Which fields does a ChatGPT product feed require?

The Citead team, updated

In short. OpenAI’s Stable product feed requires item_id, title, description, url, brand, seller_name, image_url, availability, and price, with conditional fields for specific uses.

Definition: A ChatGPT product feed is structured catalog data supplied to OpenAI for shopping and search. As verified on 17 October 2026, OpenAI's product feed specification requires nine fields per item in its Stable file-upload format, with additional requirements for particular integrations.

Key facts

  • OpenAI's Products reference separates the supported Stable upload schema from a Draft schema intended for planning and feedback.
  • OpenAI's Get Started guide limits feed onboarding to approved partners and asks merchants to validate required fields for every record.
  • The file-upload overview describes a full catalog snapshot and recommends updates at least daily.

What belongs in every product row?

The Basic product data table in OpenAI's Products reference marks these fields as required:

FieldRequired content
item_idA stable identifier unique to the item or variant within the feed.
titleThe product name, including the selected variant where relevant.
descriptionA factual description of the item.
urlThe product detail page, with the variant selected where possible.
brandThe brand shown on the product page.
seller_nameThe seller supplying the offer.
image_urlA direct URL for the main image of the selected variant.
availabilityAn explicit supported stock-status value.
priceThe regular item price with its currency.

A CSV export can use this header, matching the Stable field names:

item_id,title,description,url,brand,seller_name,image_url,availability,price

The same reference accepts in_stock, out_of_stock, pre_order, backorder, and unknown for availability. A missing, empty, or unrecognized value rejects the row. Its price example is 79.99 USD. Product and image URLs must be publicly accessible.

When are more fields required?

For variants, meaning selectable versions of the same product, OpenAI's variant rules require group_id, listing_has_variations=true, and variant_dict. Each selection has its own row and distinct item_id. The shared group_id differs from each item identifier.

The same reference labels gtin and mpn, identifiers assigned to trade items and manufacturer parts, as optional in the OpenAI format. Ads processing conditionally requires is_ads_eligible=true. Checkout participation adds is_eligible_checkout, seller_privacy_policy, and seller_tos, alongside a separately enabled checkout integration.

When supplied, separate dimensional measurements require dimensions_unit, and weight requires item_weight_unit. Third-party marketplace offers conditionally require marketplace_seller and feed setup.

A Google-compatible feed has a separate profile that OpenAI must confirm for the registered feed. Its required columns are id, title, description, link, image_link, availability, price, and brand; the registered merchant name supplies seller identity. This profile also conditionally requires gtin or mpn unless identifier_exists=no, and availability_date for preorder or backorder items. Its zero-price exception also requires google_product_category and subscription_cost for supported mobile devices.

How should a feed be checked before upload?

OpenAI's file-upload overview recommends beginning with around 100 items and including every required field in every sample row. Its listed ingestion failures include missing fields, outdated field names, and malformed values.

For ongoing delivery, the same guide asks merchants to overwrite a stable filename or shard set with the latest full snapshot. A shard is one file containing part of the catalog. The Get Started guide also describes API updates throughout the day alongside a daily file upload and directs merchants to use the specification for their chosen delivery method.

FAQ

Can I use the Draft schema for a production file upload?

OpenAI's Products reference directs you to Stable for supported file uploads. Draft is published for planning and feedback.

Does completing the required fields finish onboarding?

OpenAI's Get Started guide requires approved-partner access. Field validation is one step in its integration process, followed by upload and ongoing updates.

Sources

  1. OpenAI, "Products", accessed 17 October 2026
  2. OpenAI, "Get Started", accessed 17 October 2026
  3. OpenAI, "Overview", accessed 17 October 2026

Check how a page reads to answer engines. Open the GEO score analyzer

Test a page and see what to fix first

The analysis usually takes less than 10 seconds and needs no account.

400,000+ pages scored in the public benchmark