To get your store into Google's Universal Cart, you meet Google's Merchant Center baseline, enrich and structure your catalog so an agent can describe and buy your products, choose a checkout path, and then measure whether agents actually surface you. The Universal Cart runs on the Universal Commerce Protocol (UCP), Google's open standard for the agentic shopping journey; being in it means an agent can discover, buy, and track an order from your store inside Search or the Gemini app without the shopper ever loading your site.
This is the practical path, in order. It assumes you sell online and want to be transactable by AI agents on Google's surfaces; the catalog steps apply to any agentic surface, and the surface-specific steps are called out as such. Specifics are dated to mid-2026 and will move — the durable work is the catalog groundwork.
Step 1 — Meet the Merchant Center baseline
Google's stated gate for the Universal Cart is an active Merchant Center account with checkout-eligible products. In practice that means each product carries accurate price and availability, valid identifiers, and the shipping and returns data a real purchase needs — the same data quality Google Shopping has always required, now as the entry ticket for agent checkout.
Start by auditing feed health: resolve disapprovals, fix missing GTINs or attributes, and confirm price and stock match your site. An agent cannot complete a purchase against a product it is not allowed to sell.
Step 2 — Enrich the catalog so an agent can describe you
Eligibility gets you in the door; richness gets you chosen. AI agents field long, conversational queries and can only recommend what your product data actually says. This is the same enrichment covered in depth in preparing your catalog for AI agents — do that work here rather than repeating it:
- Conversational context — what a product pairs with, who it suits, how to use and care for it.
- Social proof as data — review themes exposed as crawlable text, not a star widget.
- Objection handling — product-level FAQs answering the questions that decide a sale.
Generate this from your own reviews, search logs, and support tickets — never from a general-purpose model's guesses, which invent confident, wrong detail you then spend a year correcting.
Step 3 — Expose it as structured, parseable data
UCP standardises primitives — product discovery, checkout, discounts, fulfilment, order management, identity linking — but it can only act on data it can read. Make sure every product fact lives in server-rendered HTML with structured data labelling what it is, not locked in images, click-to-reveal tabs, or JavaScript an agent will not execute.
This is where being agent-buyable and being AI-visible converge: the clean, structured catalog that lets UCP transact is the same asset that lets answer engines cite and recommend you.
Step 4 — Choose your checkout path
The Universal Cart offers two ways to complete a purchase, and the merchant stays the merchant of record in both:
- Native checkout — the shopper pays with Google Pay in a few taps, inside the Google surface.
- Transfer to your site — the cart moves to your own checkout to complete.
Native checkout removes friction but keeps the shopper in Google's surface; transferring keeps them in your funnel but adds a step. Pick per your margins, your data needs, and how much of the experience you want to own — this is a genuine strategic choice, not a default.
Step 5 — Layer payment authorisation with AP2
For agent-initiated payments — where the agent, not the person, pushes the button — UCP is designed to work with AP2, Google's Agent Payments Protocol, which attaches a cryptographically signed proof that the human authorised the purchase. Google said it would bring AP2-backed payments to these surfaces starting with Gemini. You largely accept this rather than build it, but knowing UCP hands payment off to AP2 tells you where the trust boundary sits: UCP runs the journey, AP2 vouches for the money.
Step 6 — Measure whether agents actually surface you
Being transactable is not the same as being chosen. Once your catalog is UCP-ready, the open question is whether agents and answer engines actually discover, describe, and recommend your products — and whether they do it accurately. Track which queries surface you, how you are described, and whether your own pages become the cited source, across every engine and over time.
That feedback loop — did being buyable translate into being bought? — is exactly what Buffy Intel is built to report. For a fuller side-by-side of the standards underneath all this, see ACP vs AP2 vs x402. Questions: [email protected].