Your product feed is now your storefront

In March 2026 OpenAI shut down ChatGPT Instant Checkout, the most-hyped agentic-commerce product of the previous year. But it was not for lack of demand. Real-time inventory sync across millions of SKUs, fraud screening and tax handling proved prohibitive for a checkout that lived inside a chatbot with no e-commerce expertise. So OpenAI shifted to routing purchases out to merchant apps instead. Days later, Google began rolling its own native checkout into Gemini, the same week Shopify's Q3 earnings call reported AI-attributed orders up roughly elevenfold and AI-driven traffic up sevenfold across its merchants since January.

OpenAI's rollback shows that the bottleneck on agentic commerce was never the intelligence of the model. It was the quality and structure of retailers' product data. The checkout rail that failed was the one that depended on data nobody had cleaned. The rails that are quietly working all consume the same feed.

So, the urgent question for retailers is not whether to wait for the protocols to settle. It is whether the product data already being read, today, by the surfaces that survived the shake-out, is good enough to win the sale.

Why the feed became the storefront

For twenty years, the job was to rank a page. A retailer optimised a product description, earned some links and hoped to land in Google's top results so a human would click through, look at the photography, read the reviews and buy their product. The website was the storefront. The page did the selling.

An AI agent does not shop that way. When a shopper asks ChatGPT or Gemini for "a quiet dishwasher under £600 that fits a 60cm gap," the assistant does not browse ten retailer websites and admire the merchandising. It reads structured data: the title, the price, the availability, the dimensions, the GTIN, the return policy, and whatever machine-readable signals each retailer has published. Then it compares, filters and recommends. The photography that took a studio day to produce is close to invisible. The 500-word brand story on the about page is noise.

This is the shift retailers keep underpricing. The thing being evaluated is no longer the page a human sees. It is the data a machine reads. A retailer can have the best website in its category and still be passed over, because its feed says "availability: in stock" on a product that sold out on Tuesday, or because the variant the agent needs is buried in an unstructured description rather than declared as a named attribute.

The numbers a board should look at

The stakes are not hypothetical. Adobe Analytics reported AI-referred traffic converting 42% above the channel average. Shopify, on its November 2025 third-quarter earnings call, said AI-attributed orders across its merchants had grown roughly elevenfold in a year, with AI-driven traffic up around sevenfold. The traffic is small today and compounding fast, and it converts better than the average because an agent only arrives at a product page once it has already decided the product fits.

The behaviour is already visible in the market that matters here. McKinsey found that across France, Germany and the UK, 38% of consumers now use AI tools to research products or decide what to buy, and 67% of UK consumers had used AI in the previous three months. Adyen's 2026 retail research puts around 44% of UK shoppers as willing to trust AI to shop on their behalf once preferences are set, and Checkout.com found 40% of Britons would let an agent handle routine shopping such as groceries and household restocks. An Omnisend study of 1,046 UK shoppers, conducted by Cint in July 2025, reported that 55% of British consumers already use generative AI when shopping online, with ChatGPT the preferred tool by some distance.

There is a relevant gap in those figures worth mentioning. Willingness to research with AI runs above two-thirds; willingness to let an agent buy sits at about four in ten. That gap is not a sign the channel is years away. It is a measure of how much trust the current buying experience has failed to earn, and a large part of that failure is product data: agents that return the wrong variant, quote stale availability, or cannot confirm a return policy. The research traffic is already here and already converting above average. The retailers whose feeds let an agent buy with confidence are the ones who will turn the researching two-thirds into the buying four-in-ten, and then past it.

Search rewarded the retailer with the best page. Agents reward the retailer with the best data.

You cannot optimise what you cannot see

Before a retailer changes anything, they need to know how it currently appears to these systems, and that turns out to be the hard part. A retailer can read its own website. It cannot, by default, see what ChatGPT says when a shopper asks for the kind of product it sells, which of its SKUs surface for which queries, on which engines, or whether a competitor is being recommended in its place.

This is what the AI-visibility layer measures. Think of it as analytics for the answers that machines give about a brand: share of voice across ChatGPT, Gemini, Perplexity and Google's AI surfaces, how often the retailer is cited, which products appear for which buying intents, and where rivals are winning the recommendation. It is the diagnostic instrument for a channel that is otherwise invisible, because the queries happen inside someone else's chatbot and never touch the retailer's own logs.

However, retailers must simultaneously recognise that visibility monitoring describes the symptom; it does not cause the cure. A high share-of-voice score is evidence that the underlying product data is good, not a substitute for fixing it. A dashboard that tells a retailer it is absent from 70% of relevant queries has diagnosed a feed problem, not solved one. And the category is filling up with standalone tools that do the measuring and then leave the retailer to find someone else to do the fixing. A retailer that ends up with a visibility subscription, a separate feed agency and no one joining the two, has bought a problem not a solution.

Reading the diagnosis

The value of seeing how a brand appears to AI is entirely in what a retailer does next. A visibility report, read properly, is a feed-remediation brief in disguise. Three patterns come up again and again.

The first is the absent SKU. A product the retailer sells, and sells well to humans, simply does not appear when an agent is asked for that category. Almost always, this traces to a feed problem: a missing GTIN, an attribute the agent needs to qualify the product that the catalogue never declares, a variant that is invisible because it lives in free text rather than structured data.

The second is the wrong-attribute match. The product surfaces, but for the wrong queries, because the data describing it is thin or miscategorised. A 60cm appliance that never declares its width gets recommended to someone with an 80cm gap and ignored by the shopper it would actually have suited.

The third is competitor substitution. A shopper asks for the retailer's exact product type and the agent recommends a rival, because the rival's data answers the question more cleanly. This is the most expensive pattern, because it is a lost sale to a named competitor at the precise moment of intent.

A non-technical executive does not need to fix any of this personally. The job is to read the diagnosis, identify the top-revenue categories where the retailer is losing the recommendation and hand a prioritised brief to whoever owns the catalogue. The point worth holding onto is that seeing and fixing belong together. A retailer that treats them as two separate purchases, from two separate vendors, will spend the gap between them watching the problem it has already paid to diagnose.

Past the feed: the agent on the other side of the table

A clean feed makes a retailer discoverable and selectable. It answers the buying agent's questions accurately and completely, and for the next 6 months that alone will separate the retailers who win agentic traffic from the ones who are quietly skipped. But the feed is passive. It does not argue the retailer's case, it does not handle a substitution when the exact item is out of stock and it does not close.

That is the frontier just past feed hygiene. As buying agents mature from research assistants into genuine purchasing agents, the retailers who win will be the ones with something on their own side of the conversation: a seller-side agent that meets the buyer's agent, represents the brand and actively works the order. It can offer the right substitution rather than losing the sale, surface a delivery promotion at the moment of hesitation, or hold a basket together that a passive catalogue would have let fall apart.

Microsoft Research's Magentic Marketplace, an open-source study of agent-to-agent commerce published in late 2025, found that customer agents accepted the first acceptable proposal they received between 80 and 100% of the time — a "first-proposal bias" the researchers measured across multiple model and protocol configurations. In a world where the buyer's agent tends to take the first credible answer, the first credible seller-side response wins the order. Being present and responsive on the other side of the table is the difference between being shortlisted and being bought.

That capability is still emerging, and most retailers cannot deploy it tomorrow. The feed work, by contrast, pays off immediately. Which is the sensible sequence: get visible, fix the feed so agents read the catalogue correctly today, and be ready to put an agent on the other side of the conversation as the buying agents go mainstream.

Why waiting is expensive

The competitive risk is that this advantage compounds for whoever starts first. Agentic surfaces learn which retailers give clean, complete answers and lean on them. A rival that fixes its feed this quarter starts accumulating that preference while a slower competitor is still deciding whether the channel is real. By the time it is unmistakably real, the gap is a year of learned trust wide.

For two decades, the retailer with the best page won the click. The retailer that wins the agent is the one whose data answers the question before a human ever sees a page, and soon, the one with an agent ready to answer back. The website is no longer the storefront. The feed is.

Clinchr exists to do all three jobs in one place: show how a retailer appears to AI today, fix the feed so agents read the catalogue correctly, and, as buying agents go mainstream, put a seller-side agent on the other side of the conversation to win the order.