Agentic Commerce: How to Get Your Store Ready for ChatGPT Shopping and AI Agents

A shopper types into ChatGPT: "I need waterproof trail shoes, size 9, under $120, that can arrive by Friday." The assistant searches, compares a handful of products, checks stock and delivery dates, and offers to buy the pair it recommends without the shopper ever opening a store's website. That's agentic commerce, and in 2026 it has moved from demos to real checkouts.

Whether your products show up in that conversation depends on things you can work on now. Most of them are unglamorous: clean product data, a crawler you haven't accidentally blocked, policies written where a machine can find them.

A plain definition

Agentic commerce is shopping where an AI agent does the research and completes the purchase on the shopper's behalf. The person sets the goal and approves the payment. The agent does the browsing.

Compare that with ordinary AI search, where a chatbot recommends a product and links out. In the agentic version, the agent also needs to place the order, which means it has to talk to your checkout in a structured way. That second step is the new part.

For a store owner, the change is who your page has to convince. A human skims photos and reviews and forgives a vague delivery estimate. An agent working to a price limit and a delivery date has no reason to forgive anything, so it moves on to the next store whose data answers the question.

What changed in 2025 and 2026

Two protocols arrived within about four months of each other.

In September 2025, OpenAI and Stripe released the Agentic Commerce Protocol, or ACP. It's the protocol behind Instant Checkout in ChatGPT, where a shopper can buy from a merchant inside the chat.

Then in January 2026 Google announced the Universal Commerce Protocol, UCP, its own approach to letting agents shop across merchants. We don't know yet which protocol will win, or whether stores will end up supporting both the way many already accept more than one payment method. For planning purposes, assume you'll want to support at least one and design your checkout so adding a second isn't a rewrite.

How an agent reads a store

Your hero banner is wasted on an agent. It reads pages the way a parser does, and the parts it uses are mostly invisible to a human shopper.

Discovery comes first. For ChatGPT's search features, the crawler is OAI-SearchBot, and your robots.txt decides whether it can visit. It's easy to block every AI bot as a precaution and then wonder why your products never appear in ChatGPT answers. Check yours. You can allow OAI-SearchBot for search while keeping other crawlers out, because each one is listed by name.

Once on a product page, the agent looks for structured data. Schema.org Product markup in JSON-LD tells it the name, brand, price, currency, availability and, where you have one, the GTIN. If that markup is missing, the agent has to guess from the visible page, and a guess about price or stock is exactly what an agent buying on someone's behalf will avoid.

A small example. Your page shows a sale price of ₹2,499, but the JSON-LD still says ₹2,999 because the markup is hard-coded in a template nobody updated. A person sees the sale. An agent sees two prices, can't tell which is real, and may skip the product or quote the wrong one.

Product feeds fill in the rest. A complete feed with every variant, accurate stock and current prices lets a platform list your catalogue without crawling every page. Gaps in the feed (missing sizes, a stale sale price) turn into wrong answers.

And the agent needs to know what happens after the order. Shipping costs, delivery times and the returns policy should be on plain, crawlable pages. A returns policy that only exists as an image, or behind a chat widget, is invisible to it.

Speed matters in a plain way. An agent comparing several stores will give up on pages that time out, and a product page that only fills in its price after a slow script runs may look empty to it. Fast, stable pages that put the important content in the first HTML response are the safe bet.

Readiness checklist

If you want to know how to get products on ChatGPT shopping and in front of other AI shopping agents, work through these. They're roughly in order of effort, smallest first.

1. Open your robots.txt and confirm OAI-SearchBot isn't blocked, unless you've decided you want it blocked.
2. Add schema.org Product markup to every product page with name, image, description, brand, price, priceCurrency and availability. Include the GTIN where you have one.
3. Make sure the price and availability in your markup match what the page shows. Mismatches look like errors, and agents are cautious about errors.
4. Give each variant (size, colour) its own identifier and stock status in your data.
5. Publish a complete product feed and keep it in sync with your inventory, ideally from the same database the storefront reads.
6. Write your shipping and returns policies as normal HTML pages with specific numbers: delivery days by region, the return window, who pays return postage.
7. Check that product pages load fast and render their key content without waiting on client-side JavaScript.
8. Use clear, factual product titles. "Men's waterproof trail running shoe, Gore-Tex, black" helps an agent more than a campaign name.
9. Look at your checkout code and ask how hard it would be to accept an order from an outside system through an API. That's the step a protocol like ACP or UCP will need.

None of this is exotic. Items two through seven also help ordinary search and comparison shopping, so the work pays off even if agent traffic stays small for you this year.

Where a custom store has room to move

On a hosted platform, protocol support usually arrives as a platform feature or an app. That's convenient when it ships. When it hasn't shipped, or it handles your edge cases badly (bundles, made-to-order items, region-specific pricing), you wait.

A store you own the code for can take the other route. A developer can read a protocol's specification and integrate it directly in your own checkout code, against your own order and stock logic. The same goes for structured data: you decide what Product markup appears on every page, generated from the same records that drive the visible price, so the two can't drift apart.

To be fair to the platforms, many themes output reasonable Product schema out of the box, and a platform that ships protocol support early can get a small store live faster than a custom build would. The trade is control. We hand-code stores in Next.js, React and TypeScript, with payments through Stripe or Razorpay, so the checkout and the data layer are code you own in your own repository, and changing them is a normal pull request.

What to ignore for now

A lot of advice in this area is guesswork sold as a framework. Some things we'd park:

• llms.txt as a ranking fix. Google has said it doesn't use llms.txt for Search. Adding one does no harm, but don't expect it to put your products in front of shoppers.
• Rewriting all your product copy "for AI". Agents need accurate facts and structured data. Clear writing for humans already does most of the job.
• Paying for a tool that promises guaranteed placement in ChatGPT answers. Nobody outside OpenAI controls that.
• Building separate pages only for bots. Keep one set of pages that serves people and agents with the same information.

Getting your brand cited in AI answers is a related job with its own methods, and it's what our SEO & Performance service works on. For agentic shopping specifically, start with the data and the checkout.

Questions people ask

What is agentic commerce in simple terms?

An AI agent researches products and completes the purchase for a shopper, who approves the payment. The agent talks to the store through structured data and, increasingly, a checkout protocol.

How do I get my products on ChatGPT shopping?

Let OAI-SearchBot crawl your store, add complete Product structured data, keep a full product feed, and publish clear shipping and returns pages. Buying inside ChatGPT through Instant Checkout runs on the Agentic Commerce Protocol, so your checkout also needs to support it.

What's the difference between ACP and UCP?

ACP was released by OpenAI and Stripe in September 2025 and powers Instant Checkout in ChatGPT. UCP was announced by Google in January 2026. Both aim to let AI agents place orders with merchants in a standard way.

Should I block AI crawlers?

That's a business decision. Blocking OAI-SearchBot keeps your pages out of ChatGPT's search features, which also means your products can't be recommended there. You can allow search crawlers and decide on the others case by case.

Do small stores need to bother yet?

The structured data, feed and policy work is worth doing at any size because it helps regular search too. A full protocol integration can wait until you've seen whether your customers shop this way.