What an agent-ready product feed is

A product feed was originally built for advertising and recommendations: ID, title, price, image, category. An agent choosing a product on the shopper’s behalf asks more questions: what colour is this variant, how much is delivery, can it be returned. If the answer is not in the data, the agent either picks another store or gets it wrong. In OpenAI’s words, the feed exists so that ChatGPT accurately indexes and displays your products with up-to-date price and availability.

What the specifications require (as of September 2026)

Requirement OpenAI Product Feed Spec (ChatGPT) Google Merchant Center and UCP
Required fields item_id, title, description, url, brand, seller_name, image_url, availability, price id, title, description, link, image_link, availability, price; brand for new products
Identifiers gtin, mpn are optional GTIN strongly recommended, MPN when there is no GTIN
Variants group_id, variant_dict item_group_id
Text length title up to 150 characters, description up to 5,000 title up to 150, description up to 5,000
Format and updates JSONL, CSV or TSV; full export over SFTP at least once a day Price and availability match the landing page, structured data and checkout
Checkout Separate integration, is_eligible_checkout flag native_commerce attribute; UCP checkout in AI Mode and Gemini — US, Canada, Australia

How it differs from an advertising feed

  • Delivery and returns. The shipping, accepts_returns, return_deadline_in_days and return_policy fields — an agent needs them before payment, not after.
  • Reviews as numbers. review_count and star_rating — the agent compares products on them.
  • Seller. seller_name, and for checkout, links to the privacy policy and terms of sale.
  • Eligibility flags. is_eligible_search and is_eligible_checkout define which products can be found and which can be bought.

Feed data has to match the schema markup on the page and what the shopper sees at checkout. The purchase itself then runs through a protocol — ACP or UCP.

Feed readiness checklist

Check Ready when
Identifiers Every variant has a stable ID; GTIN is filled in for every product that has one
Titles and descriptions Brand, model, key variant parameter; the description is factual, not promotional
Variants Colour, size, volume — separate rows sharing one group_id
Price and availability Match the website and checkout; exported at least once a day
Delivery and returns Cost, delivery times, return window in days
Images The main photo shows this exact variant
Data source Attributes come from a PIM rather than being edited by hand in the feed

The feed is the first step towards agent-readiness: without it, checkout protocols have nothing to stand on.