Is Your Online Store Ready for AI Shopping?

Open the sitemap of almost any large Shopify store and you may find an entry that was not there a year ago: a small file called agents.md. We looked at two well-known stores, Allbirds and Brooklinen, on 20 September 2026. Both serve one, and the two files are almost identical apart from the store name. That points to the platform generating them, not the owners writing them. The file tells AI shopping assistants how to search the store's products, fill a cart and start a checkout.

That changes what "getting your store ready for AI search" means. A lot of the technical plumbing is now handled by the platform, at least on Shopify. So the useful question is no longer "how do I get my products into the system?" It is "what is left for me to do that the platform cannot do for me?"

This post answers that. It covers what Shopify now does for you, how to check whether your store has the same, and the two things no platform can write on your behalf: complete product facts, and the facts about the company behind the shop.

What Shopify Now Does for You

Three pieces are worth knowing by name, because you will see them in headlines and vendor emails.

Agentic Storefronts. Shopify describes it as "a sales channel in the Shopify admin that connects a merchant's products to AI shopping platforms." The channels it lists include ChatGPT, Microsoft Copilot, and AI Mode in Google Search and the Gemini app. Shopify says it is switched on automatically for eligible merchants, and that you can choose which AI partners receive your product data or let Shopify send it to current and future ones.

Shopify Catalog. This is the shared product library behind that channel. Shopify calls it "the structured data layer that makes products readable to AI agents." In plain words: your products are copied into a standard format so that an AI assistant does not have to guess what your page layout means.

The Universal Commerce Protocol, or UCP. Shopify says it co-developed this with Google, and describes it as "an open standard that defines how AI agents transact with merchants," covering the cart, the checkout, payment and after-sale steps. Think of it as a shared set of rules, so an assistant can shop in any store that speaks the same language instead of learning each store separately. The protocol's own site, ucp.dev, lists Google, Shopify, Amazon, Walmart and Stripe among the companies building it.

The agents.md file is the front door to all this. It is written for machines. It points an assistant to the store's address for the protocol, to the product search, and to the sitemap.

Check what your own store serves

This takes two minutes. Add each of these to your store's web address and open it:

1. /agents.md. On the Shopify stores we checked, this returns a plain text page headed "Agent Instructions" with your store's name.
2. /.well-known/ucp. On Allbirds this returns a small block of data listing which versions of the protocol the store supports. If you get an error page, the store does not publish it.
3. /sitemap.xml. On the two Shopify stores we opened, the sitemap lists an extra "agentic discovery" section next to the usual products, collections, pages and blogs.

If you are not on Shopify

We only checked Shopify, so do not assume your platform does the same. A large brand's own storefront we tried, Gymshark, returned a "not found" page for agents.md. That tells you nothing bad about the brand. It tells you the file is something a platform chooses to produce, not something every store gets.

There are two honest routes if your store sits elsewhere. One is to check whether your platform has an equivalent (open the same three addresses, and look in your platform's settings for anything about AI channels or product feeds). The other is to send your products somewhere yourself. Shopify says its Agentic Plan "lets businesses on other ecommerce platforms sync products to Shopify Catalog" without moving platforms. OpenAI also has its own product feed, but access is currently limited to approved partners. Our post on ChatGPT Shopping and product data walks through that route and what your page markup is and is not doing.

What That File Says, and What It Leaves Out

Here is the part that is easy to miss. We read the whole agents.md file on both stores. It is short, and the two are almost word for word the same apart from the store name. It covers four things:

• How an assistant that shops for a person should behave (for example, checkout needs the buyer's approval).
• The steps for using the protocol: look up products, make a cart, start a checkout.
• Where to read products without logging in.
• A "Store Policies" list with three links: the privacy policy, the terms of service and the refund policy.

Now hold that up against the questions a real shopper asks an assistant before spending money:

What the shopper asks	Does the file answer it?
What do you sell?	Yes, through the product search it points to
Can I buy it through you?	Yes, through the checkout steps
What is your refund policy?	It links to the page, but does not say
Do you ship to my country, and how fast?	No shipping page is listed
Who runs this company?	No
Where are you based?	No
How do I reach a real person?	No

That is not a criticism of Shopify. A file built for every store cannot know your address or your founder's name. But it shows where the automatic part stops. The platform moves your products to the assistant. It cannot tell the assistant who you are, and it cannot make a thin product description into a complete one.

Shopify says as much in its own words. Its guide states that AI agents "read structured data" such as titles, descriptions, images, pricing, inventory and shipping speeds, and that "the quality and completeness of that data determines whether a product surfaces in a conversation or gets passed over." In the same guide, its answer to "do I need to change my product data?" is no, because Catalog "automatically structures, cleans, and syndicates" it. Both statements can be true. Tidying what you wrote is not the same as adding what you never wrote.

So the rest of the work sits in two places.

Part One: Product Facts Only You Can Supply

A platform can copy a product description perfectly and still pass along a description that answers nothing. "Cozy and warm. A perfect gift." is copied faithfully, and it tells an assistant nothing it can use to answer "is this blanket wool?" or "will it fit a king bed?"

The "what would I still have to ask?" test

Pick one of your best-selling products. Copy only its title and description into a plain text file. Leave out the pictures, the size chart and anything that appears after you click a tab.

Paste that text into an AI assistant and ask: "I am about to buy this. List every question I would still need answered before I could decide. Do not guess the answers."

The list that comes back is your gap list. For a blanket it might be: what is it made of, how big is it, can it go in a washing machine, what is included. Each question is a fact you know (from the label, the supplier sheet, the photo of the size chart) that a machine reading your page cannot reach. It is more useful than any general checklist, because it comes from your own product.

Three habits cause most of the gaps:

• The size chart is a picture. Shoppers read it without thinking. A machine reading the page text gets a blank where the measurements should be.
• The fact is only inside a tab or pop-up. Some themes load that content after you click. If it is not in the page when it first arrives, an assistant that reads the raw page may never see it.
• The description is written for a person standing in front of the shelf. "Perfect for everyday" means something to a shopper looking at the photo. It carries no facts for anyone else.

Fix the template, not the thousand pages

A store with 2,000 products does not have 2,000 problems. It has one product page template and a set of product fields. If your description box is where every fact goes and the box is empty of facts, you have one habit to change. If your theme hides a "Materials" field, you have one setting to change.

The efficient order is:

1. Decide which four or five facts every product in your range needs (material, size, care, what is included, who it is for).
2. Make sure your product page template shows each one as plain text, near the top or in the main description.
3. Fill the gaps for your top 20 products first, then work outward.
4. Keep the facts as a repeatable pattern, so a new product cannot go live without them.

On Shopify, extra facts usually live in product fields you define yourself; our Shopify markup guide explains where to put them and how to get them into the page code as well. The markup itself is a separate job from writing the facts, and we do not repeat it here.

Part Two: The Company Behind the Shop

Here the automatic file has the biggest hole. A shopper asking an assistant "is this a real company, and can I trust them with my card?" is asking about you, not your product.

Our free readiness check tests this with six fixed questions: what the company does, where it is, which areas it serves, what it offers, who runs it, and how to contact it. The check gives an AI model nothing but your own site's text and sees whether it can answer. The Answer pillar guide explains the method in full, so we will not repeat it. What matters for a store is the pattern of results.

A store homepage is a banner, a grid of products and a newsletter box. That answers "what do you do" and "what do you offer" without any help. The other four questions depend entirely on whether you have written the answer down as text somewhere a reader of the page will see it.

What to put in writing

• A line on the homepage that names the company and where it is. A footer counts, but only as text. An address that exists only on a separate page nobody links to does not.
• An About page with real facts. Who started it, when, where the products are made or shipped from. Not a mission statement.
• A shipping page that states your real terms. Countries you ship to, how long it takes, what it costs. Shopify's agent file lists a refund policy link but no shipping page, so this one is yours to write and to make easy to find.
• A returns page in plain sentences. The window, the condition, who pays. Do not rely on the fine print inside the legal template alone.
• A contact page with a way to reach a person. An email or phone number as text, not a form alone.

Then connect it to the outside world. The profiles you have elsewhere (your social accounts, your business listing, any marketplace page) help a machine confirm that the company on your site is the same one it reads about elsewhere. Our guide to building your brand's entity profile covers how to set that up once and reuse it.

Part Three: Can They Get to the Pages at All?

Everything above assumes an assistant can open your pages. Two store problems are worth a quick check: a security service that blocks automated visitors, and prices that only appear after the page finishes loading. Both affect every kind of site, so our crawler access guide covers them properly. If your agents.md opens fine for you but you have never checked whether a crawler can reach your pages, start there.

A Worked Example

This is a made-up store, used to show the steps. A small shop sells wool blankets on Shopify.

Before. The platform is doing its job: the products are in Shopify Catalog and the store serves an agents.md file. But the product descriptions read "Cozy and warm. A perfect gift for anyone." The size chart is a picture. The homepage is a banner, a product grid and a newsletter form, with the address only on a contact page that the menu does not link to. There is no shipping page, only a line in the checkout.

The "what would I still have to ask?" test on the best-selling blanket comes back with the material, the size, the washing instructions and what is in the box. Asked from the homepage text alone, "who runs this company and where are they?" gets no answer either.

After. The owner makes five changes, in this order:

1. The product template gets four fixed facts near the top of the description: fibre content, size in centimetres and inches, washing instructions, and what is in the box.
2. The size chart picture is repeated as a small text table under it.
3. The 20 best-selling products are filled in. The rest follow over the next month.
4. The footer gets the company name, city, email and phone as text.
5. A short About page names the founders, and a real shipping page lists the countries and delivery times.

Running the same tests again, the assistant now answers the material, the size and the care from the product text, and it can say where the company is and how to contact it. Nothing in that list needed a developer, a new app or a change to the protocol the platform handles. None of it guarantees an assistant will recommend the blankets. It removes the reasons an assistant would have to skip them.

Common Mistakes

Assuming "automatic" means finished. The platform handled the delivery of your products. It did not check that the products said anything useful.

Never looking at what you are sending. Shopify says you can choose which AI partners receive your product data. If you have never opened that setting, you are relying on a default you have not read.

Fixing product by product. If every product has the same gap, the cause is the template or your writing habit, and one change fixes them all.

Putting facts only in pictures. Size charts and ingredient panels are invisible to a reader that only takes in text.

Treating the file as your pitch. The agents.md file is written for machines and, going by how alike the two we read are, generated by the platform. Your story, your terms and your contact details belong on your own pages, where people and machines both read them.

Assuming every platform matches Shopify. We verified this on Shopify only. Check your own platform rather than trusting a general claim.

How to Check Your Store This Week

1. Open the three addresses (/agents.md, /.well-known/ucp, /sitemap.xml) and note which ones work.
2. Run the "what would I still have to ask?" test on your best-selling product, using only its text.
3. Ask the six company questions using only your homepage text, or run the free check below to do it for you.
4. Find your shipping, returns and contact details from any product page in two clicks. If you cannot, an assistant probably cannot either.
5. List the gaps and fix the template first. Then fill in the top products.