Three Protocols, Three Layers
Most "UCP vs ACP" articles treat them as rivals. They overlap less than that framing suggests. The Universal Commerce Protocol covers the whole shopping journey from discovery to checkout. The Agentic Commerce Protocol started as a checkout standard and, since OpenAI retired Instant Checkout in March 2026, mostly carries product feeds into ChatGPT. The Agent Payments Protocol does neither: it proves that the user actually authorised a payment. A single purchase can touch all three.
At a Glance
| UCP | ACP | AP2 | |
|---|---|---|---|
| Full name | Universal Commerce Protocol | Agentic Commerce Protocol | Agent Payments Protocol |
| Led by | Google, co-developed with Shopify, Etsy, Wayfair, Target, and Walmart | OpenAI and Stripe | |
| Announced | January 11, 2026 (NRF) | September 29, 2025 | September 16, 2025 |
| Layer | Discovery, cart, checkout, post-purchase | Product feeds, checkout, delegated payment | Payment authorisation and proof of consent |
| Main surfaces | Google AI Mode, Gemini, Shopify stores, any self-serve UCP agent | ChatGPT | Any agent; extends A2A and MCP |
| Backers | 20+ endorsers at launch incl. Adyen, American Express, Best Buy, Mastercard, Stripe, Visa; Shopify also lists Amazon, Meta, Microsoft, Salesforce | PayPal supports it for payments; Target, Sephora, Nordstrom, Lowe's, Best Buy, Home Depot, Wayfair feed ChatGPT through it | 60+ partners at launch incl. PayPal, Mastercard, American Express, Adyen, Coinbase, Worldpay |
| How merchants join | Publish a business profile at /.well-known/ucp, or get it through Shopify | Send a product feed to OpenAI; checkout endpoints optional | Usually through your payment provider |
| License and status | Open standard; Shopify made it self-serve June 17, 2026 | Apache 2.0, beta, latest stable spec 2026-04-17 | Open protocol |
UCP: The Full Shopping Journey
UCP gives agents one way to search a store, build a cart, check out, and handle support. Stores publish a business profile, agents publish a platform profile, and the two negotiate which capabilities they share; checkout is required, cart, orders, discounts, and fulfillment are optional. Shopify is the biggest single deployment. Its Spring '26 Edition removed the approval gate, and on August 31 it retired its older Storefront MCP cart tools in favour of UCP. See Shopify's UCP rollout.
ACP: ChatGPT's Product Pipe
ACP was built so ChatGPT could complete purchases. That part did not work: Walmart reported in-chat conversion at about one third of its own site, and only a few dozen merchants at most went live. OpenAI kept the protocol and moved checkout back to merchants. ACP's product feed is now what decides which products ChatGPT can put in front of a shopper. The April 2026 spec added cart, feed, orders, authentication, and MCP support. See the ACP adoption tracker.
AP2: Proof the User Said Yes
AP2 answers one question: did the user authorise this purchase? It does this with a chain of signed mandates. An intent mandate records what the user asked for and the limits they set, such as a price ceiling, and a cart mandate records the specific basket they approved. Payment providers and merchants can check those records before settling. AP2 does not move money itself and is designed to sit above cards, bank transfers, and stablecoins alike.
Which One Should a Merchant Implement?
On Shopify, you already have UCP and ChatGPT syndication by default; your job is product data quality. Large retailers off Shopify increasingly run both: an ACP feed for ChatGPT and a UCP profile for Google and other agents. AP2 usually arrives through your payment provider rather than as a separate project. If you can do only one thing, make your product data complete and consistent, because every one of these protocols passes it straight to the agent.
What It Means for Brand Visibility
Protocols decide whether an agent can see and buy your product. They do not decide whether the agent picks it. That choice still depends on product data, reviews, third-party coverage, and how the model already describes your brand. Support the protocols your channels need, then measure how often agents on each surface actually recommend you.