ChatGPT and Google both decide what to show shoppers largely from structured product data, so your feed is the single most useful thing to fix for AI agents. OpenAI's spec requires nine fields per product and rejects rows with bad availability data. Google reuses your Merchant Center feed and adds one attribute for checkout. This guide lists what each platform asks for as of November 2026, based on their published specs.

Required fields in OpenAI's feed spec
9
Accepted OpenAI feed formats
JSONL, CSV, TSV
Default checkout eligibility in ChatGPT feeds
Off
native_commerce, for Google's Buy button
1 attribute

What does OpenAI's product feed require?

OpenAI's product feed spec lists nine fields that must have values on every row:

Field What to put in it Common mistake
item_id Your stable unique ID for the sellable item Changing IDs when you re-export
title Product name with key attributes Marketing slogans instead of the product
description Plain-text description HTML or truncated text
url Product page URL Redirects or tracking-only URLs
brand Manufacturer or brand Leaving blank for own-brand items
seller_name Your store name Platform name instead of yours
image_url Main product image Placeholder or watermarked images
availability Stock status value from the spec Missing or invalid value gets the row rejected
price Price with currency Price that does not match the page

Optional fields worth filling

The spec defines optional fields that help agents compare and recommend:

  • Variants: group_id, listing_has_variations, variant_dict, offer_id. Without them, a jacket in five sizes looks like five unrelated products.
  • Identifiers: gtin and mpn, which let agents match your listing to the same product elsewhere.
  • Pricing: sale_price. Update it when the sale starts and ends.
  • Shipping: shipping_price or shipping. OpenAI's page tells you to confirm which one your integration accepts.
  • Returns: accepts_returns, return_deadline_in_days, return_policy.
  • Reviews: review_count, star_rating.

Shoppers increasingly ask agents for "free returns" or "arrives by Friday". If those facts are not in your feed, the agent has to guess or skip you.

The three eligibility flags

Flag Alias Default Effect
is_eligible_search enable_search True Product can appear in ChatGPT; false also disables checkout
is_eligible_checkout enable_checkout False Only works if search is on and checkout is enabled for your integration
is_ads_eligible is_eligible_ads Off unless a feed-level default applies Ads eligibility

The spec is explicit that setting the checkout flag "doesn't complete checkout onboarding". Checkout requires the separate integration described in ACP for merchants.

How do you deliver the feed to OpenAI?

  1. ExportGenerate JSONL, CSV or TSV from your catalog.
  2. ValidateCheck the nine required fields and availability values on every row.
  3. UploadSend the file to OpenAI's SFTP endpoint, or use the Feeds API.
  4. Refresh on changeRe-send when prices, sales or stock change.
  5. Watch rejectionsFix rows OpenAI rejects, starting with availability.

OpenAI's commerce docs describe the feed as the first step of an ACP integration and list a Feeds API with products and promotions endpoints alongside SFTP upload. For CSV and TSV, quote cells that contain commas, quotes or newlines, double any embedded quotes, and serialise nested objects as JSON. The spec sets no fixed refresh schedule; it says to update when a sale starts or ends and when availability changes.

How do shoppers see this?

ChatGPT's shopping research compares products and points shoppers to where they can buy. A clean feed is how you end up in that comparison. See how it looks from the other side in ChatGPT shopping research.

What does Google need for AI Mode and Gemini?

Google's UCP getting-started page says you reach shoppers in AI Mode and Gemini using your existing Merchant Center account and shopping feeds. There is no separate AI feed. To make a product purchasable, Google's help page says the listing must use the native_commerce (checkout_eligibility) attribute: "Only product listings using the native_commerce(checkout_eligibility) product attribute will display the 'Buy' button."

That attribute is only useful once you are accepted into UCP checkout, which as of November 2026 is limited to selected merchants with products eligible in the US, Canada or Australia. Details are in UCP and AP2 for merchants.

What about Shopify, Copilot and other agents?

Shopify merchants do not build feeds by hand. Shopify's March 2026 announcement says Agentic Storefronts keep products, inventory and pricing synced across ChatGPT, Microsoft Copilot, AI Mode and Gemini, and that products are discoverable in ChatGPT by default. Brands not on Shopify can add products to Shopify Catalog through its Agentic plan. See Shopify and agentic commerce.

Browser agents such as Claude Duet and Meta Muse read your product pages directly. For them, the equivalent of a feed is good on-page structured data: consistent prices, visible stock status and variant selectors that work without hovering.

How do you keep the feed accurate?

Agents repeat your data to shoppers with confidence. A price or stock error in the feed becomes a wrong promise in the chat, then a support ticket or a return. Practical rules:

  • Generate the feed from the same source as your product pages, not a separate spreadsheet.
  • Push stock changes for fast sellers more often than the rest of the catalog.
  • Put sale start and end into your publishing schedule, since dates in the feed do not trigger changes.
  • Audit a sample weekly: compare feed price, page price and checkout price for 20 random items.
  • Keep return policy fields in line with your published policy, because shoppers will use guides like returning an item your agent bought.

Peak season

Black Friday falls on 27 November 2026. Shoppers using agents for price tracking will trigger purchases the moment your feed shows a drop, so make sure sale prices go live in the feed at the same time as on the site.

Bottom line

Fill OpenAI's nine required fields, add GTINs, variants, shipping and returns, and update whenever price or stock changes. Reuse your Merchant Center feed for Google and add native_commerce once you are approved. Everything else in making a store agent-ready builds on this one file.