ACP vs UCP: The AI Shopping Protocols in Plain English

Why there are two standards instead of one

AI shopping protocols exist to solve a single, practical problem: an AI assistant like ChatGPT or Gemini needs a reliable way to read a business's real services, prices, and availability, and in some cases to hand off a purchase or booking request, without guessing or making something up. Two of the biggest companies building AI products — OpenAI and Google — have each backed their own approach to that problem, alongside payment and e-commerce partners. That's not unusual for an early-stage technology category; the same thing happened with video formats, mobile payment standards, and smart-home protocols before one approach consolidated or the major players agreed to support each other's work.

For a small-business owner, the practical takeaway is simple: you don't need to pick a side in this fight to benefit from it. A service that maintains one connection that works across assistants can keep your catalog readable no matter which protocol a given assistant ends up using, which matters a lot more than understanding the standards war itself.

ACP: the OpenAI and Stripe approach

The Agentic Commerce Protocol (ACP) is an open standard originally built by OpenAI and Stripe to let an AI assistant complete a purchase on a shopper's behalf — browsing a catalog, building a cart, and handing off payment — inside a chat conversation instead of redirecting the shopper to a separate website to check out. It's published as an open specification rather than a closed OpenAI-only tool, and other companies, including payment processors beyond Stripe, have joined as contributors or integration partners since it launched.

ACP's early public face was "Instant Checkout" inside ChatGPT, letting a shopper buy a product from a participating merchant without leaving the chat. The specific consumer-facing feature has already changed shape at least once since launch, which is a useful reminder for a business owner: the branded product sitting on top of ACP may be renamed, relaunched, or quietly retired, but the underlying protocol — the plumbing that lets an assistant read a catalog and request a transaction — is the more durable part worth paying attention to.

UCP: the Google-backed approach

The Universal Commerce Protocol (UCP) is Google's open standard for the same underlying problem — letting an AI agent discover a merchant's products and move through checkout — developed with e-commerce platform partners rather than built solo. It's commonly discussed alongside a separate but related Google-backed standard, the Agent Payments Protocol (AP2), which focuses specifically on how an AI agent proves it has a user's authorization to pay, using signed digital "mandates" rather than handing raw payment details to an assistant.

The distinction worth remembering, even without memorizing the acronyms: UCP is aimed at the discovery-and-checkout journey (can an agent find your services and start a transaction), while AP2 is aimed specifically at the payment-authorization step underneath it. ACP, by contrast, bundles discovery and payment handling into one specification. Different companies drew the lines between "catalog," "cart," and "payment" in different places, which is a good part of why these standards don't map onto each other one-to-one.

So what's actually different between ACP and UCP?

Strip away the branding and the practical differences for a small business come down to a few things:

• Who's behind each one. ACP traces back to OpenAI and Stripe; UCP traces back to Google and its e-commerce platform partners. Which assistants and which checkout tools support which protocol generally follows those lines, though cross-support between the camps has started to appear as the space matures.
• What they bundle together. ACP ties catalog discovery and payment handoff into a single flow. UCP separates "can an agent find and understand your offering" from "how does the agent prove it's allowed to pay," leaning on AP2 for the second part.
• How mature each one is. Both are genuinely new. Neither has the kind of universal adoption that something like a credit card network or an email protocol has — both are still in the phase of signing up launch partners and expanding merchant support, not the phase of being a settled default.

None of this is a knock on either protocol. It's just the normal, messy middle of a new technology category, where the practical shape of "how AI shopping works" is still being negotiated between a handful of large companies — and a small business has no leverage to speed that up or pick the winner in advance.

Where MCP fits into this picture

There's a third name worth knowing that isn't a competitor to ACP or UCP so much as a layer underneath both: the Model Context Protocol (MCP), an open standard for how an AI assistant reads structured information and calls tools in the first place. MCP started as an Anthropic project for Claude and has since been adopted broadly enough that it shows up referenced inside the specifications for both ACP and UCP as one of the ways a business's catalog can be exposed to an agent, rather than staying locked to a single company's ecosystem.

That's a useful detail for a small business owner specifically because it means "connect once, read by many" isn't just marketing language — it reflects how these specifications are actually being built. A catalog published through MCP can, in principle, be the thing a checkout-focused protocol like ACP or UCP points to when it needs real prices and service details, rather than each protocol requiring its own completely separate data format from scratch.

What competing standards mean for your business, practically

Here's the part that actually matters if you run a business rather than follow this space for a living: your customers don't all use the same AI assistant. Some ask ChatGPT, some ask Gemini, and a growing number use Claude or another assistant entirely. If your business's catalog is only readable by whichever protocol one specific assistant favors, you're invisible to every customer using a different one — and you won't necessarily know which protocol "won" until long after it would have helped to pick correctly.

That's the core argument for not betting on a single standard yourself. Building a custom integration against one specific protocol is a real technical project — the kind of thing that made sense for a large retailer with an engineering team, and makes much less sense for a business trying to get review requests out and keep its Google Business Profile current. A service that handles the catalog-and-connection work across multiple protocols and assistants removes that decision from your plate entirely: you maintain one structured, accurate catalog of your real services and prices, and the connection layer handles translating it for whichever assistant and whichever protocol a given customer happens to be using. If you want the deeper case for why small businesses specifically should care about this now rather than later, our guide to agentic commerce for small businesses goes into the reasoning in more detail.

Common questions about ACP and UCP

Is ACP or UCP better for small businesses?

Neither is clearly "better" in a way that should drive a small business decision — they solve overlapping versions of the same problem, backed by different companies, and both are still expanding which assistants and merchants they actually support. The more useful question for an owner isn't which protocol to pick, it's how to avoid depending on picking correctly at all.

Do I need to choose one protocol to connect my business to AI?

No. A connection service that reads your catalog once and makes it available across protocols and assistants means you're not choosing between ACP and UCP any more than you'd choose between two different web browsers rendering the same website — the underlying content should work either way.

Will one of these protocols eventually replace the other?

Possibly, but it's not something a small business should wait around for. Standards battles like this one have historically taken years to resolve, and sometimes resolve into coexistence rather than a single winner, with large platforms supporting both. Waiting for clarity means waiting to be visible to AI assistants at all in the meantime.

What happens if I do nothing and wait for a winner?

You stay invisible to AI-assisted shoppers in the meantime, and there's no guarantee the wait ends with a clean answer. Even if one protocol ends up dominant, the businesses that already have an accurate, structured catalog connected will have a head start — there's nothing to "redo" if you started with a connection method that already works across approaches, while there's real catching up to do if you waited and guessed wrong.

The bottom line

ACP and UCP are two different, still-evolving answers to the same question: how should an AI assistant read a business's real offerings and move toward a transaction. Both are backed by major companies, both are actively changing shape, and neither has settled the question of which one — or whether both — will matter most in two years. For a small business, the practical answer isn't research-and-pick. It's connecting your catalog once, through something that keeps working as these standards keep moving, rather than building toward a standard that might not be the one your customers' assistants end up using.