AI assistants like ChatGPT, Gemini and Copilot can now find products for shoppers and, in some cases, help them buy. For that to work, the assistant and your store need an agreed way to exchange information, such as what you sell and how an order gets placed. These agreed rules are called protocols.
For most ecommerce teams, the practical question is which AI assistants you want your products to show up in, and what product data each one needs. This guide covers both, as of October 2026.
What are ACP, UCP, MCP and AP2?
ACP, UCP, MCP and AP2 are the four protocols you'll hear about most. They're often compared as if you have to pick one, but they do different jobs:
- ACP and UCP let AI assistants see your products and, where it's switched on, check out. ACP is used by ChatGPT. UCP is used by Google, Microsoft and Shopify.
- MCP is a general way for AI assistants to connect to other software. ACP and UCP can both use it.
- AP2 is about payments. It gives you and your payment provider proof that the shopper approved what an AI assistant is buying for them.
| ACP | UCP | MCP | AP2 | |
|---|---|---|---|---|
| Full name | Agentic Commerce Protocol | Universal Commerce Protocol | Model Context Protocol | Agent Payments Protocol |
| Who's behind it | OpenAI and Stripe | Google, Shopify and Stripe, with Amazon, Meta, Microsoft, Salesforce and others involved | Created by Anthropic, now looked after by the Linux Foundation | Created by Google, now looked after by the FIDO Alliance |
| Where you'll see it | ChatGPT, mainly for finding products | Google AI Mode and Gemini, Microsoft Copilot, Shopify | Behind the scenes in ACP, UCP and most AI assistants | Inside some UCP checkouts, still early |
| Product data it needs from you | A product feed | Your catalog, plus a Google Merchant Center feed for Google | None of its own | None directly |
With all of them, you stay the seller. Payments and returns go through you, just as they do on your own site.
What is the difference between ACP and UCP?
Both let AI assistants find your products and, where it's switched on, buy them. The main difference is which assistants use them.
ACP comes from OpenAI and Stripe and launched in September 2025. It's how ChatGPT gets product information from stores. You send OpenAI a product feed, which is a file or regular data upload listing your products and their details. Today it's mostly used to get products found in ChatGPT.
UCP launched in January 2026. Google leads it, alongside Shopify and Stripe. It covers more of the shopping journey: searching your catalog, adding to cart, linking a shopper's account, checking out and tracking orders. It's used by Google's AI Mode and Gemini, Microsoft Copilot and Shopify. Our guide to Google UCP covers its launch in more detail.
You don't have to choose between them. They're used by different AI assistants, so many stores will end up using both.
What happened to ACP after ChatGPT Instant Checkout?
When ACP launched in September 2025, it powered Instant Checkout, which let shoppers buy from Etsy and Shopify stores without leaving ChatGPT. In March 2026, OpenAI changed course. Shoppers now finish their purchase on the store's own website, and ACP is used mainly to help ChatGPT find and recommend products.
ACP is still being updated. In OpenAI's current product feed rules, your products can appear in ChatGPT search by default, and checkout inside ChatGPT stays off unless it's switched on for you separately. For how Instant Checkout worked when it launched, see our post on ChatGPT Instant Checkout.
How do MCP and AP2 fit in?
MCP connects AI assistants to other software. The Model Context Protocol isn't about shopping on its own. It's a standard way for an AI assistant to connect to a tool or a data source, whether that's a database or a store's catalog. ACP and UCP can both run through it. Shopify is a good example: its AI tools use MCP to let AI assistants search Shopify stores, following UCP's rules.
AP2 handles trust in payments. When an AI assistant buys something for a shopper, the store and the payment provider need to know the shopper really approved it. The Agent Payments Protocol creates a signed record of what the shopper agreed to buy, so the payment can be checked against it. It can be added to a UCP checkout. Most stores won't set it up themselves. It will come through their payment provider or commerce platform.
Which agentic commerce protocol do you need?
Start from the AI assistants where you want your products to show up.
| Where shoppers are | Protocol behind it | What you need to do |
|---|---|---|
| ChatGPT | ACP | Apply through OpenAI's merchant form. It's open to approved partners for now. |
| Google AI Mode and Gemini | UCP | Products in your Google Merchant Center feed can already appear in AI Mode and Gemini. To let shoppers buy without leaving Google, join Google's waitlist. |
| Microsoft Copilot | UCP | Set up Copilot Checkout through Microsoft Merchant Center. It's currently a pilot for select US merchants, and Microsoft's guide covers who can join. |
| Shopify stores | UCP, through MCP | Shopify makes your catalog available to AI assistants for you, with controls in your Shopify admin. Shopify's agentic storefronts guide covers the settings. |
| Other AI assistants | UCP or ACP, often through MCP | Ask your commerce platform what it supports, or have your developers set up UCP. |
| Payments by AI assistants | AP2 and card network programs | Your payment provider handles this. |
Setting these up doesn't guarantee your products will appear. OpenAI says being eligible doesn't mean a product will be shown, and Google is rolling out checkout to a limited group of stores first.
What product data does each protocol need?
Comparing OpenAI's product feed rules with UCP's catalog rules, the core details overlap:
| Product detail | ACP (ChatGPT) | UCP |
|---|---|---|
| An ID for each product and each option, such as each size or color | Required | Required |
| Title | Required | Required |
| Description | Required | Required |
| Price | Required | Required |
| Product page link | Required | Optional |
| Images | Required | Optional |
| Stock status | Required | Optional |
| Barcode number, such as GTIN | Optional | Optional |
| Category | Optional | Optional |
If you already send products to Google Merchant Center, you have most of what ChatGPT asks for. The main addition is a seller name, and OpenAI can accept Google-format feeds once it has confirmed yours. Both also accept more detail than the minimum, and OpenAI's best practices guide says optional details can improve the quality of ChatGPT's answers.
Google's checkout. Google uses the product feed you already send to Google Merchant Center. On top of that, you add your return policy and customer support details to your account, mark which products can be bought through Google, and add any legal or safety notices, such as California's Prop 65 warning. Some products can't be sold this way, and that includes subscriptions, personalized items, pre-orders, used items and age-restricted products.
MCP and AP2. Neither asks for product data of its own. MCP carries whatever ACP or UCP needs. AP2 records the items and total the shopper approved, so the product and price details an AI assistant shows need to match what you charge.
For which product details help products get recommended, see our guide to product data for AI shopping agents.
How does Hypotenuse AI help get product data ready for AI assistants?
Whichever protocols you use, AI assistants can only recommend your products based on the details you give them. Hypotenuse AI's SEO and GEO tools fill in and check those product details, so complete information is ready for every feed. Your ecommerce platform, ERP and payment provider handle price, stock, checkout and payments.
| What AI assistants need | How Hypotenuse AI helps |
|---|---|
| Complete product details | Fills missing specs, dimensions, materials and other attributes from images, PDFs, supplier feeds and web sources, with a source and confidence score on every value. |
| Categories and tags | Organizes specs into structured fields, applies tags and classifies products using your internal taxonomy or marketplace requirements. |
| Shopper language | Adds shopper-facing tags such as use case and occasion alongside spec attributes, including the alternate terms shoppers use. |
| Structured data | Generates structured markup and FAQ schema so products surface in rich results and get cited by AI search. |
| Channel checks | Validates every attribute and content field against brand and channel requirements before export. |
FAQs
Is UCP replacing ACP?
No. They're backed by different companies and used by different AI assistants. UCP covers more of the shopping journey and has wider industry support, while ACP is what ChatGPT uses. Many stores will use both.
What is the difference between UCP and AP2?
UCP covers the shopping and checkout between an AI assistant and a store. AP2 only covers the payment step, recording that the shopper approved the purchase. AP2 can be added to a UCP checkout.
Does UCP use MCP?
It can. MCP is one of the ways UCP connects AI assistants to stores. Shopify, for example, uses MCP to let AI assistants search its stores.
Is ChatGPT Instant Checkout still available?
Not as it first launched. Since March 2026, purchases that start in ChatGPT mostly finish on the store's own website, and ACP is used mainly to help ChatGPT find products. Checkout inside ChatGPT stays off unless it's switched on separately.
Do I need my own MCP server to sell through AI assistants?
Not if your ecommerce platform provides one. Shopify, for example, makes its stores' catalogs available to AI assistants for them.
.png)
.avif)