SellerSort
Try it free

ACP vs UCP: OpenAI’s and Google’s agentic commerce protocols compared

ACP, the Agentic Commerce Protocol, connects merchants to shoppers in ChatGPT. UCP, the Universal Commerce Protocol, is Google’s standard for purchases on AI Mode in Search and the Gemini app. Both are open standards, both are invitation-only for now, and both start from the same thing: an accurate product catalog. They differ in where the catalog lives. ACP takes a feed sent to OpenAI by SFTP or API; UCP reads the product data already in Merchant Center. For most stores the choice is not one or the other, because the work that matters most carries over.

What this comes down to

  • ACP connects merchants to ChatGPT; UCP covers purchases on Google’s AI Mode in Search and the Gemini app.
  • OpenAI onboards approved partners through an application form; Google runs a waitlist and approves each integration.
  • ACP takes its own product feed by SFTP or API, while UCP builds on the product data already in Merchant Center.
  • In OpenAI’s feed the category is a path in product_category; in Merchant Center it is google_product_category, ideally as an ID.
  • Accurate titles, prices, stock, categories, return terms and contact details serve both, so that work is not wasted either way.

SellerSort Team · Updated

ACP and UCP side by side

ACP and UCP compared
AspectACP (OpenAI)UCP (Google)
Where shoppers buyChatGPTAI Mode in Google Search and the Gemini app
Access todayApproved partners, through an application formA waitlist; Google approves each integration
Product dataA feed sent to OpenAI: SFTP snapshots, the API, or bothYour Merchant Center product data, plus a few UCP attributes
Category fieldproduct_category, a path from broad to specificgoogle_product_category, an ID or a path
CheckoutA separately enabled integrationNative checkout on Google, or a cart handed to your site
Products kept outOpenAI’s prohibited product listGoogle’s ineligible list, with native_commerce(checkout_eligibility) left FALSE

Both specifications are new and both companies say they are changing. Check the dates on the source pages before building against either.

The protocols differ in plumbing; the catalog they need is largely the same one.

How the product data gets there

This is the largest practical difference. UCP sits on top of Merchant Center, so a store that already sends a Google feed has most of its product data in place and adds a few attributes, usually in a supplemental file. The UCP setup guide lists them.

ACP needs a feed of its own. OpenAI offers two routes, and recommends using both:

  • File upload. A full snapshot of the catalog pushed by SFTP at least daily, as Parquet or compressed JSONL, CSV or TSV.
  • API. Endpoints for feeds, for product upserts keyed on the product ID, and for promotions, which can only be sent this way.

OpenAI’s suggested rhythm is the daily file for the full catalog and API calls for changes during the day. The ChatGPT product feed guide covers the fields.

Where the category goes in each

Both protocols lean on Google’s product taxonomy, in different shapes. Merchant Center takes google_product_category as a numeric ID or a full path, and Google recommends the ID. OpenAI’s file format takes product_category as a path, and its API describes a category as a value plus the taxonomy it comes from, where google_product_category is one of the named options.

So the same categorization serves both, as long as you keep the ID and the path together. The taxonomy download has them side by side, and the Bulk Mapper returns both for every row.

Products that stay out of both

The two lists are written differently but overlap heavily. OpenAI’s prohibited list names adult content, age-restricted goods such as alcohol, nicotine and gambling, weapons and prescription medicines, among others. Google’s checkout restrictions name alcohol and tobacco, weapons, adult content, healthcare and pharmaceuticals, and add commercial cases of its own: subscriptions, pre-orders, used items, personalized goods and more.

In both cases a correct, specific category is what lets you filter these products out before they reach a checkout they are not allowed in.

Which to prepare for first

Since both are invitation-only, the honest answer is the one you can get into, for the market your customers are in. OpenAI’s standard upload currently targets the US. What does not depend on that decision is the shared groundwork:

  • Titles and descriptions that state what the product is.
  • Prices and stock that match the product page, updated at least daily.
  • A specific product category, with the ID and the path both on hand.
  • A return policy and contact details a shopper can find.

A store with those in order is left with integration work, not data cleanup, when access opens. The guide to what AI shopping assistants read maps the rest.

Questions

What is the difference between ACP and UCP?

ACP is OpenAI’s protocol for selling to shoppers in ChatGPT, fed by a product feed sent to OpenAI. UCP is Google’s protocol for purchases on AI Mode in Search and the Gemini app, built on the product data in Merchant Center.

Can any merchant use ACP or UCP today?

Not yet. OpenAI says onboarding is available to approved partners through an application form, and Google asks merchants to join a waitlist and approves each integration.

Can I use the same product feed for ChatGPT and Google?

Partly. OpenAI accepts a Google-compatible CSV, TSV or TXT file once it confirms that format for your feed, but the fields and delivery differ from Merchant Center. The underlying data, including categories, is the same.

Which category format do ACP and UCP use?

Merchant Center takes google_product_category as an ID or a path, with the ID recommended. OpenAI’s file format takes product_category as a path, and its API lets you name google_product_category as the taxonomy.

Read next