ChatGPT Shopping: how it reads product data

A shopper types "ceramic housewarming gift under $100" and gets a row of products with images, prices, and links. Your catalog either appears in that row or it does not, and most merchants have no idea which of their data ChatGPT was reading when it decided.

Unusually for this subject, OpenAI has published a fair amount of it. This page works from those documents rather than from inference.

ChatGPT gets product data from three places, not one

OpenAI's help center lists what ChatGPT considers when choosing which products to surface: "Structured metadata from first-party and third-party providers (e.g., price, product description) and other third-party content," plus "model responses generated by ChatGPT before it considers any new search results," plus "OpenAI safety standards and product policies" (OpenAI).

The same article states plainly that "product results are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships."

For the deeper shopping research flow, OpenAI lists three inputs: "merchant product data provided through the Agentic Commerce Protocol (ACP)," "publicly available product information," and "other relevant retail sources" (OpenAI).

So there is a feed path, a data-provider path, and a read-your-page path. They are not alternatives. They run together, and a merchant can be strong on one and invisible on another.

The feed path, and who can use it

OpenAI's commerce documentation is direct about access: "Onboarding product feeds in ChatGPT is currently available to approved partners" (OpenAI Developers).

For those with access, the mechanics are documented. Two delivery methods, file upload or API. The recommendation is "to provide the entire feed once a day via file upload, and then send updates throughout the day via the API," with promotions data available through the API only. And the schema is not a new invention: "If your feed uses a Google-compatible product data format, OpenAI will use that formatting" (feed spec).

Two facts change the picture for most small merchants. OpenAI's merchant page states that if you sell through Shopify or Etsy, "your catalog is already integrated, and no additional setup or application is required." And the help center adds that for Shopify merchants, product data reaches ChatGPT through Shopify Catalog with "no additional work required from individual merchants."

Then there is the question OpenAI answers in its own FAQ, which carries the most useful sentence on the whole page for a small business. The FAQ asks whether you need a product feed if ChatGPT already crawls your site, and answers: "No, but feeds give you greater control over how your products appear, helping ensure more accurate and up-to-date information" (OpenAI merchants).

No, but. Your site is being read either way. The feed is about control and freshness, not about eligibility.

Which is why your product page still decides a lot

If you take one thing from OpenAI's documentation, take this line about shopping research: results are "organic and based on publicly available retail sites, reading product pages directly, citing sources, and avoiding low-quality or spammy sites."

Reading product pages directly. Your page is an input, in real time, for a real purchase decision.

And the failure mode is documented too: "some retailers block automated access to their sites. When this happens, shopping research will either skip those sources or rely on other sites with similar products."

Skip you, or use someone else selling the same thing. That is not a ranking penalty. It is a door.

What a page needs in order to be usable here is unglamorous, and we cover the mechanics in how to structure product pages for AI shopping engines:

• The product name, price, and stock state present as real text, not painted in by script after load.
• Product schema with a complete offer carrying price, currency, and availability, kept in sync with what the page shows.
• Specs in a real table with units written out, so a comparison can be built without guessing.
• Automated access allowed, at robots.txt and at whatever bot protection sits in front of it.

Old way: the product page was a shop window for a human who would then decide. New way: the product page is also a data source for a machine building a comparison that person may never scroll past.

What OpenAI says about buying, as of today

Agentic checkout is the part most commonly overstated, so here is the current published position rather than the version from twelve months of headlines.

OpenAI's merchant FAQ states: "We're moving away from a standalone Instant Checkout experience in ChatGPT and prioritizing better shopping discovery and merchant-owned checkout experiences. Users can discover and evaluate products in ChatGPT, but purchases are completed on merchant-owned websites or apps."

The Instant Checkout option still appears in the shopping help article "for some eligible products and merchants," so the surface has not vanished. The stated direction is toward discovery in ChatGPT and checkout on your own site. There are also "no fees on purchases that start in ChatGPT," and shopping is "currently live in the U.S."

This is worth stating flatly because the practical implication runs the other way from the hype. If checkout happens on your site, your product page is not less important in an agentic world. It is the destination.

The damaging admission

Here is exactly what Citedon's scan does and does not query, because this is the page where the distinction matters most.

Our scan queries ChatGPT through OpenAI's API with web search enabled. That is the text answering path: a real model reading the live web and returning cited sources. It tells you whether your page can be retrieved, parsed, and named.

It does not query the shopping product carousel, and it does not query the shopping research flow. It has no visibility into the ACP feed pipeline, into Shopify Catalog, or into the third-party data providers OpenAI names. If your price is wrong inside a product card, our scan will not surface that, and we are not going to imply it will.

What the scan measures is whether the page an engine would read directly is readable: content in real text, valid product markup, offers that match the visible page, crawlers not blocked. That is the input you actually own on every one of the three paths above.

We cannot promise you a place in a product carousel. OpenAI decides that, the policy is theirs, and it is changing while you read this.

One more admission, and it is a real one for this audience. If you are on Shopify, our automated fix layer does not apply to you. Citedon applies fixes through a connected plugin on WordPress, with a preview and a per-fix approval. On Shopify, Wix, Webflow, or a headless storefront, the scan gives you the diagnosis and you apply the changes in your own theme. For a Shopify merchant, OpenAI's own documentation says your catalog is already integrated, so the highest-value work is usually making the page itself parseable, not chasing a feed. What you can actually change on that platform, through the theme, metafields and robots.txt.liquid, is set out in the Shopify AI readiness checklist.

Why this needs re-checking

Every fact on this page was fetched from OpenAI's documentation on 2026-08-06, and several of those pages were themselves updated within the previous week.

Instant Checkout is being de-emphasized. Feed access is limited to approved partners with a self-serve portal stated as coming. Shopping is US-only with expansion planned. Every one of those sentences has a shelf life.

Your catalog moves faster still. Prices change, variants get added, and a theme update can quietly drop the markup that made an offer legible. A product page that read cleanly last quarter is not a product page that reads cleanly now.

How to check what your product pages send

Pick the product you would most want ChatGPT to surface, and find out what a machine reading only that URL can extract: name, price, availability, and specs, or a styled layout with the facts locked inside it.

Run a free scan on that product URL to see how ChatGPT, Perplexity, Gemini, and Claude read it today, and which structural pieces are missing.