 reviews are presented - usually with a mix of positive/negatives. There seems to be some new filtering to take out the ‘reddit crazy’ - those reviews that are just written by someone, well, on reddit.

• Explore more - As mentioned above, this is a gateway into the new pinteresty wall - kind of an escape hatch - if you made it this far on this SKU and it’s not what you wan’t this allows you to go back up to ‘Find’ or even ‘Research’ from the Buy part of the funnel.

• Follow-up - This we saw in preview and it’s cool to play with now that it’s live. Here’s an example→

In this example, I’m looking for a Dyson Airwrap and got confused about which accessories come with this SKU. When you click on Follow-Up, it has this hovering pop up chat that has ‘selected’ in it (see red box) the SKU in question - a visual clue we are asking about this specific product’ and I ask my question. Then the answer, also in the hovering box is presented. This is pretty cool to see in practice and I now want to put 3 of them in there like Rufus and Gemini have and see how that works. It seems like the comparison matrix would be a great place to put some check boxes and when you check them it puts them into this type of setup.

ChatGPT Offer Card: “Best Price” Badge

This one is harder to produce but we’ve had enough people send it to us we feel confident it’s not just a test, but probably here to stay (at least for 2 weeks haha)

Here’s an example we found in the wild:

Here you’ll see are two offers Poshmark at $30 and Mercari at $35 and Poshmark has the new “Best price” call out. Note that it does seem to take into consideration pricing and only showing up if the price difference is well north of 10% savings. I’m sure this will be tweaked a lot over time.

BCG: Agentic Scenarios Every Marketer Must Prepare For

I thought this BCG article was pretty interesting and had some thought provoking scenarios and likely conclusions for how Agentic Commerce plays out. Here’s an example - they have this diagram detailing four possible agentic futures→

… in which you have Super-App in the upper quadrant where you have AI purchase decision and high market concentration.

You’re Caught Up!

Now you’re caught up and ready for more news, let’s see what the next week holds for us!

Subscribe now

Today we’re dropping a surprise podcast that is different than anything we’ve done before and a first of it’s kind!

Subscribe now

Shoptalk Spring 2026

I was invited by Ben Miller, VP of Content and Strategy to participate in the first ever shoptalk debate format talk. The topic - Agentic Commerce Will AI agents transform retail?

For Agentic Commerce:

On the pro agentic side we had Ekta Chopin, CDO at ELF Beauty and myself. If you’re not familiar with Ekta, she’s a thought leading CDO and has a great substack for other CDOs about AI and various other topics I recommend here:

Anti Agentic Commerce

On the ‘anti agentic side’ we had Sarah Marzano, Principal Analyst, Retail Ecommerce and Retail Media at Emarketer and Andrew Lipsman - independent analyst and consultant for media, ads and commerce.

How we made this happen…

Before the show I asked and received permission from Shoptalk and all the participants to turn the raw audio into a podcast. In the AI era, speed matters so we are getting this in your hands only 2 days after it happened. This debate took place at 12pm PDT on Tuesday March 24th at Shoptalk’s AI Stage.

How to Listen:

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

How to “Watch”:

BONUS: Spicy Shoptalk Recap in Under 25 Minutes With Rick Watson

Enjoying our Content?

You can’t find awesome Agentic Commerce content like this anywhere else, so:

Subscribe now

Or check out our other channels here→

• 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!

The release is sparse and is here.

Here is a screenshot they showed (sorry the lighting is a challenge)

Here is a brief video they showed:

Joe the ChatGPT sleuth has found these live in the wild, here’s a great example of the compare function.

Build With ReFiBuy Announced!

Also this morning at ReFiBuy we announced the availability of our new “Build With ReFiBuy” offering. We now have APIs, MCP and CLI integrations with our product catalog Evaluate, Enrich/enhance, Distribute, Monitor and Measure core capabilities. This allows partners, Agent Builders and dev-forward customers to utilize our capabilities in many different contexts. Details here.

More news as it breaks here at Shoptalk!!

Retailgentic at Shoptalk Reminder

Two quick reminders:

• Agentic Commerce Debate - I’m speaking at Shoptalk Tuesday March 24th at noon in the big Agentic Commerce debate - Agentic believers come cheer on Ekta Gupta and I - Naysayers - don’t miss lunch, you’ll want to head to the cateteria early 😄

• ReFiBuy booth - ReFiBuy’s booth is SU3 - it’s in the startup area by the food/eating area in the exhibitor hall.

Subscribe now

News Item 1: ChatGPT’s Checkout Confusion

I’ve received many requests from readers in comments, LinkedIn and DMs asking about my thoughts and predictions on the controversy around ChatGPT’s instant checkout changes/confusion. All I can say about the why is that is I have some conflicts I have to manage through.

There’s enough public information available , that I think you can form your own opinions and make a good prediction of what’s REALLY going on if you look at it through the right lens and the right information thats been presented. Here’s my assembly of the information presented and you can form your own opinions.

Information Asymmetry in the Ecommerce Industry

One general observation before we dig in here, I’ve talked to hundreds of reporters and folks outside the industry about ecommerce over the last 30yrs. Similar to the disparity between the consumer UX (so simple and easy!) and ecommerce back-end implement details (insane complexity required to deliver that simple front-end UX), it is extremely hard to get nuanced, complex information about ecommerce across to people not building in the space. Also complex, multi-faceted details aren’t really great for sound bytes and awesome headlines. Can you imagine this headline: “MerchantX says if only the canonicalization, variation matrix and supplemental contextual+conversational datafeeds were easy to implement with the identity linked loyalty programs, it would be easier to do X, Y and Z” - sadly not so riveting when you get into the weeds.

Timeline Refresher

• 9/29/25 - OpenAI ChatGPT announces Instant Checkout

• 11/30/25 - Walmart launches on Instant Checkout

• 1/25/26 - Retailgentic discovers ChatGPT multi-item/single-merchant cart system here and on 1/28/26 more details here.

• 2/8/26 - Retailgentic discovers the ChatGPT self-service merchent center here.

• 2/22/26 - Retailgentic discovers 2 more features for ChatGPT instant-checkout: more info and ‘follow-up’. detailed here.

• 3/4/26 - Information runs this piece: “OpenAI Scales Back Shopping Plans for ChatGPT”

• 3/18/26 - ChatGPT’s head of ChatGPT, Nick Turley, when asked a direct question about Instant checkout dodges the question and says they are focusing on discovery right now. Our coverage here.

• 3/19/26 - Google announces UCP expansion with: Multi-item cart, improved discovery endpoints, loyalty integration and

• 3/4/26-3/22/26 - Headlines about Instant checkout and the death of Agentic Commerce flood the zone as the Naysayers pile on and dance on the grave of Agentic Commerce.

My three favorite examples of the ChatGPT Checkout/Agentic commerce 🪦🕺

Why Walmart is Moving from Instant Checkout to App Checkout?

The Tech Buzz article here and a Wired article out here, what we have presented from these two reports is a tale of Walmart not being happy with Instant checkout, surfacing problems and a ‘solution’ that raises more questions than it answers.

Instant Checkout Problems Raised (by a named source I’ll point out)→

• Conversions - "Conversion rates—the percentage of users following through with a purchase of an item shown to them by ChatGPT—have been three times lower for the selection sold directly inside the chatbot than those that require clicking out, according to Daniel Danker, who oversees design and product for Walmart."

• Consistency - We learned that our customers want consistency across every touchpoint," a Walmart spokesperson said in a statement.

• Flop - According to Daniel Danker, “Put simply, Instant Checkout has been a flop.” Not much actionable over here, Mr. Danker is clearly not a happy camper, that much is clear.

Three other common Instant Checkout Issues I’ll throw in the mix here are:

• Data errors- Frequently you’ll click on a Walmart widget A instant checkout and get widget B. This is Walmart sending a janky datafeed, that ones on them. (sorry it’s obvious).

• Merchant wait times - Most merchants I talk to are very upset they filled out the ChatGPT form and never got an email/call or anything.

• Crawler errors - Frequently in the discovery process, there will be mapping issues, pricing errors and incorrect shipping information. We know how to solve all this - $400B goes through Amazon’s marketplace alone and it’s the same exact type of system. These are known systems, we can build them. We, as an industry, HAVE built them. We will build more (e.g. TikTok Shops).

Reported Solution:

• Next week, Walmart’s chatbot, Sparky, will begin operating within ChatGPT—essentially a chatbot inside a chatbot. A similar setup will arrive in Google’s Gemini chatbot next month."

This is a perfect example of information asymmetry - this makes zero sense to me - I look forward to seeing how the consumer going ChatGPT→Sparky→Walmart makes any sense and improves conversions. 🤔

Scot’s, aka Captain Obvious, Solution:

I’m going to just say this, it’s so insanely obvious but nobody is saying it, so I will.

The way to solve all these problems is these features:

• Loyalty program integration, multi-cart, and more pdp details - Solves conversion and consistency. If I’m a Walmart+ subscriber (I am) and I can’t use that benefit on Instant Checkout, I’m going to abandon that cart. Everybody knows the economics of 1 item/1 cart don’t make any sense also. More items in a cart, more buy-in, more conversions. 🤦‍♂️

• Self service merchant center - Once you have a self-service merchant center, the lines and forms go away AND the data quality goes up. Add a refined discovery endpoint (like UCP just did!) and bang, problem solved. The next step, right out of Amazon’s and even eBay’s 3PM playbook is you then create merchant-level incentives via strikes/scorecards in the self-service merchant center to make sure they enforce canonicalization (👀 at you Walmart datafeed) , data freshness and other key pieces to the CX. I lived 20+ years of life worried about SNAD, INAD strikes, ODR rates and the like at ChannelAdvisor.

Dang, if only ChatGPT had these features 🧐

The only piece we didn’t solve in the marketplace era was loyalty programs and merchant-of-record. Instant Checkout solved MOR first, and it brings with it the unlock to loyalty. You’re one inch from loyalty when you solve MOR which IMHO is the harder problem.

I will say this as clearly as possible. These are all solved problems (Google UCP now has gone that last inch, so I’m counting it as ‘solved’ now too btw).

Ironically, Walmart knows exactly how to solve these problems (except MOR/loyalty), because Walmart solves them today right there it’s own third-party marketplace.

Analysis of the Information Piece

From the information piece, what we have at the core of that piece is this quote:

“We’re evolving how we approach commerce in ChatGPT to better meet merchants and users where they are,” an OpenAI spokesperson said in a statement, adding that the company is prioritizing improving product search and discovery in ChatGPT. “Instant Checkout is moving to Apps, where purchases can happen more seamlessly. We appreciate our partners for learning with us and look forward to sharing more as we continue building in this area,” the spokesperson said.

A couple of things to point out:

• The quote is attributed to an un-named “OpenAI spokesperson”

• The mystery person explicitly says: “…we continue building in this area.”

Unfortunately, we don’t know if the reporters at Information asked these critical follow-up questions:

• “What about the multi-cart experiment”

• “What’s going on with the merchant center?”

• "How about the expanded experiences we’re seeing being tested such as more info and follow-up?

• “Any update on loyalty program support?”

• “Is this just a Walmart thing or an everyone thing?”

• “Ok, you are going to focus on discovery, are you going to stop checkout all together?”

The Instant Checkout Dog Not Barking…

Editorial Note: This is an idiom based on a classic Sherlock Holmes story. Details here for my many international readers that I may have lost on this reference. 👍

Swirl all this together and do you see the dog not barking yet? Let’s look at it from 30k-feet:

I. Instantly Checkout 1.0 is obviously just a first step

Literally everybody in the industry knows that Q4_25 Instant Checkout wasn’t complete. In this June 2025 articleutm_source, I introduced the Five levels of agentic commerce and very purposely put ‘loyalty program integration’ as part of ‘Advanced buy’ because it and multi-cart are obvious ‘must haves’ for this to work at all.

II. ChatGPT is clearly building features to address IC 1.0 shortcomings

We have very clear and transparent evidence in Q1 2026 of ChatGPT building out multi-item cart, a self-service merchant center and mini-PDP enhancements for the product and offer cards to improve conversion by giving the shopper much more product information.

III. Google announces UCP enhancements: Loyalty, multi-cart, discovery and self-service merchant center.

The UCP team at Google, as surfaced in my interview last week with Ashish, has clearly seen the need for these things from the three stakeholders of UCP partners, shoppers and the sea of merchants interested in these programs.

I hope this presentation of information has enabled you to see the dog not barking here and helps you start asking some deeper questions about what’s really going on here.

Subscribe now

News Item 2: ChatGPT's Shopping Home Page Take II

Speaking of ChatGPT and discovery, Joe the Agentic Sleuth has an update on the new ChatGPT Shopping Home Page. This one is best seen as a very short < 30s video clip here→

Here you see that since our first reporting way back on March 18th (here if you want to compare before/after) it’s already much cleaner and clicking on the discovery cards actually fires a prompt and shows relevant results now.

I know what you’re thinking - kind of crazy that they are still working on this disaster of a flop, what gives?! 🤪

News Item 3: Meta Decides to Go ACP?

Several folks that watch the ACP Github repositories like crazy have told me of evidence in that repository that META is now an active ACP participant, but so far, no evidence of them in UCP. Several of these folks have basically come to the somewhat logical conclusion from this chain:

• Jan 29 - On their Q4_26 Earnings call Zuck said they are doing Agentic Commerce. Our detailed coverage here. No direction on UCP/ACP mentioned.

• March 2- We discovered that Meta AI is Live with Agentic Commerce results, but no checkout - yet.

• March 15-18 - Reports of ACP activity by Meta go from zero to low roar

At this point if you are getting in the Agentic Commerce game, you have two, or you could call it three options:

• Option 1: Adopt existing protocol:

  • Option 1a. Adopt existing ACP

  • Option 2b. Adopt existing UCP

• Option 2: Build your own new protocol

It seems kind of obvious to me given they are so strong on the discovery side with Instagram and all the surfaces, they would go Option 1. At the same time, Anthropic, being anthropic I believe will go Option 2 and when they do it will blow our minds. They have an uncanny ability to jump 18 months into the future. I suspect time travel is involved and Dario is from 2040, I could be wrong that one, this is not an official Retailgentic prediction.

Conclusion: I think we’re at 75.4% probability we get a ACP announcement out of Meta AI in the not too distant future. I’ll point out the math on their next earnings call would put it in late July. I think that’s the ceiling on something - that’s like 5yrs in AI time.

News Item 4: McKinsey Reveals Feb 2026 Consumer Agentic Commerce Preferences

We haven’t had a great, reputable n=2000+ consumer since November of 2025 and the long wait is over. McKinsey has a great research paper here and they are at n=4008 - those last 8 - 👍

There’s a TON of great info in here and I recommend you go check out the whole thing, the two pieces of data that caught my attention as newsworthy and worth discussing at cocktail parties, board rooms and Shoptalk are:

Datapoint 1: What Shopping Activities are AI Consumers Engaging In?

I’ll note while trying not to say any more snarky comments about the first topic that 23% are now using it to checkout and buy. Wow. I was told it’s a mirage AND a hallucination, I guess it’s heading towards the ‘Mass Delusion’ phase. (Ok, to my credit that was snarky about naysayers generally, I’m warming up for Tuesday at noon).

Datapoint 2: What Categories Are AI Shoppers Shopping?

This is interesting, the last data from November had a bit different sequence with grocery very high (31% in McKinsey, and 49% in MS AlphaWise below). Note that McKinsey has Electronics at 38% - by far the highest i’ve seen that and AlphaWise is at 25%.

News Item 5: Is Amazon Working on an Alexa+ Phone?

I’m pretty sure that Jason RetailGeek and I are the only two people holding on to our vintage Amazon Fire phones (the first, and last 3D phone!) The Fire phone came out in June of 2014 and a short 14 months in August of 2015, it was cancelled.

Imagine everyone’s surprise after that famous debacle to read the exclusive reporting by my OG buddy Greg Bensinger, currently at Reuters who in this piece, wrote:

• Amazon’s new phone project is codenamed ‘Transformer’

• Focus on AI integration, Alexa features, and mobile personalization

• Led by Amazon’s ZeroOne group, inspired by minimalist Light Phone

Four people are sourced, but not named since they are not approved to speak publicly about it.

Friend of Retailgentic, Wall St. analyst Colin Sebastian had three interesting points in his analysis→

What do you think - can Amazon nail an AI device and create the ‘next platform” or will the mystery device be a repeat of the Fire phone?

News item 6: LogicBroker Releases Industry Data.

This one is a bonus because it’s for the fellow industry people that read this that are building agents and companies in the same ecosystem as what we’re building at Retailgentic. I live in this weird world, where on LinkedIn, X and the press, I’m seeing constant messaging about how Agentic Commerce is a dead end, will never work - you now the drill. Then in the day job/reality-land, we’re bombarded by interest from brands and retailers across the globe having a ‘hair on fire moment’ trying to figure out what’s going on.

Then you see n=4008 consumer survey data that 23% of people are using Agentic Commerce to checkout. Make it all make sense?!

This is a survey that is n=600 which for consumer is not enough, but this is 600 executives at Brands and Retailers, which is a significantly smaller TAM than ‘consumers’, so it’s actually a large sample set from what I’ve seen historically.

Founded back in 2010, LogicBroker started as a dropship network and has added oms/order routing type capabilities and is now touting AI capabilities. I mention this because it would be unusual for them not to ask a lot of questions about this aspect of AI and merchants.

Three charts to call out:

Chart 1: This Report Covers Retailers of all Sizes:

I always check these industry surveys for merchant size - there are millions of online merchants, most are < $1m/yr - this report has a good mix of large and medium, and not a lot of the smaller merchants.

Chart 2: Merchant Leadership Belief and Investment in Agentic Commerce

These three stats jump off the page:

• 91% expect AI to influence at least 20% by 2027.

• 57% are implementing agents

• 47% are investing over $1m in 2026

If you are, like me, in the enablement biz, this helps you prove to the team and potential investors out there that this not a ‘nice to have’, it’s real, with real budgets and real projects that are getting funded.

If you are a leader in retail/brands, you may feel like you’re on the cutting edge and alone, but you are not, we are at a tipping point of 50% and in another 6 months I suspect we’ll be into the 70% zone. This pairs nicely with the McKinsey report showing strong consumer adoption, it’s logical that smart retailers and brands will follow their data and come to similar conclusions.

Chart 3: What Areas of Investment Are Leaders Focused On?

You can see the top two are:

• AI Powered Product Discovery - They don’t specify if this is off-site or on-site, I believe in my conversations with retailers, they are pausing replatforming and a lot of new on-site work to invest in the new world of off-site discovery. This leads at 50%

• AI Chatbots for service- These agents are in the customer service function and helping drive a lot of economic value for retailers and faster/better answers to shoppers. A close second at 48.5%, but Discovery edged otu the win (Yeah discovery!)

Those are a couple of tidbits, the whole report is really excellent - thanks to Logicbroker for doing this.

The full report is here and is sourced as: “Logicbroker/Midsail Research “The State of Agentic Commerce Adoption.” 2026: 2026 Market Research Report.

The State Of Agentic Commerce Report

8.92MB ∙ PDF file

Download

Download

That’s a Wrap!

Unless there’s really big news, that’s it for this upcoming week, we’ll be back in full Retailgentic force next week and hope to get some Shoptalk highlights out while they’re fresh!

Google - Welcome to Retailgentic!

Buckle up Retailgentic’ers, we have a big one for you today!

In this episode, we sit down with Ashish Gupta, Vice President and General Manager for Merchant Shopping at Google and Engineering Fellow at Google, where he shapes technical strategy across both ads and commerce. Over two decades at Google, he has worked on some of the company’s most complex large-scale systems, from data infrastructure and ad systems to YouTube Ads and now Agentic Commerce via UCP.

That mix of deep technical leadership and operational responsibility makes Ashish uniquely positioned to explain how Google is approaching the next phase of commerce: one where AI agents, merchants, and shoppers need a shared infrastructure layer to work together.

Plus, he’s a really fun person to get to know!

Listen/Watch Our Interview:

This episode explores the shift from traditional search and ecommerce flows into a new agentic commerce environment where shoppers are using AI surfaces to discover, compare, and increasingly delegate parts of the buying journey.

Ashish explains what Google was seeing internally that led to the development of UCP, why the protocol was designed to be open and ecosystem-wide, and why the ambition goes far beyond checkout. He also shares how Google is thinking about discovery, conversational product data, merchant readiness, and the long-term role of AI across the shopping journey.

For retailers, brands, marketplaces, and commerce tech teams, this conversation is a helpful window into what Google believes is changing now, and what foundations businesses should be investing in before the market fully shifts.

Highlights

• How Google’s large-scale ad and data systems shaped his approach to commerce

• Why UCP emerged now, and the two major trends that inspired it

• Why Google believes agentic commerce needs an open, ecosystem-wide standard

• Why the future of UCP extends far beyond checkout

• How discovery is starting to take shape within the protocol

• Why richer, more conversational product data matters in AI shopping experiences

• What retailers need to do now to improve visibility on AI surfaces

• How Google is thinking about the relationship between agentic commerce and advertising

• Why agentic commerce is still early, but already moving from concept to reality

Agentic commerce is still early, but this conversation makes one thing clear: the infrastructure decisions being made now will shape who gets discovered, considered, and chosen in the next era of shopping.

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

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

Check out the materials we discuss:

👉 Google’s AI shopping gets simpler with Universal Commerce Protocol updates: https://blog.google/products-and-platforms/products/shopping/ucp-updates

👉 Retailgentic’s announcement:

👉 Retailgentic’s UCP announcement: https://www.retailgentic.com/p/flash-googlenrf-announces-universal

Loving this Convo? We’re just getting started.

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!

Happy UCP/Google Day…

Welcome to Google/UCP day here at Retailgentic! 😄 First, this morning we are out with breaking news on UCP/Agentic Commerce and as fate would have it, we also have an incredible podcast dropping later today with Google’s VP/GM of Merchant Shopping (aka Captain UCP@Google), Ashish Gupta. Stay tuned for that, it’s a banger and a great compliment to today’s news and a perfect warm-up for Shoptalk.

To tee the pod up propertly, let’s jump deep into the news that Google just broke.

Today’s Big UCP News

On a blog post, here, titled: “AI shopping gets simpler with Universal Commerce Protocol updates.” Ashish announces four, count them four, major updates to UCP and the Google program(s) around it.

The new features are:

• Multi-item cart capability

• Enhanced catalog endpoint

• Loyalty program integration (via a capability called ‘Identity Linking’)

• The launch of a self-service merchant center

Let’s dig into each of these in detail.

Editorial note: Deep Agentic Commerce Nerds like, ahem, your's truly are watching the UCP github like hawks and have seen these pop up so we had a bit of an idea something was up, that being said, seeing some draft things on a github is one thing, getting a message from ‘Captain UCP@ Google’ is another. We also have more context on Google’s plans around the capabilities thanks to the blog post.

UCP Multi-item Cart

As a refresher, we’ve talked a TON about multi-item/multi-merchant cart and it’s importance for Agentic Commerce to really make any economic sense for the merchant and add value for the consumer to unlock all the ‘cart goodies’ that come along with this capability (lower shipping, free shipping unlocks, etc.)

Also as a reminder, we found evidence that ChatGPT has a multi-cart capability, we’re not sure what they are going to do with that now that they are focusing more on discovery. Also note in this post we went deep into the matrix of multi-item vs multi-merchant carts and aren’t going to repeat all that here, but I do want to re-post the quadrant diagram so it’s fresh in your mind:

Google/UCP’s Cart News

Ashish starts off with:

As a reminder, UCP is a two sided coin with a front-end and a back-end. By having something in the protocol, in no ways guarantees that Google or other UCP builders will utilize that capability.

On the UCP side you can go here and learn all about it. As demonstrated live in last week’s podcast with Vendee Labs , there was nothing in UCP from stopping someone from building a multi-item/multi-merchant cart on top of the basic protocol, but this beefs up the protocol and gives it a cart back end. This means less work on the front-end and also a standard way that agent builders and merchants can build, process, and checkout carts vs. a wild-west of different scenarios.

Unfortunately, Ashish did not mention if/when we’ll see multi-item carts on the Google user interface (AI Mode is usually where these debut) side, but I’ve got to guess this one is coming soon. We’ll keep our eagle eyes 👀 on the lookout.

Enhanced Catalog Endpoint News

Back to Ashish:

Because e-commerce is so dynamic - prices and stock levels changing (at the variant level where it gets super complicated), this is a critical feature to improve the user experience. Now agent builders using UCP can have a more interactive user-experiences. For example, let’s say a shopper is looking at adidas women’s shoes in size 6 and the SKU she looking out is now sold out in that size, the agent builder can call this endpoint and say - “quick, I need all the in-stock size 6’s women’s running shoes” and now you can say: “I’m sorry Sarah, here’s some alternatives and I have verified they are in stock.”

This level of functionality is starting to get waaaaay beyond what we saw in the marketplace era of ecommerce and I’m 100% here for it - this is how Agentic Commerce creates significantly better experiences - with deep back-end integration.

It’s going to be interesting to see how , on the front-end UX, Google utilizes these endpoints -we’ll be on the lookout for examples.

New Loyalty Program Integration (Identity Linking)

Here are Ashish’s comments on this one:

Like Multi-item cart, I believe that loyalty program integration has been a huge friction point not only for the obvious merchant-side, but also on the shopper side. If I’m a loyal Walmart+ subscriber, I’m not going to checkout without that and have down-graded shipping- I paid for up-graded shipping - WTH?! Or if I’m stacking points in my Ulta Beauty Rewards Program to get this killer sampler pack, I don’t want to buy something off Google AI Mode that doesn’t ‘stack my points’ for me. I’m incented to leave and go to Ulta vs. the more streamlined Agentic checkout experience.

I predict mult-cart and loyalty are going to 100X the adoption of Agentic Commerce on both sides of the equation for Holiday ‘26 and then we’ll see what else the Naysayers come up with 🤣.

Saving the Best for Last: Merchant Center!👏👏👏👏

Back to Ashish:

Both ChatGPT and Google appear to have announced and launched ACP/UCP and had more bespoke ‘design partner’ type integrations/launches. This is a great way to build, because you can get to a v2, v3 faster and learn while you’re launching - always the most agile and fast way to do things, especially in the AI Era where you are frequently ‘specing while building’.

The addition of a merchant center (also something we discovered ChatGPT is/was working on here waaay back in Feb!).

For this one, it’s all on the Google side and here’s a snippet of the landing page:

This landing page links to two things that merchants reading this will want to bookmark now:

1. Implementation guide/requirements here.

2. Sign-up page here.

Important Note: Don’t sign-up unless you are ready to 🚀! You’ll notice this very clear message in the form with an also clear warning:

Need help getting ready? Let us know, we are already rocking and rolling here at ReFiBuy HQ.

🔥Let’s GO!!! 🔥

See everyone next week at Shoptalk…

Subscribe now

P.S. Your move Team ChatGPT….

P.P.S. This is also a significant up-grade to the types of things Amazon’s Shop Direct program can do, will they counter? 2026 is the best ecommerce timeline!

Friend of Retailgentic, Joe the Agentic Commerce Sleuth, has done it again! While there’s a lot of noise about ChatGPT’s shopping checkout and other plans that has caused a lot of chaos and confusion, they are 100% doubling down on discovery. Joe has found they are testing a new new landing page for shopping that helps ‘kick start’ the user to see what’s available and how to best use the powerful shopping capabilities inside of ChatGPT.

Subscribe now

👀FIRST LOOK 👀 at ChatGPT’s New Shopping Landing Page

Here you’ll see a new landing page - https://www.chatgpt.com/shopping that has:

• Title: “Try shopping on ChatGPT”

• A standard prompt box at the top with the hint: “Shop for anything”

• Then a series of product tiles, with a hero tile that spans 4 tiles at the top and rotates followed by a matrix of 6 rows by 4 colunns or 24 “categories” of items that are ChatGPT-shoppable.

Presumably when you click on a tile, it will auto-run a pre-canned prompt for you that the user can see and somewhat ‘prime the pump’ for teaching people how to shop using ChatGPT.

Here’s what the rest of the grid looks like→

You can see there’s a little bit of a taxonomy hint in here, so maybe this will ultimately be a browse hierarchy where you could click into home, shoes, fashion and see a grid specific or inspired by that category. You could also imagine there being a ‘top prompts’ or products area that is more dynamically generated vs. this more static presentation.

Nick Turley Interview on BG2

Nick Turley is the head of ChatGPT at OpenAI and he recently (March 15th recently) gave a wide ranging interview the folks at Altimeter on the BG2 podcast→

At 41:30 (here) he talks about where they are going with Shopping. Notice here it’s more of a focus change to discovery than “we’re nuking checkout”.

He then goes on to say:

• There’s a lot to do to make discovery work

• That’s the current focus - nailing discovery

Also a recurring theme in this interview is they are seeing tremendous market ‘pull’ in ChatGPT for a variety of experiences that they then go build against. Nick explicitly says at the end of the segment: “

If I’m ChatGPT, I carve 2026 up like this:

Q1/Q2 - nail discovery

Q3 - beef up Instant Checkout, open up merchant center to fill the virtual shelves

Q4 - 💰

Thanks Joe!

Thanks to Joe for finding this one.

Subscribe now

Welcome to the fifth and final part of a thematic article series sharing what we’ve learning through 2026 Q1 as retailers+brands implement and scale their Agentic Commerce strategies. First a quick recap:

• We started the series covering Keyword Jail and the Golden Age of Storytelling - we have a deficit of product information - content and context in the Agentic Commerce era.

• Then we looked at what I call the Iceberg Problem - We have waaay (100X+) more product stories and content to provide to tell than brand/category level.

• Next in the ‘Good Bad and Ugly of AEO/GEO - AEO/GEO is good for sentiment analysis and understanding how your brand story is landing with AI, but not for Agentic Commerce.

• Finally, in the last post we covered Agentic Commerce Optimization Update - Strategies and Tactics to Optimize ChatGPT/ACP, Google Gemini/UCP, Copilot and Perplexity

In this last article, between the presentations I’ve given at kickoffs, industry events, for brands and retailers and our overall team at ReFiBuy has talked to hundreds of brands and retailers over the last six months on this topic. As a team, we boiled down the top strategic (big picture) and more tactical (faster hits, take a week) questions and then refined them, then refined them further. The result? We have identified the top 6 Strategic and 7 Tactical questions we receive, for a lucky 13 in total.

In this post we’re going to go through them all and answer them for you. This should help you think through your 2026 Agentic Commerce plan and pick your battles for Holiday 2026 which we believe is going to be a huge step-up in Agentic Commerce driven transactions (10%!?).

Enjoying Our Content, Dig Deeper with the ReFiBuy ACO Guide…

Looking for even more detail? Check out our ACO Guide here today!

6 Strategic Agentic Commerce Optimization FAQs:

Strategic Questions: Questions that are bigger picture in nature

1. How do I structure an Agentic Commerce team and what skills should I make sure are on this team?

2. My company is large and bureaucratic, how do I cut through layers of red tape to implement some of the strategies you talk about with Agentic Commerce Optimization?

3. How do I create and frame a business case to pitch to my CEO/CDO/CMO/CFO?

4. What separates the Agentic Commerce leaders from the laggards?

5. What Phases should I consider as I build a Project Plan for implementing Agentic Commerce Optimization?

6. How do I partner with our merchandising and SEO teams to implement a Multi-layer PDP strategy?

Subscribe now

7 Tactical Agentic Commerce Optimization FAQS

Tactical ACO FAQS - Shorter questions with quick answers implementable in under one week

1. How do I write a product title that is Agentic friendly?

2. What are a couple of silver bullet changes I can make that quick wins with low effort?

3. How do Agentic Commerce Engine differences impact optimization strategies. Corollary: Is this something I do once and get all 6 engines or is it more complicated than that.

4. How do we measure success?

5. What are the trade offs of just sitting this out?

6. Do we have to rewrite ALL of our product information?

7. How do I know what <ChatGPT/Gemini/Perplexity/Copilot/META> think about my product vs. my competitors?

How to Leverage This Article…

Unlike our usual mid-length content, this one is a going to be a long one. You’re certainly welcome to read it top to bottom, but it’s also designed to be a bookmarked reference for you. We recommend scanning it to become familiar with the questions covered, then as you execute on your plan in 2026, you’ll have this as a resource to refer back to. After we publish, the links above will turn into links into the article (anchor links) and you’ll be able to jump straight to your topic.

S-Q1: How do I Structure an Agentic Commerce team and What Skills Should I Make Sure are on the Team?

One of the most common questions we get from retailers exploring agentic commerce is: “What kind of team do I need?” We recently walked a major retailer through this, and the answer surprised them — it’s not a technical focus, it’s a marketing focus. The people who thrive in ACO roles are folks who already understand catalog optimization, ‘digital channel management and optimization’, and digital ai shelf positioning.

The most critical capability? Agility. You need someone who can spin up a catalog data feed, test new product attributes, and iterate in days — not someone stuck in a 6-month approval workflow just to change a product description. Think of it as a parallel catalog strategy specifically for answer engines, separate from your main commerce feed.

Medium-term as these platforms mature, you’ll want to specialize. Today covering ChatGPT, Gemini, CoPilot and Perplexity is relatively the same task. But each engine has its own catalog structure, its own matching quirks, and its own optimization “nudges.” By holiday 2026, the retailers winning in this space will have platform-specific specialists, someone managing the emerging paid/ad opportunities, and a partner relationship lead keeping tabs on the rapidly expanding vertical and horizontal agentic commerce ecosystem. Google traffic is already down 40-50% for some fashion and beauty brands as search shifts to answer engines → the team you build now is how you claw that back in the short-term for Holiday 26, for Holiday 27, you’ll need to specialize by engine.

S-Q2: My company is large and bureaucratic, how do I cut through layers of red tape to implement some of the strategies you talk about with Agentic Commerce Optimization?

Another frequently asked strategic question we receive is how retailers can cut through the red tape in their org to implement agentic commerce optimization. Here’s what we’ve seen work → treat your AI catalog as a separate experiment or sandbox. At most companies, touching the core product catalog and thus the PDPs requires a workflow that runs through six different orgs and can take six months to get a single change approved. That complex change management process is a death sentence in Agentic Commerce. Your competition will literally run circles around you. Plus, the targets are moving way too fast — engines are updating their catalog structures, new commerce features are launching, and your priority products are shifting week to week. If you’re locked into a traditional change management flow, you’re going to get over-run on four sides.

Our advice is to carve out a separate process specifically for AI answer engines, make it clear to leadership that this doesn’t touch the site or the consumer-facing catalog, OR SEO and give your team the authority to enrich, optimize, and iterate without waiting in line. Once you prove things out in the experimental agentic sandbox, you can pull the best stuff back into your PDPs - either human-readable or machine-readable (SEO) and broader catalog — but you need the ability to optimize for AI in isolation to get there.

S-Q3: How do I create and frame a business case to pitch to my CEO/CDO/CMO/CFO?

There are three ideas we have for helping you frame the business case for creating a team and investing Agentic Commerce Optimization.

Revenue Recovery Framing

One way to think about the business logic (this concept is borrrowed from Avinash Kaushik, ex-Google Analytics guru turned newsletter writer who we covered in detail here.) is what we call the “recovery model.” Here’s the reality → traditional Google traffic is declining 20-50% for a lot of retail categories as search shifts to answer engines. But consumers didn’t stop shopping. Those intent signals didn’t disappear, they moved up to the agentic layer.

Retail has a new front door. Are your products behind that door?

The question your finance team can help figure out is:

We’ve seen brands build models where they believe they can claw back up to 100% of that lost traffic. That’s a compelling number to put in front of a CFO. And it reframes the conversation from “here’s a speculative new channel” to “here’s how we recover revenue we’re already losing.” Layer on top of that the fact that ads are coming to these engines by holiday — and you need talent on your team who understands ROAS from the old CPC world is going to need to translate that into the agentic world and make the case for more budget as the signal strengthens.

The New Opportunity Framing

Deloitte actually put out a solid one-pager that works as a CFO cheat sheet for sizing this opportunity — worth grabbing if you need to arm your finance team with a framework they’ll trust from a super reputable source. Here’s the P&L bridge→

The entire papers is here→

Agentic Commerce Redefining Retail Economics

5.59MB ∙ PDF file

Download

Download

And your favorite podcast, Retailgentic, had the authors Saurabh Vijayvergia and Brian McCarthy on the 2/12/26 episode.

Video here→

The Audio version is here→ (all of your favorite podcast players are supported).

https://retailgentic.transistor.fm/episodes/saurabh-vijayvergia-brian-mccarthy-from-deloitte-on-trust-catalogs-and-the-real-agentic-commerce-stack

Influenced vs. Direct Framing

Today, most of the value from Agentic Commerce is in driving traffic back to your site — direct checkout through answer engines is still ramping up. That can make it harder to justify headcount internally when finance wants to see a direct revenue number. Our advice is to run both tracks in parallel: optimize for visibility and click-through now, which is your influenced revenue play, and position yourself early for checkout when it becomes material. The retailers who wait for direct revenue to materialize before investing are going to be way behind by holiday 2026, which is when we think this really starts to ring the register.

None of these three approaches is mutually exclusive and most recently we had a customer blend all three to create a ‘triangulation’ on the opportunity which made it extremely bullet-proof and defensible.

S-Q4: What Separates the Agentic Commerce Leaders from the Laggards?

Let’s start by looking at A couple of Agentic Commerce Leaders that are very public in their support and worth watching.

Category-level Agentic Shopping Adoption.

Due to demographics and Google’s category-by-category decline in clicks plus the Agentic Commerce engines categorical focus, we see the following categories that were highlighted in the Morgan Stanley November 25 AlphaWise survey reveal the leaders and laggards in Agentic Commerce:

Grocery is the exception because it was dead-last on this list from Q1-Q3 2025 and then popped to the top - more on that in a future post. Next you have personal care, fashion and beauty followed closely by Pet.

Speaking of the beauty category, our first leader is deep in the category:

Ekta Chopra - CDO at e.l.f Beauty

In fact at ShopTalk Spring on Tuesday March 24th, Ekta and I are on the “Pro Agentic Commerce” side of a debate vs. Andrew Lipsman and Sarah Marzano representing the “AI Will Not Transform Retail” side (aka the Naysayers) - come watch us duel it out!

Also, Ekta publishes a regular substack I highly recommend called AI Chef. In there she helps answer a lot of these strategic questions.

How to talk to the board about Agentic Commerce and more. You can subscribe here→

Steve Madden - Josh Krepon and Colleen Waters

Steve Madden has been one of the earliest and most aggressive adopters of Agentic Commerce Optimization. Josh and Colleen frequently speak at industry events and Colleen will be speaking at Shoptalk as well.

Diageo - Roger Dunn - Global Retail Media Lead

Roger frequently posts his throughs on Retail Media and also Agentic Commerce on his LinkedIn feed and he’s definitely a thought leader at a top ‘house of brands’ in Agentic Commerce.

Ulta Beauty

Finally, we’re back to….beauty. Ulta made ripples in Agentic Commerce in Holiday 25 when their GEO mentions rocketed ahead of Sephora and Amazon (self inflicted damage there).

And just last week, Ulta announced they are launching a TikTok Shop. This shows a readiness to really lean in on new opportunities and take calculated risks that can result in outsized results.

Agentic Commerce Leadership Commonality

What separates the Leaders in the Agentic Commerce from the Laggards are a couple of factors. Leaders have top-down commitment. They are supported by the board, the CEO and the leadership team. We’re seeing dedicated cross-functional teams of 10-15 people, active Slack channels, weekly/daily stand-ups, and top-initiative prioritization across the org.

Leaders are also getting their products visible today, leaning into how build a much rich, contextual product-level story will ultimately have a leg up from a visibility and revenue perspective.

In the early days of marketplaces, mobile, TikTok Shops, retail media networks and other big disruptive new digital retail technologies leaders, leaned in, and got ahead of the game while laggards took a ‘wait and see’ approach.

S-Q5: What Phases should I consider as I build a Project Plan for implementing Agentic Commerce Optimization?

We get asked all the time: “Okay, we’re bought in on ACO — but how do we actually roll this out?” We’ve landed on a three-phase approach that balances speed with organizational readiness. Let’s dig in.

Phase 1: Learn (Weeks 1-6) — Get Smart Before You Get Big

This is where most retailers underestimate the work. Before you start pushing optimized feeds to answer engines, you need to build internal muscle. Start with a subset of your catalog — your top 20-50 hero products — and use them to train your team on how agentic commerce actually works. Set up monitoring so you can see where you’re showing up (and where you’re not) across ChatGPT, Gemini, Perplexity, and Copilot. In parallel, get your brand guidelines and content policies documented — these engines are going to surface your products in ways your brand team has never seen before, and you need guardrails before you start pushing upgraded product catalog data out. Don’t skip this phase 1. We’ve seen retailers try to jump straight to optimization and end up burning cycles fixing things and pulling the org along.

Phase 2: Integrate (Weeks 4-12, overlapping with Phase 1) — Connect the Pipes

This is the plumbing phase. You’re determining what integrations need to be established — product data flowing in from your PIM, eCommerce platform, or spreadsheets, and agentically optimized data flowing back out to answer engines. The workflow is straightforward: import your catalog daily, run it through agentic optimization(content generation, attribute expansion, product evaluation and scoring), then distribute it to the target answer engines and then back into your internal systems. You need your IT stakeholders aligned on the integration path, but the enrichment work itself should live in a separate environment where your ACO team can move fast without getting stuck in a six-month approval queue. The biggest risks in this phase are approval process delays and internal alignment across teams — so get your technical contact and business champions identified early.

Phase 3: Optimize (Ongoing from ~Week 10) — Now You’re Running the Business

This is where it gets fun. Your full catalog is now being optimized — both by your team reviewing and approving agentic commerce product-level context and content additions, and increasingly by taking what you have learned from performance data. You’re doing context engineering at the product level: expanding attributes, building out Q&A content, and sending platform-specific “nudges” to improve canonicalization and visibility on each engine. You’re also standing up the ongoing operational cadence — someone looking at rankings daily, reviewing what’s working, and iterating. Over time, this is where you start pulling the best SKU-level attribute additions and contextual clues back into your PDPs and broader catalog. And this is where the team structure conversation kicks in: you’ll want your marketer running day-to-day optimization, a product/project management liaison interfacing with the broader technical org, and eventually platform-specific specialists as the channel scales. By holiday, if you’ve executed well through these phases, you should be seeing significant revenue signal from agentic commerce.

Don’t try to boil the ocean on day one. Learn, integrate, optimize — in that order. The retailers who are winning in agentic commerce right now all followed some version of this playbook. The ones who are still stuck in committee trying to figure out whether to start are going to get walloped by the competition in Holiday 2026.

S-Q6: How do I Partner with our Merchandising and SEO Teams to Implement a Multi-layer PDP strategy?

Two parts to this one:

1. What is a Multi-layer PDP strategy?

Here’s a framework we’ve been developing with large retail partners → the future PDP isn’t one page. It’s three layers:

→ Layer 1: Human-Readable — The beautiful, shoppable experience your customers see today

→ Layer 2: Legacy SEO — The metadata, schema, and structured data that traditional search engines still need

→ Layer 3: LLM-Optimized — Rich context, expanded metadata, Q&A content, and deep attribute depth designed specifically for AI agent consumption

Most retailers are only running Layer 1 and maybe a thin Layer 2. Layer 3 is where the competitive advantage lives right now.

2. How do we Partner to Implement Multi-Layer PDP?

Bring in those teams early. Conduct a leadership learning session to gain buy-in from those folks who may be wary of AI and how the platform will give recommendations. Collaborate and create alignment with those teams on how they want to structure the approval process. Be sure to ensure their continued involvement and support by adding guardrails, giving examples for teams to review, and ensuring the right folks are involved earlier on rather than later makes the difference.

This ties into S-Q5 - bring them into the learning phase, make them a part of the process, not collateral damage.

T-Q1: How do I craft a product title that is Agentic friendly?

When creating titles ensure they have enough context in the title to be understood by an LLM. Avoid vague titles like “Riverbank” - it could be a piece of furniture or shoes. If you have a multi-layer PDP strategy and you have room, include some hints/clues, such as a model number to help the destination LLM map your catalog item to the destination’s catalog.

T-Q2: What are a couple of silver bullet changes I can make that deliver quick wins with low effort?

Three of the most common silver bullet-esque actions you can take to deliver quick wins:

• Make sure you are not blocking Answer engine bots and their corresponding agent-bots from crawling your site.

• Provide your product reviews in a datafeed, or make sure you aren’t using javascript for crawlers so they can ingest all your reviews

• Create a basic Q+A based on the human knowledge you have in your org about your top selling products.

T-Q3: How do Agentic Commerce Engine differences impact optimization strategies. Corollary: Is this something I do once and get all 6 engines or is it more complicated than that?

One of the biggest “aha” moments we’re having with retailers/brands is this → unlike the ‘legacy feed era’, with agentic commerce you can’t optimize once and distribute everywhere. Each AI shopping engine has a fundamentally different catalog structure, different inputs in their “offer card” sort order and, soon, will have very different ad platforms all requiring their own bespoke optimization focus.

→ ChatGPT is building a new commerce catalog from scratch, which means matching issues require different nudges to get your products surfaced correctly

→ Google’s AI shopping/UCP is working with a catalog (Google Shopping) polluted by years of legacy shopping feed data that was very ‘dirty’ - lots of bad data and noise in there — cleaning that up is its own massive project. On top of that, UCP requires supplemental feeds with expanded attributes beyond what you’re sending today

→ Perplexity/PayPal - As best we can tell, the underlying catalog here is from the Honey browser extension. In certain categories it’s decent and in others it needs some work. Since the big changes at PayPal, it’s retailers are finding it hard to get information about the program from PayPal/Perplexity.

→ Microsoft CoPilot - CoPilot’s catalog is based on Bing Shopping, so it’s very similar to Google’s catalog - derived from an old-school comparison shopping engine that wasn’t transactional, so there’s a lot of clean up to do there.

→ META - This one is so new, we’re still digging into understanding more about the character and ‘shape’ of the catalog that we can see from the consumer-view.

The big picture - Merchants need bespoke optimization per platform. Same product, different nudges, different catalog strategies. This is fundamentally different from the “generate a basic catalog for Google Shopping feedspec and you’re done” era.

T-Q4: How do we Measure Success?

The Agentic Shopping Engines are a moving target and we’re learning more every day. For example, we’re finding that in analytics and panel data referral data for LLMs is wildly under-reported. Retailgentic friend, Kiri Masters at Retail Media Breakfast Club, calls this Dark Search and has a great piece on it here.

That’s a long way of saying, if you are measuring traffic/citations, you are chasing a ghost 👻.

In the previous post in this series, "Agentic Commerce Optimization Update” we shared these are the top three goals for ACO:

For readers that read all the way to the end we threw in a preview for exactly THIS QUESTION here→

For our World of Agentic Commerce, we believe:

• Product Card Mapping - Product cards are a hallmark of discovery across the engines, if you aren’t mapped correctly and showing up in the best light in the Agentic Commerce combined catalog, you aren’t in the arena at all. Focus here first.

• Own the Product Card - Once you are mapped correctly, you next need to understand where you’re offer is on the ‘Offer card’ - the list of offers associated with every product card. Think of this multiplied by all of your SKUs as your comprehensive AI digital shelf. Based on how marketplaces have developed, we fundamentally believe the goal here is to be number one on this offer table. You’ll notice that across the five live answer engine Agentic Shopping Discovery mechanisms, the highest offer has many benefits around exposure. As more and more people buy from here, we believe the trend will be that the top spot receives a disproportionate % of traffic and sales. For example, if there are 5 offers, instead of there being a 20% equal distribution, it will go something like:

  • Top offer: 50%

  • Second offer: 30%

  • Third offer: 15%

  • Fourth offer: 4%

  • Fifth offer: 1%

Focusing on these metrics at the Research/Find phase will logically drive transactions either on-agentic or off-agentic at the ‘Buy’ phase.

T-Q5: What are the Trade Offs of Just Sitting this out?

To answer this question, we have to look into our crystal balls. On these digital pages, I’ve covered 7 companies from Wall St. firms to ARK Invest to McKinsey, Accenture, Bain and Deloitte - the consensus of those predictions is that by 2030, 20-30% of ecommerce will be Agentic Commerce with some coming in as low as 15% and others coming in as high as 40%.

I’ve been making predictions publicly and transparently for ~15yrs and gotten decent at it - somewhere in the 60-70% zone and 10% of my misses are timing related. But at the end of the day - your guess is as good as mine. There are naysayers that believe this is all a mirage or hallucination and people will never shop this way. Another reminder: we’ll be debating two of the most staunch and vocal naysayers at ShopTalk- details above.

For most retailers, what is moving them off the dime as the saying goes is the decrease in traffic from Google. That’s a ‘hard push’ causing retailers to wake up that we will face a zero click world soon. If you’re seeing your SEO/SEM traffic decrease, I encourage you to look at question S-Q3 and build a replacement model and project that model forward at the same ‘slope’ - that’s the trade off cost of ‘doing nothing’.

T-Q6: Do we Have to Rewrite ALL of our Product Information?

Your product data typically does not need to be fully re-written:

• Start with phases and look at your top sellers

• Start with expanding basic attributes and adding reviews

• Work in that sandbox until you have reached the % of top sellers in your phased approach

• Then work on bringing the data back into the ‘metadata’ layer and cherry pick attributes you want on your PDPs and work that sub-set through your change management process for on-site, human-visible changes.

T-Q7: How do I know what <ChatGPT/Gemini/Perplexity/Copilot/META> Think about my Product vs. my Competitors?

The good news is it’s right there on the offer page→

In this example, Walmart hs found itself a the bottom of the offer list and PetMeds has found itself at the top.

You can immediately see that PETCO is not mapped correctly (It’s the 20 count when you click through not the 45 count) - this means PETCO’s 20-count SKU is not mapped correctly either.

Anyway, taking that one out, there’s reasons for the sort order if you think about it this way - if you were building the algorithm, what inputs do you have? How would you prioritize them? What have we seen marketplaces do in the last era of ecommerce with a similar decision matrix? One hint - look at what ChatGPT thinks the Walmart shipping fee is 🤔.

What are YOUR FAQs?

That concludes our top 13 FAQs with answers (6 Strategic + 7 Tactical) -we hope this helps you navigate our rapidly changing world and think through the big picture strategic decisions and then get moving with actionable answers to your tactical questions.

Subscribe now

Stay tuned! We’ll keep sharing what we’re learning from the front lines as more retailers move from strategy→planning→execution for ACO.

Dr. Darryl Carlton brings a long career in IT, AI, governance, and large-scale systems design, with experience spanning retail, telecom, and major enterprise transformation projects.

Ross Gardiner brings deep expertise in technology planning, delivery, and complex systems execution across global markets.

Together, they’re building Vendee Labs around a core idea: if AI agents are going to reshape e-commerce, the buying experience itself needs to be rethought from the consumer side.

Listen/Watch Our Interview:

In this episode, we take a deep dive into what buyer-side agentic commerce could actually look like. The conversation explores the origins of the company, why Darryl and Ross focused on reducing friction for buyers, and how that unexpectedly opened up solutions for sellers, payment providers, and enterprise procurement teams as well.

The episode also features a live demo of Vendee Labs’ platform, showing conversational discovery, real-time product retrieval, multi-vendor carting, and a multi-item ACP checkout flow.

Highlights

• Why Vendee Labs is approaching agentic commerce from the buyer’s perspective

• The founding story behind the company and the friction that inspired it

• How ChatGPT and Claude changed the direction of the product

• A live demo of conversational commerce across real merchants

• What a multi-vendor, multi-item ACP checkout could mean for online shopping

• Why autonomous commerce needs legal and governance frameworks, not just AI

• The vision for a Universal Protocol Bridge across ACP, UCP, and future standards

• How strong permissions and audit trails could reduce chargebacks and disputes

• Why poor product data creates broken outcomes in agentic shopping

• Where Vendee Labs sees opportunity across consumer, B2B, procurement, and white-label use cases

Vendee Labs is betting that the future of commerce won’t just be agent-powered—it will be buyer-centered, trust-driven, and built to work across whatever protocols come next.

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

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

Loving this Convo? We’re just getting started.

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!

If you’re a deep ecommerce person with years of Amazon knowledge, you can click here to jump right to the news. Most of our Retailgentic readers are newer to this Amazon World, and I’ve prepared this super-fast backgrounder to orient you a bit before we get into today’s big news.

Brief Backgrounder

I’ve been working with Amazon (as a partner) since before they had a marketplace and for those of us that lived through all of this - where we are today makes sense. But I’ve noticed one of the TOP FAQs when I meet Retailgentic readers/merchants at tradeshows is a lot of questions about the differences between 1P, 3p, Shop Direct and Buy-for-Me. Here’s my attempt to try and add clarity there for everyone.

Amazon’s Four Buckets of Inventory

This table lays out the four types of inventory - 1P, 3P, Shop Direct, and Shop Direct/Buy for Me→

A couple of caveats here for those of you that want to get really Amazon-nerdy on me:

• Shop Direct fulfillment - Yes, technically the merchant could use FBA’s MCF program, I’m gonna guess that the venn diagram of those circles is the null set.

• CS for 3P - The Merchant handles most of these, but if Amazon fulfilled via FBA, they take over shipping questions.

• Permission model / 3P - I’m aware that many categories are gated (frequently counterfeited I think is the lingo).

• The ‘no ads yet’ for Shop Direct models is me speculating that’s how they will look to monetize this eventually, should the program continue out of beta.

The point is Amazon is assembling a very unique set of four different pools of inventory that give them access to merchants of all sizes and relationship types. But the ultimate goal seems to defend against ChatGPT, Gemini and other ‘Agentic Commerce Answer Engines’ ‘Out-selectioning’ Amazon. To understand this strategy, we need to go a bit deeper in the history of how we got here.

Brief History of the Four Buckets

1P/First-Party/”The OG”/VendorCentral

Amazon was founded in July of 1994, by Jeff Bezos and was born with a traditional retail model. What we call the wholesale model, or first-party, commonly abbreviated 1P. In this model:

• The retailer buys products from the manufacturer(brand/merchant) at a wholesale price.

• They then re-sell that product at a (hopefully) markup

• That difference minus the carrying costs, fulfillment and all other costs is the retailer’s profit.

• VendorCentral has (near always) been the portal that wholesalers primarily use to work with Amazon.

3P/Third Party/MarketPlace/SellerCentral

In the very early 2000’s Amazon added used-books through a marketplace-like system. Then after a big blow up with Toys-R-Us, around 2004-2006, Amazon started to scale the marketplace outside the media (Books/movie/videogames) categories.

Then FBA was released in Sept 2006 and also that year they changed the buy-box algorithm so third-parties could famously “own the buy box” and we were really off to the races at that point.

Today - for the full year 2025, Wall St analysts estimate Amazon’s third-party marketplace at $575B in GMV 🤯

Note: GMV: Gross Merchandise Value - the measure of the transactional volume through the system, of which Amazon collects a % (called a take-rate), which is their revenue component of those sales.

Shop Direct

About a year ago, in Feb and then April 2025, Amazon rolled out:

• Shop Direct - The name of a program that allowed for any merchant’s inventory to be shown to Amazon’s 100s of mullions of buyers outside of 1P and 3P for the first time ever and the qualified traffic delivered directly to the PDP. Details here.

• Buy for Me - A subset of Shop Direct inventory that can be purchased on behalf of the shopper by Amazon’s agent. Initially this was restricted to merchants that opted in and took Amazon Pay. My understanding is this program has been expanded to include some Shopify stores as well, but I haven’t found those in the wild yet. Details here.

Subscribe now

Understanding Amazon’s Agentic Commerce Strategy

Last thought before we get to the news. Amazon, being Amazon and given their ‘king of the hill’ position in ecommerce, has a unique strategy for Agentic Commerce that is also one of my TOP FAQs when I meet people, because it can be perplexing if you don’t see their point. I’ve found it’s helpful to compare to Walmart visually to help understand:

Walmart’s Agentic Commerce Strategy

First, here’s Walmart’s strategy, which I’ll call: Agentic Commerce: Open for business

Walmart has thrown open the virtual doors for crawlers, ACP and UCP to come and get their inventory and bring in new customers. Walmart is betting that customers shopping via the Agentic Commerce ‘Answer’ Engines are incrementally new customers, and, maybe even Amazon’s customers. What’s even better, is Walmart’s 3P merchants come along for the ride, giving them even more exposure and opportunities to sell with zero effort.

Amazon’s Agentic Commerce Strategy

On the other hand, Amazon has taken the ‘The Agentic Doors to Amazon are Closed’ approach:

Amazon’s view is their inventory and associated data are the keys to the kingdom that must be protected. But by doing this, by definition, it puts Amazon at a disadvantage to ChatGPT and other engines because they will pull in inventory from any brand or retailer, therefore over time, their inventory selection will EXCEED Amazon’s. How can Amazon expand their selection to be at parity and even ahead of ChatGPT? Enter: Shop Direct.

If Amazon can have all of the Amazon inventory PLUS all the same merchants ChatGPT has (minus Walmart probably), then they can’t be out-selectioned.

Today’s Amazon Shop Direct News

That brings us to Amazon’s news today. They are announcing an expansion of the Shop Direct program to accept feeds. Today they announced are accepting feeds from some of the third-party pre-Agentic legacy feed providers and soon, they will be opening up a self-service merchant portal so merchants can upload their own product data feeds.

Prior to today, Amazon was crawling the ‘target’ Shop Direct merchant’s site to pull in unique inventory. For price, more product details and inventory levels, datafeeds are always more reliable than crawlers.

Amanda Doerr, Amazon’s VP of Core Shopping at Amazon said:

Highlights From My Conversation with Amanda Doerr.

The previous quote is from the Amazon News post. Yesterday, I was able to catchup briefly up with Amanda about the news and picked up a couple more tidbits of interest for Retailgentic readers:

First, I asked what their plans are to monetize this offering and she said:

Ultra-Hybrid Mode?

At ChannelAdvisor, we started to see merchants use BOTH the 1P and 3P - we called this Hybrid mode. Each merchant had their own reasons for what SKUs went in which bucket (see the table at the start), but at the end of the day - it meant maximum selection for Amazon’s buyers.

That led me to ask Amanda: “Do you see a world where customers may have some inventory in three or four of these different buckets?”

Here I’m paraphrasing:

Yes. In fact, I think that this is also a great way to extend the amount of selection that merchants have on Amazon.

We look at this both as net new incremental coverage for brands that we might not have at all today, as well as making sure that we’re rounding out all of that selection.

I do see a world where we will have 1p and 3p selection as well as maybe some particular items they want available to amazon shoppers via shop direct.

I’m going to go ahead and call this Ultra-Hybrid mode which is a really big development for those merchants that want to really embrace the Amazon ecoystem.

Here’s a fictional, but not too far away from reality example:

Under Armour has a set of inventory that Amazon wants to move a lot of - say their OG dry-fit shirt. Some of the other inventory, maybe shoes, or golf, Under Armour chooses to place in 3P. Finally, maybe they have some items that can be customized using a unique ‘only on Underamour.com customization feature, or inventory exclusive, but who doesn’t want free access to hundreds of millions of qualified buyers? They go ahead and send that inventory in a datafeed and now they are Ultra-Hybrid:

• 1P: Super fast bestsellers that Amazon wants to price and move

• 3P: Under Armour wants to sell their whole line of team and individual sports

• Seller Direct: Under Armour wants buyers to see their customizable line

• Seller Direct/Buy for Me: Perhaps there are ‘dot com exclusives’ or “licensed brand items not able-to-be-sold-on-Amazon” that, unlike complex customization, are able to be purchased by Amazon’s agents.

The end result: 100% of Under Armour’s inventory is now exposed to Amazon’s buyers which wasn’t an option before.

What’s Does This Mean for Overall Agentic Commerce?

While ChatGPT, Gemini, etc.. are figuring out V1 of their Agentic shopping experiences, Amazon is shoring up their selection fortress by filling in brand holes that could be a weakness should ChatGPT, Gemini or any of the other players get traction.

Sometimes the best offense is a strong defense and that’s Amazon’s current strategy. Will it slow down the attackers at the gate? Only time will tell!

Holiday 2026 and 2027 are going to be very interesting to see how this all plays out, stay tuned!

Subscribe now

Subscribe now

Welcome to a very special episode of the Retailgentic Podcast! In November of 2025, Peter Steinberger released ClawdBot into the world and here we are three short months later. Since then Peter renamed the project (at Anthropic’s request) to OpenClaw and was hired by OpenAI for an undisclosed amount of $.

In the last thirty days, on a daily basis, Retailgentic fans like YOU 🫵 have been relentlessly asking: “Scot, umm, haven’t you heard of OpenClaw?!” Embrace the CLAW🦞 Scot, come on!

Excuse Me, Do YOU Know About OpenClaw and Agentic Commerce?

Befuddled by the onslaught of inbound inquiries by aggressive Claw-sters🦞, I was on a trip to Silicon Valley last week, and stumbled upon what I thought must be the home of OpenClaw. Here I would find answers!

Alas, it turns out Luke had no idea what I was talking about, but did say I was the fifth person that week to ask about it and he didn’t know what was going on, but would I like to see the menu? 🤷‍♂️

Ryan Eade to the Rescue🦞

Rumor had it that my local tech-friend and super AI enthusiast Ryan Eade was deep into the OpenClaw world, so I scheduled a coffee meeting to see what that was all about and when he showed up singing this song dressed like this, I knew we had found our man!

The Mt. Everest 🌋 of Agentic Commerce Tests for OpenClaw…

After doing my research on X and the OpenClaw forums, I challenged Ryan and his armada of Claws to demonstrate these 3 super-hard Agentic Commerce tasks:

Task 1: Locate a the best price on a product both online and offline. **But an Important Twist: Have OpenClaw find detailed local inventory!

Local inventory at stores like Walmart, Target, BestBuy, Dick’s Sporting Goods are notoriously hard to execute on because an agent has to meticulously, find the local stores within an X mile radius, iterate through them, changing the store locator, executing the search, finding the item and pulling the availability and price. While this is a very parallelizable problem (agent swarm!), I have yet to see a demo of anything even close in browser-scraping agentic commerce (which OpenClaw is).

Task 2: Buy Something From Amazon 😱

As covered in deep detail in these virtual pages, Amazon is very very very very anti-bot. Could Ryan and his Claw defeat the tech magic that Amazon employs to ward off the evil Lobsterbots?

Task 3: BOSS LEVEL: VisionClaw!

I thought, hmmm, what could be harder than shopping on Amazon?! Then after hours of searching X, I found a hacker that had taken a pair of Ray-Ban META AI glasses, hacked his OpenClaw in as a personal voice assistant and then on top of that he created a shopping skill that lets him look at any physical object, perform an image/object search, speak back the results for further refinement, or you can order from that list verbaly.

Claw-powered Augmented Reality Real-time Agentic Commerce!?!

OH MY CLAWD🦞 !!

This would be awesome, but, I was very suspicious that the one demo video I found could be fake AI generated video - it’s hard to tell these days. I’ve seen these types of AR/VR demos and they are super slow, flakey and never as crisp as this demo. It felt too good to be true. Plus META glasses are notoriously closed, how is this possible?

I presented this to Ryan, he asked for a minute and feverishly texted his OpenClaw-powered agentic assistant and they return with a hardy: “Challenge: Accepted!”.

He didn’t know yet, but to make it EVEN harder, I threw a curveball in. Ryan’s Clawd would have to find my go-to hard to find-and-match product - Veggiedent dog treats. Those of you that have followed my first live streams of Agentic Browsers and other crawler-based systems, have seen them break instantly at the complexity of the seemingly simple Veggiedent dog treats. (It’s because they come in variations of dog size AND they have product variations fresh/something/something-else).

Note: They look harmless, but these VEGGIEDENT dog treats strike fear into the hearts of Agentic Commerce bots around the World!

OpenClaw Disclaimer and Skills Used In This Demo

In this podcast, Ryan takes us through all of the above. Can he pull it off? This is where I tell you, you have to watch or listen to find out!

Before we jump into how to watch/listen to this bonkers demo, first a couple of critically important notes:

Disclaimer: OpenClaw and Agentic Commerce Are Inherently Insecure.

Ryan is a security expert and while he implements some clever systems here to minimize the footprint, he did this in a very controlled environment, that wasn’t purely off-the-shelf OpenClaw, but some extra security rails were added. If you decide to experiment with OpenClaw and give it access to any credit card or bank details, you have been warned, this is highly risky.

If you want to learn more about why this is risky and why our friends at Nekuda don’t think it’s the future of Agentic Commerce, I highly recommend their substack on the topic here→

Unlocking Agentic Commerce

Why OpenClaw Won't Buy You Anything Soon

TL;DR: - OpenClaw is the first open-source personal agent that feels mainstream. The obvious next question is payments: can I let my lobster buy things for me? - Not reliably. There are two “human route” ways to do it today, via guest checkouts or logged-in flows, and both are fragile at checkout. - Guest checkout via browser automation (i.e. inject…

Read more

2 months ago · 7 likes

Some OpenClaw Skills Required

For those of you aware of the risks and looking to learn more at your own peril, here are the OpenClaw skills mentioned in this podcast. Disclaimer2 - we have not validated that these are risk free, a common attack vector on OpenClaw is prompt injection via skills, be careful, you are on your own, this is in no ways a valiation of these skills to be risk free, on top of an already risky system.

With that disclaimer, here are the skills:

Skill 1: Rye Universal Checkout

Details here. Rye offered the best solution for solving the Amazon problem.

Skill 2: SERP AI

Details here. This skill dramatically speeds up the execution of the OpenClaw agent in the Research→Find phase of the purchase cycle.

Skill 3: Buy-anything

Details here. After saying/or typing his bill to/ship to 1000 times, Ryan decided to find or build a skill that would persist that and make it easy. buy-anything does that.

Skill 4: VisionClaw

Finally, the skill that makes VisionClaw possible is here. Note: Requires Ray-Ban META glasses.

Where to Watch

Where to Listen

Addendum

Unfortunately due to our recording software, Streamyard, we could not get the Visionclaw demo to stream correctly, but I can 100% verify it worked flawlessly on the veggiedents the first time it had ever seen them. I literally had to pick my jaw up off the floor, there were 2-3 times in this whole experience where I felt like I was on the edge of the reality I knew and a new one that - it’s hard to explain how mind-blowing some of these experiences are.

I asked Ryan to record a stand-along video so you could see the entire experience and here it is:

And here’s the output on his Claw’s (RonClawd) slack channel:

Zuck Follows Through on His Q4 META Conference Call Comments

Way back (in AI time), on January 28th, Mark Zuckerberg said: “Our ads today help businesses find just the right very specific people who are interested in their products. New Agentic shopping tools will allow people to find just the right very specific set of products from the businesses in our catalog.”

At the time I predicted:

I’ve often said the hardest thing about predicting in the AI age is the freakin’ timing. my 3/15 window was off by 2 weeks, but thankfully I’ve learned my lesson and threw in there “possibly earlier” there - phew.

33 Days Later….

Let’s get to it! Only 33 days later from Zuck uttering those words and here’s what we have (details on how to try and replicate this at the end):

We have discovered that on meta.ai there is a new ‘shopping’ ‘verb’ available (in the above you can see it in the top screen shot. Then when you click on that, it puts meta.ai into ‘Shopping research’ mode. You can see from the above it says: “Personalized product discovery”.

Here you can see Meta is offering 4 canned prompts.

Now that Meta AI Agentic Commerce ‘exists’, let’s run this new fledgling Agentic Commerce Engine through its paces!

Thanks Joe

Shout out to our friend Joe who tipped us off to this! Thanks Joe!

Example 1: LED TV

Here the prompt is “I am looking for a new 65” LED TV and here’s the response:

Houston: We have Product Cards (in Meta AI Shopping Research Mode)!

By now frequent readers will see that, not unlike the existing 4 Agentic Commerce offerings, we are presented with product cards! It’s a 3x2 scroll in this example.

Unlike ChatGPT’s current product cards, there’s no ‘mapping’ or (my favorite word) canonicalization going on (yet). When you click on a product card it opens a tab to that retailer, so it’s a 1:1 ratio of retailer to card. E.g. if you click on the the first one for an LG, they have hard pre-selected ‘brandsmartusa.com as the winner of the card and don’t show you the options for that product.

The other really interesting tidbit from this example is at the top (unfortunately it scrolled off and we couldn’t get it in a shot with the results) it shows that the model is a thinking model that took 15s to pull together this information, if you click on that to have the model ‘show it’s work’ you get this→

Note that the little shopping back icon and it says: “Searched the product catalog” and also look at the next bullet: “Considering categorization strategies”.

To me that shows that the Shopping research function has some helpers - a product catalog helper and a categorization helper. Knowing how this ‘has’ to work, over time, I think there’s a near 100% chance that the categorization helper will canonicalize these offers onto the product card. Since this is so early and Facebook has a hacker mindset(the go fast an break stuff kind not the other kind of hackeer) we are seeing them build in the open and this is super MVP stuff we are seeing.

Example 2: Beauty

Let’s do one more example, this time beauty: This was a multi-turn prompt looking for a gift for my 16yr old niece and after 2 questions (the last answer being she is into makeup) resulted in product cards.

Again, on my favorite topic of canonicalization, you’ll notice that product cards 2 and 3 are duplicates, so the categorization function being very young/unsophisticated at this time, didn’t know to either pick one or collapse them. Likely there’s a small difference in the title or an attribute that fooled it into thinking they are different products, when in fact they are not.

Give Meta.AI Agentic Commerce a Spin!

Here’s the steps we followed to get this to trigger:

1. Head to meta.ai

2. Login - we found it triggers most reliably both logged in and with your instagram account vs old blue fb.

3. At that point, once logged in, you should see the shop ‘verb’, hit that

4. Congratulations you can now experiment on your own.

Let us know if you find anything interesting,

Coming up Later This Week: 🦞

Buckle up everyone, we’re going to embrace the CLAW!

In typical AI-era fashion, an X post was heard around the world earlier today from non-other than SamA→

They then hopped over to CNBC to talk to ARS→

Big Picture Terms of the Deal

Here’s the bones of this investment deal by →

• $50B - We have to start there, this is a big chunk of change, even for the $2.2T behemoth. As of their Q4 25 report in Feb, Amazon has $123B in cash, cash equivalents and marketable securities, therefore, this is effectively half their cash! Perhaps some of this was trade dollars and not cash, but its still a big bite.

  • They did mention there are tranches -the first is $15B so it’s not all up front. Future tranches are based on ‘future conditions being met’ 🤔

• Valuation - This round was done at a $730B valuation which gives us a post money valuation of $840B! 🤯

• AWS - OpenAI will run more of it’s workloads on AWS hardware, including a commit to 2 gigwatts worth of capacity on Amazon’s AI chip, Trainium.

  • OpenAI will parter to bring OpenAI to bedrock

  • AWS will be the “exclusive third-party cloud distribution provider” for OpenAI Frontier, an enterprise platform for building and managing teams of AI agents with shared context, governance, and security. Microsoft says Frontier will continue to be hosted on Azure.

    • Editor’s note - My read here is Amazon can ‘sell’ OpenAI Frontier into enterprises, but they aren’t hosting it, Microsoft keeps that juicy part of the biz because they were such a big founding investor. Frontier is OpenAI’s agentic framework.

  • Stateful runtime environment - powered by OpenAI's GPT models, available in Amazon Bedrock — enabling AI apps to retain memory, identity, tool-calling, and compute across sessions rather than starting from scratch.

    • Editor’s note: context windows on LLMs are. HUGE deal, this is a stateful context window so you don’t have to restart the RLAIF engine.

• Joint Model building - They will collab to build “customized models” to power Amazon’s own customer-facing apps. (Alexa?, Rufus?)

What Andy and Sam Said:

What Does This Mean for Agentic Commerce?!

The TL;DR is that while nothing was announced here on the consumer side, it’s a big step towards a warming between the companies. Amazon is now fiscally tied to OpenAI in several dimensions which should open the door. How big a coupe would it be if ChatGPT had some exclusivity on Amazon inventory? (answer: 🐳)

Stay tuned, I suspect by the end of the Q, or maybe in April timeframe we’ll hear more…

Tasty Tidbits from the CNBC Interview

Sam on interest in an IPO and investor access:

“We are open to going public at the right time. There are advantages to being private. There are clear advantages to being public. In addition to all of the reasons that we get more access to capital, I also think that if these AI companies are as important as we think they may end up being, we want all investors to have access to them. So yes, there is definitely some excitement from various investors of ours to go public, and it is something that we are open to doing, but we need to figure out the right time to do it.”

Sam on demand growth and revenue from the Amazon partnership:

“Yeah, we continue to see just rapidly growing demand. The Codex in particular, I think it grew like 30 something percent in the last week. And that is an indication of what’s happening with enterprise and what people are ready to do now. As we started to look at the growth that is coming, we realized we did not have enough compute to support that. We’re excited, thrilled to get this additional Trainium capacity, and I think it’ll go to great use with customers. We also think, as Andy was saying, that Amazon can deliver so much more to us in terms of new demand and new opportunities in the market, that there will be huge incremental revenue here into the whole ecosystem. And we’re thrilled to get to do that.”

Sam when asked how much runway this little investment gets them→

“I don’t know the exact date, but it’s a very long runway. I expect we will need more capital. I don’t know what the form will be. It may not – you know, there’s, like, a lot of ways we could continue to finance growth, but this is like, a long runway from now.”

Sam on AGI:

“I don’t have – I can’t tell you the exact date that AGI arrives. And also, you know, it’s gotten to the point where people define it in different ways. The field is clearly making rapid progress. So I don’t want to, like, give a timeline, but if you look at what the current models are capable of, you know, you hear people say, man, it feels like something I didn’t expect to see this fast or I didn’t expect to see this in my lifetime. The – I expect continued, steep progress from here. I think we are all going to be surprised about the amount of AI progress the field sees this year. Probably every year after that. So I don’t have a timeline to give you, but it does feel like we are making maybe faster progress than even I expected.”

Andy on AWS AI Capex:

“On AWS capex dynamics and long-term returns: “I’ll start. You know, the way the AWS business works in such a way where the faster we grow, the more capital we put out there. And because the model works, you know, we’re investing in capital in data centers, power, land, chips, hardware, networking gear. So, all this infrastructure 18 to 24 months in advance of when we can monetize it. And then we spend the money in the first year, but a lot of these assets that we’re investing in are 30 to 40 year useful life assets like data center. And so what happens is, you spend the money in year one, but then you actually monetize over a long period of time, and you really like the return on investment in capital over time. You saw that in the early days of AWS, you know, as we are growing like crazy, just on the CPU side, where we were investing a lot of capital and the free cash flow looks different than it does once the growth rates start to level off. And so, as the growth rates are really high, we’ll spend more capital, but we like the operating income and the return on invested income over a long period of time, because those assets we can monetize over a long period of time.”

Andy on AI layoffs/jobs→ (tied it to the block/square giant RIF)→

“Well, I haven’t really digested that news very much, and I think every company is going to make their own decisions on these things. And you know, my view of it is what we’ve talked about in the past, Andrew, which is just, you know, I think that this is the most transformational technology shift that we’ve seen in our lifetime. I think it’s going to impact the way we all do our work. And I do believe that a lot of the jobs that we’ve thrown human beings at the last 20 or 30 years, you won’t need as many human beings doing those same jobs. But I also think there will be other jobs created. And that has always happened in every technology shift. And you know, I mean, 15 years ago, there was no such thing as a cloud solutions architect, and today there are tens of thousands, maybe 100,000 plus, of these types”

Subscribe now

For More Information

You can read more about the news:

• Geekwire

• CNBC video interview with Andrew Ross Sorkin

• CNBC transcript (open)

• Jassy LinkedIn Post

• Official Amazon PR (zzzz)

• The Agentic Ad formats that have been announced by Google, Amazon, ChatGPT and others.

• Examples of marketplace-based Ads that I think inform how many of the

• Instacart ads (Because Fidji)

That file is here for your reference→

Refibuy Agentic Commerce Ads Podcast Presentation

1.07MB ∙ PDF file

Download

Download

Kiri Masters has been at the forefront of retail media since the first Prime Day. An entrepreneur, author, and founder of Retail Media Breakfast Club, Kiri started her journey in retail media back in 2015 when she left banking to launch Bobsled Marketing, an Amazon-focused agency born in the early wave of e‑commerce acceleration. After growing and selling her agency to Acadia in 2022, she’s turned her focus to commentary, analysis, and connecting the dots in a rapidly changing retail landscape.

This is Kiri’s second time on Retailgentic. In her first episode, Kiri joined us to unpack the early rise of retail media networks and what brands were getting wrong as marketplaces evolved from simple distribution channels into powerful advertising platforms. That conversation focused on the mechanics, how Amazon advertising scaled, why retail media margins were so attractive, and what it meant for brands navigating the shift.

Listen/Watch Our Interview:

Retail media has been a profit engine for retailers, especially onsite sponsored product ads with massive margins. But what happens when shopping shifts from browsing and searching to AI agents that “decide the SKU” before a shopper ever hits a retailer site? In this episode, Scot and Kiri map the collision course between retail media networks and agentic commerce, walk through emerging ad formats from Google, Amazon (Rufus), and LLMs, and dig into the hardest problem of all: monetizing AI attention without breaking trust.

What’s Covered:

• Why agentic shopping compresses the journey, and shrinks onsite ad inventory

• The “two threats” to onsite retail media

• How offsite retail media depends on audience signals that may dry up as browsing declines

• Why in-store media may be the most resilient channel

• Google’s “Direct Offers,” the rise of an agentic storefront, and what it unlocks

• The emerging idea of “Agentic PLAs” and retailers bidding at the SKU level (Buy Box vibes)

• Kiri’s wishlist: multimodal ads (video/try-ons), offsite audience extension, and contextual targeting over “creepy” behavioral retargeting
  

Agentic commerce won’t kill advertising, but it will force it to evolve fast. The winners will be the platforms that can monetize attention without sacrificing trust, and the brands/retailers that learn to show up in these new surfaces early.

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

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

Loving this Convo? We’re just getting started.

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 is the fourth part of a thematic series sharing what we’re learning in Q1 as

• We’ve talked about Keyword Jail and the Golden Age of Storytelling - we have a deficit of product information - content and context in the Agentic Commerce era.

• Then we looked at what I call the Iceberg Problem - We have waaay (100X+) more product stories and content to provide to tell than brand/category level.

• Finally in the ‘Good Bad and Ugly of AEO/GEO - AEO/GEO is good for sentiment analysis and understanding how your brand story is landing with AI, but not for Agentic Commerce.

Today, we’ll build on those three articles, plus we build on and revisit our (now vintage) September 3, 2025 4-part series where we introduced Agentic Commerce Optimization which you can find here if this is all new to you.

Since that 9/3/25 series where we coined ACO, it’s been 6 months the Agentic Commerce world has seen a ton of news and new implementations. Remember that was before ACP/UCP/ChatGPT apps, Perplexity/PayPal, etc. In pre-ChatGPT time, it feels like about 5 years have passed!

The overall ACO framework we predicted and shares has held up surprisingly well from that period→

How did you predict this 6 months ago?

We’ve been helping merchants optimize third-party ecommerce marketplaces since 2001 and based on what we learned, there when the very first Product Cards emerged, we saw some patterns from the marketplace world appear.

However, because the underlying engines powering Agentic Commerce are LLMs, there are big strategic differences - so it’s simultaneously similar, yet also wildly different.

Revisiting Agentic Commerce Optimization

In today’s article, we’ll revisit the concept and reasons for Agentic Commerce Optimization, starting at the ‘problem’, understanding the three core challenges and review+update the framework for solving them from the 9/3/25 ACO posts. Then next week, we’ll reveal how real-world teams at brands and retailers are engaging in ACO, how they are reconfiguring their omni-channel and digital retail orgs to accelerate into ACO and we’ll tie together why ACO is totally white hat, measurable, predictable, etc.

Subscribe now

Anatomy of Agentic Commerce

Whether we’re talking about any of the 4 Agentic Commerce solutions, while they have different ‘Research’ mechanisms and extremely different visualizations, at the end of the process when you boil it down, there are three elements:

1. The Prompt

2. The Product Card(s)

3. The Offer Card

In narrative from here is how we talk about this in Agentic Commerce. The shopper prompted the AI Answer Engine (ChatGPT in this example) for a beauty product as a gift. The shopper then picked the ‘Glossier Balm Dotcom’ product card. On the Offer Card there were three merchants: Glossier, Sephora and Kohl’s.

Glossier “Owns the product” card - in other words they are the first offer. This means two things that are very important to ACO:

1. More sales -Historically, on third-party marketplaces the top merchant offer receives a disproportionally large % of transactions (90%+ for owning the buy-box on Amazon for example) for offers on that product.

   

2. ‘Priming’ for purchase - If you look back at the product card (above)- Only one gets a mention at that part of the process, and it’s the one that owns the Product Card. Do you want to be mentioned or ‘others’? This 'Priming’ of the buyer is a factor that the ‘More Sales’ in 1.

Agentic Commerce Goals and How To Measure Them

With that three step buyer flow in mind, your goal, if your organize chooses to lean-in on Agentic Commerce can be simply stated, tracked and measured across three dimensions:

Let’s go through these in a bit of detail including how to measure them.

1. 100% Product Card Coverage - Simply stated if you have X products in your catalog, what % are on the right product card? E.g. If you have 50,000 core SKUs and 10,000 are on the right product card you have a 20% Product Card Coverage score.

2. Continuous Product Card Ownership - Out of your X SKUs, what % do you own the Product Card? To continue the example from above, retailer above:

   1. Out of the 10,000 product cards the retailer owns 3,000 so 30% coverage of Product Cards

   2. However, the SKU Coverage is 3,000/50,000 or 6% SKU coverage - we recommend tracking this as priority, because it highlights how important Product Card Coverage is.

3. Maximize Agentic Commerce Sales - The ‘output’ of steps 1 and 2 is Agentic Commerce Sales. Today, they are 95% of the equation, but based on our experience, we are anticipating many more ‘dials merchants will be able to turn’ to maximize sales (e.g. Cart optimization, loyalty programs, offers, and ads).

ACO is Sequential: Don’t Skip Steps

It’s important to highlight that these goals sequentially build on each other. For example, if you have 50,000 products and only 1,000 are on product cards, then you can’t possibly own more than 1000 Product Cards. This will, also logically, substantially hamper your goal of maximizing agentic commerce sales. This will come up in the next, post, but as you think about deploying ACO, there is a natural phased approach to launch and then an ongoing-optimization that is driven by this ‘sequential build’.

The Challenges Standing Between You and Three Goals

Core First Principle Challenges of Agentic Commerce Optimization:

1. Product Card Mapping (Canonicalization) - How do you get your products on the cards?

2. Offer Card Optimization - How do you show up number 1?

3. Measurability - Product-card and offer level Visibility - How can you measure this to track your progress and see what to do next?

Challenge 1: Product Card Mapping (Technically Coined Canonicalization)

Inside the product card mapping problem there are three main root-causes of the problems:

• The asymmetrical data problem - Imagine a game where we both had a deck of a cards and we are trying to merge them together by suit and face-value. I say to you - this one is red. That narrows it down to 26 cards. Our chances of finding the right one to map to are 3.8%.

• A data matching problem - Now we’re playing the same card game and I say, ok I’ll be really specific now, I have a King of moons, do you have a match? Now you have to decide - do I have some new generation of cards with a fourth suit? This is the first time in you’re life you’ve heard of ‘moons’ - the chances it is ‘real’ are low, so you reject this card.

• Target Catalog drift - I can’t think of a clever card analogy, but target catalogs are constantly improving and being added to/deleted/updated by the catalog owner and the merchants all inserting/mapping their products into the catalog. In the Amazon world today, this causes problems maybe ever 18 months today. In the Agentic Commerce world, this is a near-daily never announced seismic change we all adapt to as fast as possible.

Real-world Product Card Mapping Example - Data Asymmetry

Let’s get out of theoretical card games and look at an example of data asymmetry to help you understand from both the merchant and ‘Target catalog’ example what’s going on behind the scenes.

The visual above illustrates the data asymmetry problem. First a little vocabulary. Whatever system we are mapping to has their own catalog we are mapping OUR (merchant) catalog SKUs to their (target - agentic commerce in our example) catalog.

ChatGPT’s crawler has picked up these attributes from the merchant’s PDP for a shoe and picked up these attributes: Nike, Women’s, Size 8, blue/white, Air Jordan. It will also pick up the title, the price and a lot of other information, and that factors in, but for this simplification, I haven’t listed all that.

ChatGPT then tries to map that against it’s catalog and gets 16 potential matches. Now what’s it supposed to do? Pick the first one? Remember, these engines ignore your images (other than the URL), it’s far too computationally expensive (tokens💵) to do an image match - and even if it did there are probably 4-6 close matches still.

Instead it basically rejects the SKU. The end result - your product isn’t on the product card.🥺

How do you fix the mapping problem?

Agentic Commerce and Marketplaces could spend infinite time improving their catalogs, instead they use incentives (merchants want to sell stuff) to push the work to them. This isn’t going to change. Unfortunately a lot of merchants new to this setup want to go spend their valuable time trying to call a human at the target catalog company and tell them to “fix their janky messed up catalog!” - while that may feel good in the short term, it’s not going to fix the problem and help us achieve the first goal of 100% Product Card Coverage.

What you have to do is, sku-by-sku, figure out what went wrong. Some things to look for:

• What does the target catalog tell you about the card you want to map to?

• Is your product mapped to the wrong card? If yes, why?

• Do you have information on your PDP that’s hidden from the crawler?

To fix these you’re going to take these actions:

• Expand the attributes provided to help solve the data asymmetry problem

• Make sure your data you send the engines as closely as possible matches their conventions (the consumer doesn’t see this, but it can be the ‘nudge’ that helps

• Make sure your variations match the target catalog’s

• and so on.

It can sometimes take 5-10 trial-and-error cycles to get a SKU to ‘map and stick’ to the target catalog.

FAQ: Once I map to ChatGPT does that make mapping to Google/AI Mode, Copilot and Perplexity easier?

Unfortunately, no, each target catalog has very different characteristics, here’s how I’d describe them as of today’s writing:

• ChatGPT - It’s a toddler catalog, it was born about a year ago and still finding it’s sea legs. It’s maturing rapidly and making lots of improvements. Agentic Commerce/Instant Checkout will accelerate that as there’s a strong profit motive on their side to make it more robust because the catalog is the foundation of the consumer buying experience.

• Google - In a way, Google starts ahead of the game. Google’s Agentic Commerce offerings are based off of the Google Shopping graph that has (from their UCP announcement: “50b product listings, that are updated 2b times hourly.” While that’s impressive, Google has never successfully used the graph to SELL STUFF. Also anyone can send products to Google Shopping and it well known to be the most polluted. There are vast canonicalization errors and mis-prices and stock issues. When you are sending traffic it doesn’t matter. When you are sending orders, it matters a LOT. Google has a lot of work on their hands to clean this up, but I’m sure they are building agents to do this so it could be solvable in this era, not before 2022.

• Copilot - Same as Google.

• Perplexity/PayPal - Their catalog seems to be from their Honey browser extension business the acquired. It has it’s own challenges, specifically around variations.

Where to find more information…

In the 9/3/25 ACO series we had a popular post: The 13 Agentic Commerce pitfalls that documented, at that time the 13 most common mapping issues we saw.

At ReFiBuy this is what we do, that list form Sept has expanded dramatically and is now too long and become somewhat too proprietary to share here, but it’s now 6 ‘wide’ by 20-40 deep and well over 200 and growing rapidly as Agentic Commerce is growing rapidly.

Challenge 2: Offer Card Optimization

Now that your products are on the product card, you are in the arena and now it’s time to play the boss level - offer card optimization. Based on tens-of-thousands of datapoints, here some of the factors that determine where your product shows up on the product card. Note: As you can guess the weight of these varies wildly across the 4 platforms.

Offer Card Algorithm Factors

• Price

• Shipping time and cost

• Intent match from catalog details provided and context clues (We’ll dig into this more next time)

• Local availability

• Merchant reviews

• Product reviews that mention the merchant (positively/negatively)

• Do they advertise (Google has a new ad unit that seems to reward paying merchants if they pay - I have a note in for clarification).

• Click through rate (this tends to re-inforce the top position)

That’s the short list today, this is changing very rapidly and depending on the engine, there are different factors.

Challenge 3: Measurability

Finally, for one SKU, you can see how to do this and measure it. For 100-500 you can do it manually, for 1000+ SKUs it becomes untenable.

The Results…

If you can nail all three of these goals, you will achieve Agentic Commerce Optimization Nirvana and more sales and hopefully be the hero of your company:

BONUS: For Advanced Players

If you’ve read this far, you are on your way to be an ‘advanced player’ in this game. Pro tip - after you nail the three main basic metrics and get your catalog humming and your sales up and to the right, you can go deeper on each of the three headline metrics with underlying KPIs like this→

Up Next…

In the next post, we’ll cover:

• Team Structure and skill needs

• How to create and frame the business case

• What separates the Agentic Commerce leaders from the laggards

• Example timeline and phases to start and scale an Agentic Commerce Optimization strategic initiative.

• Product Content Strategy (This ties into the intent signal mentioned above

• Core differences of the engines

• Multi-layer PDP strategy

• Your ACO FAQS - answered!

Happy Agentic Commercing…