AI search glossary
Agentic commerce: when AI agents shop and check out for buyers
Agentic commerce is buying and selling in which an AI agent acts for the shopper: it searches, compares, and can start or complete checkout with the merchant, using open protocols such as the Agentic Commerce Protocol (ACP) and the Universal Commerce Protocol (UCP).
Also called: agentic shopping, agentic checkout, Agentic Commerce Protocol, ACP, AI shopping agents
Why it matters for AI search
In agentic commerce the AI does more than cite a store, it chooses which product to put in the cart. That choice rests on structured, current product data: price, availability, variants, shipping and returns. Stores whose catalogs reach the agent through feeds or protocols, with clear facts, are the ones it can recommend and buy from.
How agentic commerce works
A shopper tells an assistant what they want, for example "a waterproof running jacket under $150, delivered by Friday". The agent searches catalogs, compares options, and either sends the shopper to the store or completes the purchase inside the chat. The shopper approves; the merchant fulfils the order.
Three pieces make this possible:
- Product data the agent can read. OpenAI's commerce docs describe product feeds that let ChatGPT index and display products with current price and availability.
- A checkout the agent can call. The Agentic Commerce Protocol, built by Stripe and OpenAI, defines how an agent starts checkout with a merchant. The Universal Commerce Protocol does the same across catalog, cart, checkout and orders, and is co-developed by Google and Shopify among others.
- Payment handled safely. ACP passes payment credentials from buyer to agent through compatible mechanisms such as Stripe's Shared Payment Token; UCP points to an open wallet ecosystem.
Google is moving the same way. At I/O in May 2025 it described agentic checkout in AI Mode, where the user stays in control of the purchase.
What it means for AI visibility
Classic AI search ends with a citation; agentic commerce ends with an order. The agent has to pick one product, so vague or stale data is a direct loss. Missing stock status, unclear sizing or hidden shipping costs give the agent a reason to choose a competitor. It also ties agentic search to revenue: the same agent that researches can now buy.
How to prepare a store
- Clean the catalog. Titles, prices, variants, stock and GTINs should be accurate and current. See product feed optimization for AI.
- Mark up product pages. Use Product schema with offers, availability and reviews that match the visible page.
- Publish policies in text. Shipping, returns and warranty terms should be easy to find and quote.
- Connect to the protocols your platform supports. For Shopify stores, see agentic commerce for Shopify and the Universal Commerce Protocol on Shopify.
- Test with real prompts. Ask ChatGPT and Google AI Mode for your product category and see which stores they choose.
Related ideas
Agents connect to services through standards such as the Model Context Protocol. For how products surface in ChatGPT, see ChatGPT shopping optimization.
Questions
- What is the Agentic Commerce Protocol?
- ACP is an open standard, developed by Stripe and OpenAI under the Apache 2.0 license, for purchase flows between buyers, AI agents and businesses. Merchants implement its checkout specification so compatible agents can start checkout, and payment credentials pass securely to the agent.
- How is UCP different from ACP?
- Both are open standards for agent purchases. The Universal Commerce Protocol is co-developed by Google, Shopify and a long list of retailers and platforms, and covers catalog, cart, checkout, identity linking and orders. ACP is the protocol behind commerce in ChatGPT.
- Who is the merchant of record when an agent buys?
- The store. The ACP site says businesses stay the merchant of record and keep control over which products are sold, how they are shown and how orders are fulfilled.
- What should a store do first?
- Get product data right: accurate titles, prices, stock, variants and policies, in a feed and in Product markup on the page. Agents can only pick from what they can read reliably. Protocol integrations come after the data is clean.