Skip to main content

Learn

SalesChef gets your product data ready for AI shopping agents.

Google and OpenAI each publish a product data specification, and both ask for the same core facts: an identifier, title, description, link, image, price, availability and brand. Neither publishes how it picks products to show. This guide lists the fields exactly as the sources name them.

Google Merchant Center: the required fields.

Google’s product data specification lists these attributes as required. Field names are quoted exactly. Google says title and description must match the landing page, and price must match it too.

id
A unique identifier, up to 50 characters. Google says to use the SKU where possible.
title
The product name in plain text, up to 150 characters.
description
An accurate description, up to 5000 characters.
link
The landing page URL on your verified domain.
image_link
The main product image URL, crawlable by Google.
price
A number plus an ISO 4217 currency, such as 15.00 USD.
availability
One of in_stock, out_of_stock, preorder or backorder.
brand
Required for new products except movies, books and musical recordings.
gtin, mpn, condition
gtin (the Global Trade Item Number) is strongly recommended where one exists, numeric only. mpn is required only if the product has no manufacturer-assigned number. condition is required for used or refurbished products and optional for new.
Also named in the specification
google_product_category, product_type, additional_image_link, color, size, age_group, gender, material, pattern, item_group_id, sale_price, product_highlight, product_detail, and weight and size attributes. Some are required for certain categories such as apparel or multi-color products.

OpenAI: the product feed specification.

OpenAI’s feed specification says it covers discovery in ChatGPT. Its Agentic Commerce documentation says to start an integration by sharing a structured product feed, sent by file upload to an SFTP server or through an API. Ads and checkout have separate requirements.

Required for discovery
item_id, title, description, url, brand, seller_name, image_url, availability and price.
Required for variants
group_id, listing_has_variations and variant_dict.
Required for checkout
is_eligible_checkout, seller_privacy_policy and seller_tos. Setting is_eligible_checkout alone does not complete onboarding.
Optional identifiers and attributes
gtin, mpn, condition, product_category, material, color, size, gender, age_group, dimensions, weight, additional_image_urls and sale_price.
Optional policies and reviews
accepts_returns, return_deadline_in_days, return_policy, review_count and star_rating.
Format rules
price uses the form 79.99 USD. availability is one of in_stock, out_of_stock, pre_order, backorder or unknown. gtin must be 8, 12, 13 or 14 digits with a valid check digit. URLs must be publicly accessible.

schema.org and Google structured data on the page.

schema.org describes properties. It does not mark any Product property as required. Google sets its own requirements for rich results, and says structured data is not required to appear in its AI features.

Google merchant listings: required
Product: name, image and offers. Offer: price (or priceSpecification.price) and priceCurrency (or priceSpecification.priceCurrency).
Google merchant listings: recommended
Product: aggregateRating, brand.name, category, color, description, gtin (or gtin8, gtin12, gtin13, gtin14, isbn), material, mpn, pattern, review, size, sku. Offer: availability, hasMerchantReturnPolicy, itemCondition, priceValidUntil, shippingDetails, url.
Google product snippets: required
Product: name, plus one of review, aggregateRating or offers.
schema.org properties
Product defines name, brand, gtin13, mpn, sku, image, description, offers, aggregateRating, review, color and model. Offer defines price, priceCurrency, availability, itemCondition, priceValidUntil and seller.

The same fact has three names.

Each source names the same fact differently, and the allowed values differ. Map your own product record once, then check each output.

Identifier
Google: id. OpenAI: item_id. schema.org: sku.
Product name
Google: title. OpenAI: title. schema.org: name.
Link
Google: link. OpenAI: url. schema.org: Offer url.
Image
Google: image_link. OpenAI: image_url. schema.org: image.
Price
Google: price. OpenAI: price. schema.org: Offer price with priceCurrency.
Stock
Google: availability, with preorder. OpenAI: availability, with pre_order. schema.org: Offer availability.

What is not published.

Neither Google nor OpenAI publishes how it picks which products to show. OpenAI says eligibility does not guarantee display, and its feed specification sets no fixed refresh schedule: it says to update price and availability whenever they change. Google says indexing and serving are not guaranteed. This guide reports only what those documents say. A complete, accurate record is the part you control.

See a finished page and its shopper answers.

Your questions, answered.

What product data do AI shopping agents need?

At minimum, the fields each platform lists as required. Google Merchant Center requires id, title, description, link, image_link, price, availability and brand. OpenAI’s feed requires item_id, title, description, url, brand, seller_name, image_url, availability and price.

Do I need a Global Trade Item Number?

Google calls the gtin field (the Global Trade Item Number) strongly recommended where one exists, and requires mpn when a product has none from the manufacturer. OpenAI lists that field and mpn as optional. Either way, a correct number helps systems match your product to the same product elsewhere.

Is schema.org markup required?

schema.org does not mark any Product property as required. Google does for its rich results: name, image and offers for merchant listings. Google also says structured data is not required to appear in its AI features.

Does a complete feed guarantee my products appear?

No. OpenAI’s specification says eligibility does not guarantee display, and Google says indexing and serving are not guaranteed. Neither publishes the full selection criteria.

References

Send us one product page. We’ll show you how SalesChef sells it.

Start with the world’s hardest-working product pages (and more).