SellerSort
Try it free

Which product data AI shopping assistants read: AI Mode, Gemini, ChatGPT

AI shopping assistants read the same kind of product data a shopping feed already carries, from two places. Google’s AI Mode, Gemini and Business Agent work from Merchant Center data, and Business Agent also reads your website. ChatGPT works from a feed sent to OpenAI. Each adds a few fields of its own: conversational attributes for Google’s AI surfaces, checkout eligibility for purchases through Google’s UCP, seller_name and a category path for OpenAI. The core is shared: a clear title and description, an accurate price and stock level, and a specific product category.

What this comes down to

  • Google’s AI Mode, Gemini and Business Agent build on Merchant Center product data; ChatGPT uses a separate feed sent to OpenAI.
  • Business Agent also reads your website, so policies and size guides that only exist in a support inbox are invisible to it.
  • Conversational attributes, UCP eligibility flags and OpenAI’s seller_name are the surface-specific fields; the rest of the data is shared.
  • A specific product category matters on every surface, both for matching and for keeping restricted products out of checkout.
  • No surface publishes how it weighs individual fields, so complete and accurate data is the only advice that holds for all of them.

SellerSort Team · Updated

Which surface reads what

Product data read by AI shopping surfaces
SurfaceReads fromFields specific to it
Google AI Mode and Gemini, product discoveryMerchant Center product dataConversational attributes, all optional
Business Agent on Google SearchMerchant Center data and your websiteNone; settings live in Merchant Center (guide)
Checkout on AI Mode and Gemini (UCP)Merchant Center data and your checkout APInative_commerce(checkout_eligibility), consumer_notice, merchant_item_id (guide)
ChatGPT (ACP)A product feed sent to OpenAIitem_id, seller_name, product_category as a path (guide)

Access differs as much as data. Business Agent is rolling out to US stores; Google’s checkout and OpenAI’s feed are both by approval for now. The ACP and UCP comparison covers the two checkout protocols in detail.

Every AI shopping surface so far reads a product feed; none of them has replaced it.

The core every surface shares

Google’s required attributes and OpenAI’s required fields are mostly the same data under different names:

Required product fields in Google and OpenAI feeds
What it isMerchant CenterOpenAI
Product IDiditem_id
Nametitletitle
Descriptiondescriptiondescription
Product pagelinkurl
Main imageimage_linkimage_url
Pricepriceprice
Stockavailabilityavailability
Brandbrand, for most new productsbrand
SellerFrom the accountseller_name

A store with a clean Google feed has done most of the work for every surface on this page. The product feed guide covers the basics if you are starting from nothing.

Why the category runs through all of them

The product category is optional in both Google’s and OpenAI’s specs, yet it does work on every surface here. It tells a surface what kind of product a row is, which matters most when the title alone is ambiguous. And it is the most practical way to keep restricted products out of AI checkout: alcohol, tobacco, weapons and medicines sit in known parts of Google’s taxonomy, so a correct category turns an exclusion list into a filter.

The formats differ: Merchant Center prefers the numeric ID, OpenAI’s file format wants the path. Keeping both together solves that once. The taxonomy download has every ID beside its path, the Category Finder answers for one product, and the Bulk Mapper returns both for a whole catalog.

An order of work

This is a sensible order for a store starting now, not a requirement from any platform:

  1. Fix the core feed. Titles, descriptions, prices, stock and a specific category on every product, matching the product page.
  2. Fix the store’s trust details. Return policy, visible contact details and a working checkout. Google’s UCP requires the first two, and Merchant Center’s store requirements already ask for all three; the misrepresentation guide lists them.
  3. Add detail where shoppers have questions. Conversational attributes in a supplemental data source, and the answers on the product page itself.
  4. Apply for the protocols that fit your market. Both checkout programs are by approval, so the earlier steps are what you can control in the meantime.

One date belongs on the same list: if your store syncs to Merchant Center through its own code on the Content API for Shopping, that API shuts down in early 2027. The sunset guide has the details.

Questions

What product data does Google AI Mode use?

Google’s AI surfaces build on the product data in Merchant Center. Google also offers optional conversational attributes, such as questions and answers and variant options, that it says help AI-driven surfaces like AI Mode understand products.

Do I need a separate feed for ChatGPT?

Yes. ChatGPT reads a product feed sent to OpenAI, by SFTP or API, in OpenAI’s format or a Google-compatible text format that OpenAI confirms for your feed. Onboarding is currently for approved partners.

Is there such a thing as SEO for AI shopping assistants?

No surface publishes how it weighs individual product fields. What every one of them documents is the data it reads, so complete, accurate and specific product data is the advice that holds across all of them.

Does the product category matter for AI shopping?

It is optional in Google’s and OpenAI’s specs, but it tells a surface what kind of product a row is, and a correct category is the practical way to keep restricted products such as alcohol or weapons out of AI checkout.

Read next