Google AI Mode and Gemini Store Checkout Requirements

Google's new checkout in AI Mode and Gemini lets a shopper buy from the answer without visiting your store. Setting it up is less automatic. Part of the work is a feed job you can schedule. Part of it is a policy or legal decision that no software should make for you.

We judge workflows on this site by what work they take off your plate, so here are Google's requirements sorted that way.

All requirements come from Google's own pages: the UCP Help article, Prepare your Merchant Center account, the checkout integration guide, and the native_commerce attribute.

What the feature is

Merchants who integrate get a checkout ("Buy") button on eligible listings in AI Mode in Google Search and in Gemini. Shoppers pay with Google Pay, using payment methods and shipping details saved in Google Wallet, currently as standard funding PANs (FPANs). You remain the merchant of record.

It runs on the Universal Commerce Protocol (UCP), which Google calls "a new open standard for agentic commerce" across discovery, buying, and post-purchase support. UCP is compatible with A2A, AP2, and MCP. If AP2 is new to you, our explainer on agentic commerce and the Agent Payments Protocol covers the payment authorization side.

Access is limited. Google's Help article covers products with eligibility in the United States, Canada, and Australia. The checkout is "available for select merchants at this time," you register interest through a form, and the early access program rolls out in phases. Filling in that form is a human task, and so is waiting.

The split at a glance

Requirement	Who does it	How
`native_commerce(checkout_eligibility)` flag	Automate	Supplemental feed or Merchant API
`merchant_item_id` mapping	Automate, after a dev confirms the ID logic	Same feed job
Excluding ineligible categories	Automate the rule; a person approves it	Eligibility rule in the feed job
`consumer_notice` upload	Automate the upload; legal writes the text	Same feed job, from an approved table
Return policy (cost, window, link)	Person decides, then enters	Account settings or `returns` / `return_policy_label`
Customer support contact	Person, once	Merchant Center
Merchant Center good standing	Person	Ongoing account management
Interest form	Person	Google's early access form
Google Pay & Wallet Console account	Person	Account setup
PSP onboarding or PCI documentation	Person, with your PSP	Payments team
Native checkout API	Developer builds it, then it runs	REST integration
Sandbox and production validation	Developer, with sign-off	Merchant Center

What a script can own: the feed attributes

This is the most automatable part. Google recommends a supplemental feed for the checkout attributes, which leaves your main product feed untouched. If you push data by API instead, custom attributes go through the Merchant API in accounts.productInputs.insert. The Content API is deprecated. Any feed tooling still built on it is the first thing to replace before adding new attributes.

A nightly job can handle four attributes:

The eligibility flag. native_commerce(checkout_eligibility) is a boolean that defaults to FALSE. The job sets TRUE only for products that pass your eligibility rule and leaves the rest alone. Plain native_commerce still works as an alias, but the full form is easier to read in a diff.
The ID map. Your feed id must match the ID your checkout API expects. Where it doesn't, the job writes merchant_item_id, which takes precedence over id.
The regulatory notice. consumer_notice is mandatory for items with regulatory warnings such as California Prop 65. It has a notice_type (legal_disclaimer, safety_warning, or prop_65) and a notice_message of up to 1,000 characters. The job can attach it from a table keyed by SKU.
The exclusions. These are enforced simply by never setting the flag.

Google excludes subscriptions, installments, personalized goods, refurbished, used and final-sale items, pre-orders, products bundled with services like warranties or installation, freight and other special shipping, gifting, in-store activation, age-restricted goods such as alcohol and tobacco, prohibited content (weapons, adult, healthcare and pharma, counterfeits), services like lessons, classes and travel packages unless on travel rails, rentals, virtual items, and software-dependent goods.

Your product system already knows some of these, like whether an item is used or on pre-order. Others need judgment: is this gadget "software-dependent"? A person approves the category mapping once, and the script applies it.

What needs a person: policy and legal

The return policy is a commercial and legal decision before it's a data entry task. Google requires the return cost, the return window, and a link to the full policy. Once those are settled, you enter them at account level or via the returns or return_policy_label attributes (advanced accounts configure them per sub-account).

The consumer_notice text is the same story. A script can upload a Prop 65 warning to a thousand SKUs in seconds. We wouldn't let a model decide which SKUs need one or how the message reads. Legal writes the table. The job reads it.

Customer support info needs at least one of a website URL, email, or phone number; Google uses it for the "Contact Merchant" link on the order confirmation. Keeping the account in good standing, with products approved for free listings, is ongoing human work too.

Payments: people sign, developers wire

You don't need a Google Pay button on your own site, but you do need a Google Pay & Wallet Console account. If your payment service provider isn't integrated with the Google Pay API, the PSP onboards. If you integrate directly without a PSP, you set up encryption and upload PCI compliance documentation. None of this is agent work.

The integration: build once, then it runs

Google's Native checkout integration is a REST API that lets Google manage the checkout flow with your servers. Google renders the buyer UI for now, "with future plans to support more agentic experiences."

Authentication to your endpoint is optional. Google supports an API key in the X-API-Key header or OAuth 2.0, and recommends OAuth. Test in the sandbox, then validate capabilities in production in Merchant Center. After validation, the checkout exchange is API traffic between Google and your servers. That's the part that runs itself.

Three checkouts, three projects

The existing checkout button sends the shopper to your site to buy. UCP checkout keeps the transaction on Google's surfaces while you remain merchant of record. Google's separate agentic checkout feature buys on the customer's behalf on your own website, at the customer's direction. UCP standardizes how the agent and your backend exchange information, via API, MCP, or A2A.

The step no feed job covers

Every requirement above makes a product buyable from an AI answer. None of it gets the product into the answer, and your checkout integration won't tell you what the assistants decided to recommend.

Apply our usual test (monitor, recommend, or execute) and Promptwatch is the tool we'd put on that gap. It monitors your prompts in Google AI Overviews and AI Mode alongside ChatGPT, Gemini, Perplexity, and other engines, with citation analytics for the pages those answers cite. Agent Analytics crawler logs show whether AI crawlers, GoogleOther included, fetch your product and policy pages. Visitor analytics with conversion tracking measure shoppers who click through to your site. ChatGPT Shopping tracking covers the ChatGPT side. It recommends through Unified Actions, an AI-generated to-do list, and content gap analysis. It executes through Content Agents that plan, write, and publish GEO content to Webflow, Framer, or WordPress, with a review inbox.

Be exact about the boundary. Promptwatch doesn't track AI Mode product cards, Gemini shopping listings, or UCP checkouts, and it doesn't edit your Merchant Center feed. Orders placed through UCP checkout complete on Google and run through your own checkout API, so they won't show up as site visits.

Essential is $95/mo with 50 prompts. Professional at $245/mo adds crawler logs and 5 AEO articles. ChatGPT Shopping tracking and the automated Content Agent are on Business at $579/mo. Compare it with the rest of our agentic tool rankings, or start at promptwatch.com.