What is agentic commerce, and what does it mean for online stores?

Agentic commerce is shopping in which an AI agent acts for the buyer. The shopper describes what they want, and the agent searches for products, compares them and, once the shopper approves, places the order with a store. The store still sells, ships and handles returns. The agent decides which stores the shopper gets to see.

Facts are linked to the source we read, and the status of each agent and protocol is as of October 25, 2026. Anything marked “our analysis” is ours.

What is agentic commerce?

Agentic commerce is any purchase in which software with some autonomy does part of the shopping. “Agentic” means the AI takes actions, not just answers questions. A chatbot that lists three jackets is giving advice. An agent that checks which of the three is in stock in medium, works out the delivered price and puts one in a cart is acting.

Most purchases today sit in between: the agent researches and shortlists, and a person approves the order. Fully delegated buying, such as “reorder coffee when it runs low,” is possible under some protocols but rare.

It overlaps with conversational commerce, where a chat helps the shopper decide but the shopper still buys. In agentic commerce the agent can carry the purchase forward itself.

How is agentic commerce different from ordinary ecommerce?

The products and the stores are the same. What changes is who reads your pages and who makes the shortlist.

The biggest shift is the shortlist. An agent presents a handful of options, not ten blue links. We call that shortlist the candidate set. If your product is not in it, the shopper never hears about you.

How does an AI agent buy something?

The details vary by agent, but most purchases follow the same six steps.

1. The request. The shopper asks for something specific: “a waterproof running jacket under $200 that arrives by Friday.”
2. The search. The agent pulls candidates from web search, product feeds and partner catalogs.
3. The check. It reads each candidate for price, stock, delivery date, returns and reviews. Products it cannot verify tend to drop out.
4. The shortlist. It presents a few options with reasons, often as product cards.
5. The approval. The shopper picks one and approves a summary of the item, the total and the delivery. In every flow covered below, a person confirms before money moves.
6. The checkout. A hand-off to your checkout, a checkout inside the agent through a protocol, or a browser the agent drives through your checkout. The order lands in your admin either way.

Steps two and three are where stores win or lose, and where your feeds, structured data and page text do the work. Our guide to what AI shopping agents actually read covers which of those matter most.

Which AI agents can buy things today?

Several, and each takes a different route to the checkout. The table lists the consumer agents whose purchase flow we could confirm from a primary source or named reporting.

The routes keep changing: OpenAI built an in-chat checkout and stepped back from it within six months. On Shopify, Agentic Storefronts connect merchants to ChatGPT, AI Mode and Gemini, Copilot and Meta surfaces such as Muse from one place in the admin. Shopify's help center says the feature “is active by default for eligible stores.” Many stores are selling through agents without having decided to.

What are ACP, UCP and AP2?

They are open standards that let agents, stores and payment providers talk to each other without a custom integration for every pair. Each covers a different part of the purchase.

Coverage often frames ACP and UCP as rivals. They overlap, but at different scopes: ACP grew out of checkout inside ChatGPT, while UCP describes the whole journey and works with AP2 for payments. Visa says its protocol is meant to complement ACP. A store may support several, usually through its platform rather than its own code.

Do you need to build any of this yourself?

Usually not. Our analysis: for most stores, the protocols arrive through the platform and the payment provider, not through your developers.

• On Shopify, Agentic Storefronts handle the connection to the main agents. Your job is to check which channels are on, and that the product data they send is complete.
• For Google's surfaces, taking part in Google's UCP checkout requires an active Merchant Center account and products marked as eligible for checkout.
• For ChatGPT, what matters most today is being found and recommended, which comes down to crawler access, feeds and readable product pages. Our ChatGPT checklist covers it step by step.
• For browser agents like Muse, there is no integration at all. Your live product page and checkout are the interface, so they need to work for a visitor that reads text and fills in forms. We explain what such an agent sees in why Muse reads your site like a customer.

Custom work makes sense for large catalogs and custom platforms. Even then, the protocol is the easy part. The hard part is product data clean enough for a machine to act on.

What changes for a store?

Four things, all of which you can check this month.

• The agent reads your data before any person reads your copy. A size, delivery date or returns window that only appears in an image or a pop-up is one many agents will miss.
• The sale is decided on a summary card. “$184 delivered, arrives Thursday, free 30-day returns” is easy to approve. “Shipping calculated at checkout” is a reason to pick someone else.
• Your bot rules become sales rules. An agent session can look like a bot: no history, a data-center address, a card nobody has seen. Blocking it turns away an order.
• Your reports will undercount agent sales. Some agent visits arrive with no referrer, and an in-agent checkout may never load your site. Our guide to tracking AI traffic in GA4 covers what you can measure.

What should a store do first?

Five steps, in this order. Each one also helps human shoppers.

1. Check that an agent can read your best seller. Run the free check on one product page. It reports what an agent that does not run scripts can see: price, availability, structured data and bot access.
2. Find out which agent channels you are already in. On Shopify, open Sales channels > Agentic in your admin. On Google, check Merchant Center. Write down what is on and who switched it on.
3. Put the offer on the page as text. Delivered price, delivery date, stock by variant and the returns policy, in the HTML and in your Product schema.
4. Ask your bot and fraud providers, in writing, how they treat AI agent sessions and single-use cards, and whether they verify signed agent requests.
5. Start measuring. Set up AI traffic reporting, and ask a fixed set of shopping questions in each agent every month to see whether you are recommended. That is what the Track part of our services does.

What do we still not know?

How much is bought this way: none of these agents publishes order volumes, so we have left estimates out. How agents rank stores, whether fees or partnerships affect what they recommend, and which checkout route will win. OpenAI's step back from Instant Checkout shows the routes are still being tried out.

The direction is settled. Agents favor what they can read, verify and buy without friction, and so do people.

Written by Harry Harrison, founder of Recommended by Agents, under our editorial standards. Found an error? Tell us. Published October 25, 2026. OpenAI, ChatGPT, Stripe, Google, Gemini, Shopify, Microsoft, Copilot, Meta, Visa, Mastercard, Cloudflare and other names are trademarks of their owners, and we have no affiliation with them.