Agentic commerce for small shops, in plain English

You may have heard that people will soon shop by asking an AI assistant instead of browsing a website. That idea has a name: agentic commerce. This guide explains what it means, what is actually available and what a small shop can sensibly do about it without losing a month to it.

What it is

In ordinary online shopping, a person visits your site, picks something and pays. In agentic commerce, an AI assistant does some of that work for them. A shopper might ask for a waxed jacket under a hundred pounds in green. The assistant looks through product data from many shops, shows a few options, and in some cases helps buy one.

For that to work, the assistant needs product data it can trust. Your website is built for people, and reading prices and stock off it is guesswork. A product feed gives the assistant the same facts in a fixed shape.

What ACP is

ACP stands for the Agentic Commerce Protocol. Its documentation describes it as a REST-based standard that lets AI agents carry out purchases with online sellers: creating, updating and completing checkouts, while the seller stays in control of its own operations. The seller stays the merchant of record.

OpenAI's documentation uses the same name for its commerce work in ChatGPT, covering agentic checkout, delegated payment and product feeds. The product feed is the first and simplest piece. It is just a file.

What is real today

• OpenAI has a published product feed spec, with required fields and file formats.
• OpenAI's documentation says onboarding product feeds in ChatGPT is currently available to approved partners. Merchants can apply through a form.
• There are two ways to deliver a feed: file upload over SFTP, or an API. Daily snapshots are the minimum they recommend.
• Instant Checkout needs more than a feed. The merchant has to handle checkout and payments on their own systems, with their own payment provider.

What is not on that list: a guaranteed audience, any promise about how many sales it will bring, or a route that works without approval. We do not know how big this will be for small shops, and anyone who claims to is guessing.

Why a small shop might care anyway

• Clean product data helps everywhere. The same feed you prepare for this can feed other channels and cleans up your own catalogue.
• It costs little. If you use Shopify or WooCommerce, the data already exists. Preparing it is an afternoon, not a project.
• If approval comes, you are ready. Merchants with tidy feeds can move faster than those starting from a messy export.
• It forces you to fill gaps you have ignored: policies, returns, brand names, consistent IDs.

What you can do, in order

1. Check your catalogue against the spec and see how much is already fine.
2. Fill the store-wide gaps once: seller name, brand, policy links, returns, country.
3. Fix product-level problems at the source, in your shop admin.
4. Apply as a merchant when you are ready, and expect to wait.
5. Keep the feed fresh so what an assistant sees matches what you sell.

Things to be wary of

Be sceptical of anyone offering guaranteed placement in ChatGPT, or charging for access that OpenAI controls. There is no back door that we know of. Approval, if you get it, comes from OpenAI.

Also be careful about spending big money on speculation. A free check and a tidy feed is a modest, reasonable bet. A large project built on a guess about how shoppers will behave is not.

A fair way to think about it

Treat this like an early channel. Some past channels, such as marketplaces and price comparison sites, were worth the effort for some shops and not for others. You will not know which kind this is until you try, and trying costs little if you keep it small. Prepare the feed, apply, and see what happens, while giving your own site the attention it already needs.

The honest summary is that the people who build these systems are still working out the rules. Prepare lightly, stay flexible, and read the official documentation when something changes.