The OpenAI Product Feed Spec, Field by Field, With Shopify Mappings

The OpenAI product feed spec defines how a merchant sends product data straight to OpenAI so ChatGPT can show it in shopping results. It asks for one row per purchasable variant, with nine required fields and about forty optional ones, delivered as a full catalog snapshot at least once a day. If you sell on Shopify, you almost certainly don't need to build it: OpenAI says Shopify catalogs are already integrated into ChatGPT through Shopify Catalog. But the spec is still the clearest list anyone has published of what ChatGPT can read about a product, so it's worth knowing where each field lives in your admin.

This guide maps each field to the Shopify field, setting or metafield it comes from, and flags where Shopify has no native home. Field rules come from OpenAI's Products spec and Shopify's help center as of October 22, 2026, listed under Sources.

OpenAI product feed

A structured catalog file (or API stream) that a merchant shares with OpenAI so ChatGPT can index its products and show current price and availability. It follows OpenAI's published Products spec and is part of the Agentic Commerce Protocol.

Agentic Commerce Protocol (ACP)

An open standard, developed by OpenAI and Stripe and licensed under Apache 2.0, for how AI agents and businesses exchange product data, checkout sessions and payment tokens. OpenAI says merchants start an ACP integration by sharing a product feed.

Shopify Catalog

Shopify's structured index of merchants' products, which feeds agentic storefronts in AI channels such as ChatGPT, Microsoft Copilot and Google AI Mode. It shares title, description, options, images, price, availability and other key attributes, and updates continuously.

Do Shopify merchants need to build an OpenAI product feed?

No, not if your store is eligible for Shopify's agentic storefronts. OpenAI's shopping help article says product data for Shopify merchants is already integrated into ChatGPT through Shopify Catalog, and no additional work is required from individual merchants. OpenAI's merchant page goes further: if you sell through Shopify or Etsy, your catalog is already integrated, and no additional setup or application is required.

Shopify says the same from its side. Its ChatGPT help page says products are discovered through Shopify Catalog, the channel is active by default for eligible stores, and there are no fees for selling in ChatGPT. You manage it under Sales channels → Agentic, not by exporting a file.

Direct feeds are for everyone else, and access is gated. OpenAI's get-started guide says onboarding product feeds is currently open to approved partners who apply through chatgpt.com/merchants, and that page says applicants join a waitlist. OpenAI's March 24, 2026 announcement also names third-party providers, including Salesforce and Stripe, as delivery paths.

How does a direct feed get to OpenAI?

OpenAI recommends the full feed once a day by file upload, then updates through the day by API. Promotions can only be sent by API.

1. 01

   Apply and wait for approval

   Apply at chatgpt.com/merchants. Direct feeds are open to approved partners; OpenAI says a self-serve merchant platform is being built.

2. 02

   Validate a small sample

   Start with around 100 items, every required field in every row, and QA the first full snapshot before automating.

3. 03

   Push full snapshots by SFTP

   A complete export at least daily. OpenAI prefers Parquet (ideally zstd-compressed) and accepts jsonl.gz, csv.gz and tsv.gz, in UTF-8, with stable file names overwritten each run and up to 500,000 items per shard.

4. 04

   Send intraday changes by API

   PATCH /product_feeds/{id}/products upserts products matched by id. Its schema differs from the file: nested variants, and prices as integers in minor units.

5. 05

   Remove products deliberately

   A product missing from a snapshot is kept for up to 14 days. To pull one sooner, set is_eligible_search to false.

What are the nine required fields, and where does each come from in Shopify?

A missing or unrecognized availability value rejects the row outright. Elsewhere, OpenAI warns that not every invalid value is rejected, so a bad value can load and still mislead.

Price and sale price: the one mapping that inverts

Shopify and OpenAI name prices the other way round. In Shopify, Price is what you charge and Compare-at price is the higher original shown beside it. In OpenAI's spec, price is the regular price and sale_price is the current sale price, which must be greater than zero, strictly lower than price and in the same currency, or it's ignored. So during a sale, the compare-at price goes in price and the Shopify price goes in sale_price. Outside a sale, send the Shopify price as price and omit sale_price.

Only do this when the compare-at price is a real regular price. Shopify's Catalog Mapping has its own switch for sending compare-at prices to AI agents. Sale dates don't schedule anything in OpenAI's spec, so update the feed when a sale starts and ends. When AI shopping quotes the wrong price covers what happens when these drift.

How do Shopify variants map to OpenAI's variant fields?

OpenAI wants a separate row for every purchasable selection, tied together by a parent ID. Shopify's product-and-variant model maps onto it cleanly:

• `group_id` ← the Shopify product ID. Every variant row shares it, and it must differ from each row's item_id.
• `listing_has_variations` ← true on every variant row of a product that has options.
• `variant_dict` ← the variant's option names and values, such as {"color":"Black","size":"10"}. Use the same option names across the group, and no duplicate combinations.
• Single-variant products ← send one row without the three fields above. A group_id that equals item_id doesn't create a group.
• `offer_id` ← optional. If you use it, never put the price inside the ID.

Group only variants of the same product as presented on your site; each row carries its own price, availability, URL and images. Shopify Catalog solves the same grouping question with Combined Listings or Catalog Mapping. Our guide to variant data that doesn't confuse AI agents covers the structure in detail.

GTIN and MPN

gtin is optional but strict: exactly 8, 12, 13 or 14 digits with a valid check digit, no spaces or dashes, leading zeros kept. In Shopify it comes from the variant's Barcode. Shopify lets you add up to 20 barcodes per variant and type them (GTIN, UPC, EAN, ISBN, ASIN, or Custom, which takes any value up to 255 characters), so filter to real GTINs before export. An ISBN-10 isn't accepted; send the ISBN-13. Shopify also reserves standard metafields for UPC (facts.upc), EAN (facts.ean) and ISBN (facts.isbn) if an app writes them there.

mpn has no native Shopify field. Keep it in a product or variant metafield if the manufacturer assigns one. OpenAI is explicit: submit it with brand, and don't invent a value to stand in for a missing GTIN. Your SKU is not an MPN unless you're the manufacturer and they're the same thing.

Which optional fields matter, and where do they live in Shopify?

None are required, but they're the facts shoppers filter and compare on. OpenAI says optional fields can improve answer quality, and to leave one out if it needs brittle transforms.

Category metafields are the attributes of Shopify's Standard Product Taxonomy, and which ones appear depends on the category you assign. Where one doesn't apply, omit the OpenAI field: the spec says to leave unknown optional values out rather than send placeholders. The product attribute schema design guide covers how to decide which attributes each category needs.

What do the search, checkout and ads flags do?

The spec carries three opt-in flags, and each is independent of the others:

• `is_eligible_search` defaults to true. Setting it to false removes the item from search and also disables checkout for it. OpenAI notes eligibility doesn't guarantee display.
• `is_eligible_checkout` defaults to disabled and only works when checkout is separately enabled for your integration. It also calls for seller_privacy_policy and seller_tos (on Shopify, /policies/privacy-policy and /policies/terms-of-service).
• `is_ads_eligible` is for ChatGPT ads, off unless your feed sets an Ads default, and has no effect on organic search. See ChatGPT ads for ecommerce.

Read the checkout flag in context. Since March 24, 2026, OpenAI lets merchants use their own checkout while it focuses on product discovery, and its merchant page says shoppers go to your site or app by default, with no fees on those purchases. What happened to Instant Checkout walks through the timeline. For a Shopify store the question doesn't arise: on ChatGPT, agentic storefronts send the purchase to your own online store checkout.

What breaks an OpenAI feed most often?

OpenAI names three common ingestion failures: missing required fields, outdated or non-spec field names, and malformed values. They usually come from these habits:

• Prices in the wrong shape. $79.99 and 79.99 don't match the money format. It has to be 79.99 USD.
• Placeholder strings. null or n/a in an optional field is worse than leaving it empty. Only availability accepts unknown.
• Identifiers stored as numbers. Spreadsheets can strip leading zeros from GTINs and IDs. Keep identifiers as strings.
• Legacy column names. Old aliases such as sku for item_id still load, but send one name per value, and migrate old Custom_variant1_category pairs to variant_dict.
• Stale snapshots. OpenAI says to update the feed whenever price or availability changes, not only on the daily run.

How do I make my Shopify data feed-ready without building a feed?

Treat the spec as a field audit for Shopify Catalog, which carries what your admin holds. Catalog Mapping, under Sales channels → Agentic → Sources → Shopify Catalog → Review product data mapping, lets you choose where title, description and category come from (product fields, metafields or metaobject references) and how variants group.

1. 01

   Fix the required nine first

   Every variant needs a title with its options, a real description, an image, a vendor that holds the brand, a price above 0 and accurate stock. The title, image and price also decide Shopify Catalog eligibility.

2. 02

   Assign a product category to everything

   The category unlocks category metafields (color, size, material, target gender, age group, condition), the same facts OpenAI's optional fields ask for.

3. 03

   Clean up barcodes

   Set each real GTIN with the right barcode type and check digit. Leave Custom-type internal codes out of anything you call a GTIN.

4. 04

   Set return rules and policies

   Configure the return window and final-sale exceptions under Settings → Policies, and complete the refund, privacy and terms pages. ChatGPT via Shopify requires all three policies.

5. 05

   Connect reviews to standard metafields

   Make sure your review app writes reviews.rating and reviews.rating_count, so the aggregate exists as data and not only as a theme widget.

6. 06

   Map anything custom

   If material, dimensions or grouping live in custom metafields or tags, use Catalog Mapping where it supports the field, and keep the value in a structured field rather than prose.

The catalog enrichment for AI playbook sets the order of work for a large catalog. For how this route compares with Google Merchant Center, Microsoft and Perplexity, start with the AI shopping channels guide.

If you'd rather hand over the catalog work, from category metafields to GTIN cleanup and Catalog Mapping, that's what Agentic Commerce Setup covers for Shopify stores.