ChatGPT product feed: how to get your products into ChatGPT
Shoppers increasingly ask ChatGPT what to buy. Whether your products can appear in those answers comes down to one unglamorous thing: a product feed OpenAI can read. Here is what that takes - from OpenAI's own spec, not from hearsay.
ChatGPT does not browse your store the way a person does. When it recommends products, it draws on machine-readable catalog data - registered product feeds and what its crawlers can read on the open web. That has a blunt consequence: stores with clean feeds are recommendable, stores without them are invisible, and the quality of your products never gets a vote.
What OpenAI's product feed actually requires
OpenAI documents a flat-file schema its systems ingest and index. The core, verbatim from the spec: a UTF-8 delimited file (.tsv or .csv, gzip accepted), one product or variant per row, one header row with lowercase, underscore-separated field names - and these eight fields on every row:
Details that trip people up: price is an amount followed by a three-letter currency code, link and image_link must be valid HTTP(S) URLs, and title/description are plain values - structured alternatives are not mapped. Feeds are registered with OpenAI, including a merchant display name that OpenAI uses as the seller identity on every row, overriding whatever the file says.
Already have a Google Shopping feed? You are close
The spec documents a Google-compatible path: if OpenAI confirms your registered feed supports it, you can upload a Google-format product feed without renaming columns. OpenAI maps the supported fields into its schema and ignores the rest. Two caveats from the spec itself: it covers the core profile (not every program, market or attribute), and g:-prefixed XML names are not supported - the flat-file rules above still apply.
Practically: if your Google Merchant feed is clean, most of your ChatGPT work is already done. If it is not, fixing it now pays into both channels at once.
The five steps, in order
- Let OpenAI's crawlers in. A feed gets your products indexed; your robots.txt decides whether OpenAI's systems may read your pages at all. Blocking GPTBot or OAI-SearchBot while submitting a feed is working against yourself. Our store readiness check flags exactly which AI crawlers you block.
- Get the catalog data right at the source. Every feed inherits the store's data: missing brands, thin descriptions and dead image links do not improve on export. The product page check shows what a machine actually sees on one of your products.
- Build the feed against the spec - the eight required fields, the price and URL formats, one row per variant. If you run WooCommerce or Shopify, you do not have to hand-build this: platforms like ours generate a spec-shaped feed from the synced catalog.
- Validate before you submit. Our free ACP feed checker grades a feed against the OpenAI product feed spec - required fields, identifier weighting, formats - in your browser, no signup.
- Register with OpenAI and keep it fresh. Participation runs through OpenAI's commerce program with a registered feed and merchant name. After that, freshness is the discipline: a feed that says in-stock for a sold-out product is worse than no feed.
The same data sells twice
Everything above is catalog work, and it pays a second time: the machine-readable catalog that makes you visible inside ChatGPT is what powers an AI storefront on your own domain - where the conversation is yours, the branding is yours, and a sales agent can answer from live stock. One clean catalog, two channels. For the protocol behind the agent side, see our agentic commerce guide.
Frequently asked questions
How does ChatGPT decide which products to show?
From machine-readable product data. OpenAI ingests registered product feeds (a flat-file schema documented in their developer docs) and crawls the open web. If your catalog is not readable that way, ChatGPT cannot recommend it - however good the products are.
What fields does an OpenAI product feed require?
Per OpenAI's feed reference (checked August 2026): every row needs id, title, description, link, image_link, availability, price and brand. Price is written as an amount followed by a three-letter currency code, links must be valid HTTP(S) URLs, and the file is a UTF-8 delimited flat file (TSV or CSV, gzip supported).
Can I reuse my Google Shopping feed for ChatGPT?
Often, yes. OpenAI documents a Google-compatible path: if they confirm your registered feed supports it, a Google-format product feed can be uploaded without renaming columns - OpenAI maps the fields and ignores what it does not use. It covers the core profile, not every attribute or program.
Do I have to register with OpenAI?
Yes - feeds are registered with OpenAI, including a merchant display name that becomes the seller identity on every row. The developer documentation describes the format; participation runs through OpenAI's commerce program.
Is this the same as the Agentic Commerce Protocol (ACP)?
The feed is one building block of it. ACP describes how AI agents discover and transact against a catalog; the product feed is how your products become visible to ChatGPT in the first place. You can validate a feed against the spec with our free ACP feed checker - no signup.
What about shoppers who never leave my own store?
Same data, second use: the machine-readable catalog that makes you visible in ChatGPT is exactly what powers a conversational storefront on your own domain. If you get the catalog right once, both channels work from it.
Paste your feed URL and get a graded report against the OpenAI product feed spec - free, no signup.