ACP & UCP Product Feeds: How Sellers Prepare Data for Agentic Checkout
Agentic Commerce · 8 min read · Updated Sep 2026
Two protocols now define how AI agents complete purchases on behalf of shoppers: OpenAI's Agentic Commerce Protocol (ACP) — built with Stripe, powering instant checkout inside ChatGPT — and Google's Universal Commerce Protocol (UCP), which lets agents transact across Google surfaces with human confirmation steps. Together they mark the shift from AI recommends to AI buys.
For sellers, both protocols reduce to one asset: your product data. An agent cannot walk your store, ask a clerk, or squint at lifestyle photos. It reads your feed. This guide covers what each protocol requires, where they agree, and a practical readiness checklist you can act on this week.
ACP and UCP in one table
ACP (OpenAI + Stripe)
UCP (Google)
Where it runs
ChatGPT instant checkout; merchant-of-record via agentic checkout APIs
Google surfaces (Search, Gemini, assistant experiences)
Who fulfills
The merchant — the agent hands the order to your existing flow
The merchant — agent orchestrates, human confirms key steps
Clean, complete, current product data + reliable fulfillment promises
Same
Note the pattern: the protocols differ in plumbing, not in what they demand from your catalog. Whichever wins, the same work makes you ready for both — which is exactly what makes feed readiness the highest-leverage project of the year.
What agents actually read from your feed
Agentic checkout demands a stricter subset of your product data than a shopping ad ever did:
Identity fields: accurate titles, brand, GTIN/identifier, category mapping. Mismatches here get products excluded from agent surfaces entirely.
Structured attributes: size, material, compatibility, variant relationships. Agents compare products attribute-by-attribute; missing attributes mean losing comparisons you would have won.
Pricing & availability, live: agents check out against real-time price and stock. Stale feeds produce failed orders — and failed orders teach agents to avoid you.
Policies: shipping cost/speed, returns, warranty. Buying agents quote these to shoppers before checkout; blanks or conflicts suppress conversion.
Compliance cleanliness: restricted claims and unverifiable certifications are filtered by the platforms behind the protocols before shoppers see anything.
The uncomfortable truth: most catalogs fail 2-3 of these quietly. Product data that was "good enough" for ads — the occasional missing attribute, the variant with conflicting sizes, the return policy that lives only in FAQ copy — becomes a hard failure when a machine is the buyer.
The feed readiness checklist
Audit structured completeness per SKU. Not your top 10 — all revenue-relevant SKUs. Flag every missing attribute, every "see image" dependency.
Resolve spec conflicts. Where page copy, attributes, and variants disagree, agents notice and shoppers churn. This is the most common silent failure.
Run compliance pre-checks on claims. Before your catalog enters an agent-mediated channel, scrub restricted language and unverifiable certifications — enforcement is automated upstream of shoppers.
Tighten the pricing/stock sync loop. Whatever your feed cadence is, agent checkout punishes staleness more than ads ever did.
Make policies machine-quotable. Shipping, returns, warranty in structured fields — not PDFs, not image alt text, not "contact us."
Where Amazon sellers fit in
ACP/UCP are checkout protocols for open-web and Google/OpenAI surfaces — but Amazon sellers should not tune out. First, the same Agent-Ready data quality determines whether Amazon's own agentic features (Rufus recommendations, Scheduled Actions auto-reorder, Buy For Me) pick your product. Second, many brands run Amazon and a D2C store; the feed work you do for one is reusable for the other. The standard is converging: one Agent-Ready listing standard, everywhere agents shop.
How ListingGood helps
ListingGood checks product listings for exactly the dimensions agentic commerce punishes: machine readability, spec consistency, compliance red flags, and recommendation strength. Run the free Agent-Ready & AI Recommendation Check on any listing, or wire our MCP server into your AI assistant and let the agent run the checks itself — see the developer page.
Frequently asked questions
Q: What is the Agentic Commerce Protocol (ACP)? A: A protocol developed by OpenAI and Stripe that lets AI agents complete purchases — most visibly instant checkout inside ChatGPT — while the merchant keeps fulfillment and the customer relationship.
Q: What is Google's UCP? A: The Universal Commerce Protocol, Google's standard for agentic transactions across Google surfaces (Search, Gemini, assistant experiences). It orchestrates the purchase with human confirmation steps, with merchants fulfilling orders.
Q: Do ACP and UCP require different product feeds? A: No — both depend on the same fundamentals: accurate identity fields, structured attributes, live pricing and availability, quotable policies, and compliance-clean claims. Prepare one clean feed and you are ready for both.
Q: I sell on Amazon. Does ACP/UCP affect me? A: Yes, indirectly but strongly. Amazon's own agentic features (Rufus, Scheduled Actions, Buy For Me) reward the same data quality, and if you run a D2C store, agentic checkout readiness applies to it directly.
Q: What's the first step to become feed-ready? A: Audit structured completeness and spec consistency across all revenue-relevant SKUs, then scrub restricted claims. ListingGood's free check at /scan automates exactly this diagnosis.
Q: Does the agent handle payments and fraud? A: Payment processing is handled by the protocol layer (e.g., Stripe for ACP). Your exposure remains what it always was: product data accuracy, fulfillment reliability, and policy honesty.