Your Product Feed Is Now a Storefront for Agents

AI shopping agents don't read your store the way a shopper does. They read structured product data: a file where each product's price, stock and details sit in labeled fields. OpenAI now asks merchants for that data as a feed, and Google and Shopify have published systems that run on the same kind of data. That file, your product feed, is the first storefront an agent walks into.

I'm Bob Michaels. I started building for the web in 1992, and I now help companies make their sites and stores usable by AI agents.

I wrote this on October 27, 2026. Several of the platform pages I cite describe limited or staged availability, so read every claim here about reach as a snapshot of that date, and check the platform's current page before you plan around it.

What do AI shopping agents read when they look at your store?

AI shopping agents read a product feed, a structured file listing your products that you refresh on a schedule. OpenAI's commerce documentation says merchants share structured product data so ChatGPT can accurately surface their products in search and shopping experiences. The merchant sends a secure, regularly refreshed feed in CSV or JSON, two common formats for tables of data. It carries identifiers, descriptions, pricing, inventory, media and fulfillment options, meaning how the item gets to the buyer. The page itself carries no date.

OpenAI splits the fields into two tiers. Required fields are there so price and availability display correctly. Recommended attributes, such as rich media, reviews and performance signals, are the ones OpenAI says improve ranking, relevance and user trust. That puts your photos and reviews in the feed alongside the price.

Getting in takes two steps. In OpenAI's words: 'Integration involves sending an initial sample feed for validation, and daily snapshots.' The sample is a first version of your feed that OpenAI checks before anything else happens. After that, OpenAI asks for daily snapshots, a fresh copy of the feed each day.

Which platforms besides OpenAI are building for agent shopping?

Google, Shopify and Stripe have each published work on agentic commerce, meaning shopping that an AI agent carries out for a person. On January 11, 2026, Google's developers blog published 'Under the Hood: Universal Commerce Protocol (UCP)'. UCP is a shared set of rules for how agents and stores talk to each other, and Google calls it 'an open-source standard designed to power the next generation of agentic commerce.' Google says it developed UCP in collaboration with industry leaders including Shopify, Etsy, Wayfair, Target, and Walmart, and that over 20 global partners endorse it.

The line in that post I'd show a business owner first: 'With UCP, you own your business logic, and you remain the Merchant of Record.' The Merchant of Record is the business legally making the sale. What follows is my reading, not Google's: if the rules your business runs on stay yours, it pays to write them down where software can read them.

Shopify's Spring '26 Edition developer post, dated June 17, 2026, calls its Catalog API UCP's discovery layer, the part agents use to find products. An API is a channel one program uses to ask another for data. Shopify describes Catalog API as 'turning billions of products from millions of merchants into structured, queryable data across AI surfaces.' It also says: 'Catalog access takes just an API key, no approval needed.' An API key is a code that identifies whoever is asking.

Stripe's post of March 3, 2026 covers payment. It says: 'Any seller already processing payments with Stripe automatically supports these new payment methods for agentic transactions,' and that the capabilities are already rolling out. It says nothing about product feeds, so it changes none of the feed work below.

How far has agentic checkout actually rolled out?

Agentic checkout has rolled out in limited form, and Google's own help page says so. Google's Merchant Center help page on UCP checkout says it only applies to products with eligibility in the United States, Canada, and Australia, and for participating merchants and partners. In its words: 'The checkout feature enabled by UCP is available for select merchants at this time.' I wrote earlier about how checkout became something agents can operate.

The same page shows how a product qualifies. Only listings using the native_commerce(checkout_eligibility) product attribute display the 'Buy' button. A product attribute is one labeled field in your feed. Merchant Center, Google's tool for managing product listings, stays the central hub where you prepare the data.

The platforms also differ from one another in format, in scope and in who can take part. Being eligible on one says nothing about your eligibility on another, so I check each platform's terms separately for every store I work on.

How do I audit a product feed the way I'd audit a page?

To an agent, the feed is the page, so I give it the same audit I'd give any page on a site. This builds on my earlier piece on the second storefront your store just got. The audit has three checks.

The first check is whether each product carries its price, availability, identifiers and policies as plain data. Identifiers are the codes that tell one product apart from another. Policies are your shipping and returns terms. Plain data means a value in a labeled field, typed as text or a number. A product whose returns terms appear only inside a long description, with no field of their own, fails.

The second check is refresh rate: is the feed refreshed as often as each platform expects? For OpenAI, that means daily snapshots. Each platform sets its own expectation, so I check the feed against every platform it's sent to. A feed that falls behind the site hands agents an old price and old stock.

The third check is whether the feed matches the site. The price, availability, identifiers and policies in the feed should be the same ones the product page shows. Where they differ, the agent and the shopper are reading two versions of the same store.

What should you do when a fact lives only in a PDF or an image?

Put that fact on a web page as plain text, then into the matching feed field where the platform has one. A fact that exists only in a PDF or an image is not data an agent can read, so an agent will not state it about you. A fee schedule or a returns policy needs to exist as plain text on a page. I go deeper on this in why a fee schedule in a PDF stays invisible to AI.

Once the feed is clean, take it to each platform by its own route. For Google, the work happens in Merchant Center. For ChatGPT, it starts with the sample feed OpenAI validates. For Shopify's Catalog API, read the developer post before you assume anything about how your products reach it.

I'm a consultant who rebuilds how a company is understood on the web and by AI, through content strategy, web architecture, AI-visibility research and custom agent systems. I'm primarily interested in remote work, and I'm open to part-time, full-time, fractional or consulting work.

Common questions