Product structured data is the part of your catalog a machine reads before any person does. When a shopper asks ChatGPT or Google for a product, the answer is built from fields like price, availability and brand. If a field is missing or malformed, your product can drop out of the answer while a competitor with cleaner data stays in.
What product structured data do AI shopping agents read?
There are two main routes into AI shopping answers today. Google reads markup on your product pages. ChatGPT reads a product feed that merchants submit to OpenAI. The vocabularies differ, but the facts underneath are the same: what the product is, what it costs, whether you can ship it, and who sells it.
Neither company publishes a ranking formula, so we leave ranking aside. Both publish the fields they need, and a product that fails those lists is out before ranking starts. If you want the wider picture first, read Agentic Commerce 101.
The fields Google requires for merchant listings
Google's merchant listing documentation splits product structured data into required and recommended properties.
Required on Product: name, image and offers. The offer must be an Offer, because the merchant has to be the seller.
Required on the Offer: price and priceCurrency, the current price and its three-letter ISO 4217 code.
Recommended: a GTIN (gtin, gtin8, gtin12, gtin13, gtin14 or isbn), brand, mpn, sku, color, size, material, availability, itemCondition, priceValidUntil, hasMerchantReturnPolicy and shippingDetails.
Three details from the documentation trip up real catalogs:
- Use the narrowest GTIN that fits the exact item, and give one brand name only.
- Give one
availabilityvalue and oneitemConditionvalue, never two. - Google says a listing may not show if
priceValidUntilholds a date in the past.
Google also recommends putting your return policy and shipping rules on your Organization markup once, and using product-level versions only where a product differs. That keeps policy answers consistent across the whole catalog.
The nine fields ChatGPT's product feed requires
OpenAI publishes a product feed specification for merchants who want their products shown in ChatGPT. Its nine required fields are:
item_idtitledescriptionurlbrandseller_nameimage_urlavailabilityprice
The format rules are strict:
- Price is an amount, a space and an uppercase currency code, for example
79.99 USD. - Availability must be
in_stock,out_of_stock,pre_order,backorderorunknown. The spec says an empty or unrecognized value rejects the row. - GTIN is optional, but when you send one it must be exactly 8, 12, 13 or 14 digits with a valid check digit, with leading zeros kept and no spaces or dashes. If a product has no assigned GTIN, you leave the field out.
Never fill the GTIN field with a made-up barcode. A wrong check digit is simple arithmetic for any system to catch, and a real number copied from another product points shoppers at the wrong item. For how ChatGPT picks among products once they are in, see how ChatGPT picks products to recommend.
Where the two lists differ
In product structured data, the same fact goes by different names in the two specifications:
| Fact | Google (schema.org markup) | ChatGPT product feed |
|---|---|---|
| Product name | name | title |
| Picture | image | image_url |
| Price | price + priceCurrency | price as 79.99 USD |
| Stock | availability (recommended) | availability (required, 5 values) |
| Brand | brand.name (recommended) | brand (required) |
| Barcode | GTIN property (recommended) | gtin (optional, strict digits) |
| Seller | the merchant's own page | seller_name (required) |
The practical risk is drift. If your page markup says one price and your feed says another, at least one of them is wrong, and an assistant may quote the wrong one to a shopper. Keep one clean record per product and generate both outputs from it, so a price change reaches the page and the feed together.
If you sell on Shopify, the Shopify checkout checklist for AI agents covers the store side. If you do not, getting listed in ChatGPT Shopping without Shopify covers the feed side.
A 30-minute check of your own product data
You do not need a tool to start. Take ten products, including your three best sellers, and check their product structured data by hand.
Open the page markup
Use Google's Rich Results Test on each product URL. Note whether it finds a Product with name, image and an offer with price and currency.
Compare price and stock with your feed
Open the same product in your feed file. The price, currency and stock status should match the page exactly.
Check every GTIN
Each GTIN should be 8, 12, 13 or 14 digits with no spaces. If you cannot confirm where a barcode came from, remove it rather than guess.
Look for expired dates
Search the markup for
priceValidUntil. Per Google's documentation, a past date can keep the listing from showing.Write down what is missing
List each missing required field by product. That list is your fix order: required fields first, then brand and GTIN.
If more than two of the ten fail, the cause is usually the page template or the feed export, so one fix repairs the whole catalog. A free scan of your site shows which product structured data fields AI assistants can read. The glossary entry for AI Readiness explains what that scan measures.
Frequently asked questions
They serve different readers. Google reads the markup on your product pages, and ChatGPT reads the feed you submit to OpenAI. If you want to appear in both, you need both, ideally generated from the same product record so they never disagree.
Google lists a GTIN as recommended, not required, and OpenAI's spec says to leave the field out when no GTIN is assigned. Handmade and private-label goods often have none. Leave it empty rather than invent one.
The spec accepts in_stock, out_of_stock, pre_order, backorder and unknown. An empty or unrecognized value rejects the row, so a typo can remove a product from the feed.
No. Neither Google nor OpenAI publishes a ranking formula. Complete, accurate fields make a product eligible to be considered. Missing required fields make it ineligible.
Sources
- Merchant listing (Product, Offer) structured data, Google Search Central.
- Products: product feed specification, OpenAI developer documentation.
- Product type, schema.org.
- Rich Results Test, Google.