Agentic commerce: how AI agents are rewiring the way the world buys

Agentic commerce is online buying and selling done by AI agents that act for a shopper or a business buyer. Instead of a person browsing a website and clicking buttons, the agent finds products, compares options across stores, and can complete checkout once the person confirms. The merchant still sells the product, takes the payment, and ships the order.

Agentic AI is artificial intelligence that plans and carries out steps toward a goal, instead of only writing text. It can call tools and APIs, check results such as stock or price, and decide the next step. In shopping, that means it can search, compare, fill a cart, and hand the order to the merchant with the buyer approving the purchase.

Generative AI creates content such as text, images, or chat replies. Agentic AI takes actions: it sets a plan, calls APIs, checks the outcome, and moves to the next step. A generative model can write a product description. An agentic system can find the product, check that it is in stock, and pass the order to the store.

An AI agent is one piece of software with a job, a set of tools, and permissions, such as a shopping assistant inside ChatGPT. Agentic AI is the wider approach: models that plan, use tools, and act with limited supervision. One agent or a group of agents working together can be built on that approach.

An AI shopping agent is software that shops for a person. It takes a request such as "waterproof trail shoes for wide feet under $160", reads product feeds and reviews from many stores, ranks the options, and can complete the purchase inside the assistant. OpenAI says ChatGPT ranks multiple sellers of the same product on availability, price, quality, and whether the merchant is the primary seller.

AI commerce is the use of artificial intelligence across selling online: product search, recommendations, pricing, customer support, and catalog feeds. Agentic commerce is the part of AI commerce where software also completes the transaction. A store can use AI commerce tools today without supporting agent checkout, but agent checkout depends on the same clean product data.

In traditional ecommerce a person browses a storefront and clicks to buy, so page design, ads, and merchandising decide a lot. In agentic commerce software reads your product feed, structured data, and checkout endpoints, then picks based on facts it can check: price, stock, shipping, returns, and reviews. Your storefront still matters for people. Your data matters for agents.

It has four layers. A product data layer, such as a feed or Shopify Catalog, that agents read. A protocol layer, such as ACP or UCP, that carries the cart and order between the agent and the store. A payment layer of tokenized credentials from Stripe, Visa, or Mastercard. Your existing order, tax, fraud, and fulfillment systems sit underneath and stay in charge.

It is live but early. OpenAI launched Instant Checkout in ChatGPT on 29 September 2025 for US users buying from US Etsy sellers. Shopify says its Agentic Storefronts channel connects merchants to ChatGPT, Microsoft Copilot, and AI Mode in Google Search and the Gemini app, limited to businesses selling to US buyers. Most shopping still happens on normal storefronts.

OpenAI and Stripe co-developed the Agentic Commerce Protocol, which powers Instant Checkout in ChatGPT. Google announced the Agent Payments Protocol (AP2) with more than 60 organizations. Shopify and Google co-developed the Universal Commerce Protocol. Visa runs Intelligent Commerce with tokenized credentials for agents, and Mastercard runs Agent Pay with Agentic Tokens. PayPal is one of the named AP2 partners.

The Agentic Commerce Protocol (ACP) is an open standard co-developed by Stripe and OpenAI so AI agents and businesses can complete a purchase together. Stripe says it is open source and Apache 2.0 licensed. The merchant stays the merchant of record, accepts or declines each order, and charges the payment through its own provider. It powers Instant Checkout in ChatGPT.

AP2 is an open protocol Google announced on 17 September 2025 for agent-led payments. It works as an extension of the Agent2Agent protocol and the Model Context Protocol. It uses signed "mandates": an Intent Mandate records what the user asked for and their limits, and a Cart Mandate records the exact items and price the user approved. Partners named include Mastercard, PayPal, and American Express.

Agentic tokenization replaces a real card number with a token that an AI agent can use only within set limits. Visa calls these tokenized digital credentials and says only the consumer can tell the agent when to use one. Mastercard calls them Agentic Tokens, built on the tokenization behind contactless and card-on-file payments, and requires agents to be registered and verified first.

Conversational commerce is a person chatting with a brand, usually in a chat widget or messaging app, and still doing the buying themselves. Agentic commerce hands the search, comparison, and checkout steps to an AI agent, with the person confirming the purchase. The agent may talk to many stores in one session, not one brand.

No. Your store stays the source of truth for products, prices, and customer relationships. In both ACP and Shopify Agentic Storefronts the merchant remains the merchant of record and the order lands in your own admin. What changes is that the store now serves two readers: people who browse, and agents that read your feed, structured data, and checkout endpoints.

Start with your product data: complete titles, prices, stock, shipping, return terms, and GTINs in a feed and in Product schema on each page. Next, make sure your checkout can be reached by an agent, through Shopify Agentic Storefronts or an ACP integration with your payment provider. Then check whether AI assistants already name you, and fix the pages they skip.

Measure what you have before you spend. Pull how many orders and sessions already come from ChatGPT, Copilot, Perplexity, and Google AI Mode in your analytics. Run the buyer questions your customers ask through those assistants and count how often you are named. Then price the fixes: feed cleanup, schema, and checkout access. Most of that work also helps normal Google Shopping.

The route depends on your platform. Shopify merchants can turn on the Agentic Storefronts sales channel in the Shopify admin. Stores on other platforms can sync products to Shopify Catalog through the Agentic Plan, which Shopify says has no monthly fee. Stores on Stripe can enable agentic payments for ACP; OpenAI says that can take as little as one line of code.

Check five things. Can you publish a full product feed (OpenAI asks for CSV or JSON, refreshed daily)? Does your stock and price data update fast enough to stay accurate? Can your checkout create and update a session through an API? Does your payment provider support tokenized agent payments? Can your order system accept or decline an order and report status back?

Use the controls the payment rails already provide. Accept only tokenized credentials scoped to a set amount and merchant, never raw card numbers. Keep your own fraud checks on every order, since ACP lets the merchant accept or decline. Tell verified agents apart from scrapers at your CDN. Log each agent order with its approval record so disputes can be traced.

The agent reads the request, pulls matching products from feeds and the web, then ranks them on facts it can check. OpenAI says ChatGPT shopping results are organic and unsponsored, and when several merchants sell the same item it weighs availability, price, quality, and whether the merchant is the primary seller. Shopify lists data quality, availability, and pricing as factors.

Treat each marketplace listing as a feed an agent may read. Keep titles, GTINs, prices, and stock identical across your own store, Amazon, Walmart, and other channels, since an agent comparing sellers will spot a mismatch. Make your own store the most complete source, because agents weigh whether a merchant is the primary seller. Sync inventory so agents never buy what you cannot ship.

Yes, on the factors agents check. OpenAI says ChatGPT product results are unsponsored and ranked on relevance, so ad budget does not buy a place there. A mid-sized brand with complete product data, accurate stock, clear return terms, and real reviews can be picked over a larger retailer whose feed is thin or out of date.

Discovery moves from ranking on a results page to being named in an AI answer. That rewards pages AI engines can read and quote: Product and Offer schema, clear specs, direct answers to buyer questions, and consistent facts across your site and listings. Generative engine optimization (GEO) is the name for that work, and it sits on top of normal SEO.

The shopper approves, the network issues a limited token, and the merchant charges it. OpenAI says ChatGPT payment tokens are authorized only for specific amounts and merchants. Visa lets consumers set spending limits and conditions for agent purchases. Mastercard requires agents to be registered and verified before they can pay. Google AP2 records the approved cart in a signed Cart Mandate.

We clean and complete your product feed, add Product and Offer schema to every product page, connect your store to the agent channels your platform supports, and build back-office agents that keep inventory and orders in sync with your ERP. You see a working demo on your own data before you sign, and each phase has a fixed price quoted in writing after scoping.

Send your website URL. We check your catalog structure, schema coverage, feed quality, and whether AI assistants name your brand for your buyer questions. You get a list of fixes in priority order. You can run the free AI visibility checker first to see how often assistants mention you today.