Shopify Quietly Gave Every Store an llms.txt, an agents.md and an Agentic Sitemap

The first person to publicly notice was a merchant, not Shopify.

On May 11, 2026, a store owner posted in the Shopify community forum that their custom llms.txt redirect had stopped working. They had built a curated file by hand, pointed a URL redirect at it, and one morning the redirect was being ignored. Shopify's own auto-generated file was being served instead. Their words: the curated list "has effectively become obsolete overnight."

No email. No admin banner. The changelog entry explaining how to take the files back did not land until May 28, roughly two and a half weeks later.

If you have never looked at your storefront's /llms.txt, you now have one anyway. So does every other store on the platform.

What actually appeared at the root of your domain

Four things worth knowing about, and one of them matters a lot more than the others.

• /agents.md is now the canonical file. Shopify's changelog is explicit: stores get a default agents.md, and /llms.txt and /llms-full.txt point at the same content by default. Open yours and you will find something closer to an operating manual for a shopping agent than a curated content guide. How to discover the store, how to search products, how to build a cart, how to reach checkout.
• /llms.txt and /llms-full.txt exist and resolve. They are not empty. They are also not yours unless you make them yours.
• /.well-known/ucp is the Universal Commerce Protocol business profile, a JSON document declaring which commerce capabilities your store supports. Shopify documents it at {shop}.myshopify.com/.well-known/ucp in its agent profile docs. This is the one an actual purchasing agent reads.
• /sitemap_agentic_discovery.xml turns up in the sitemap index. This one is worth a caveat: multiple merchant teardowns describe it, and Shopify has not documented it in the help center or the dev docs anywhere we can find. Treat it as observed rather than official, and check whether your own store has it.

The UCP profile is the load-bearing file. It is machine-readable, it declares real capabilities, and agents that speak UCP use it. The markdown files are a much softer signal, which brings us to the awkward part.

Google says it ignores all of this

Google's AI optimization guidance, last updated July 10, 2026, does not hedge: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." The same page says creating an llms.txt will "neither harm nor help" your visibility, because Google Search ignores it.

That is about as blunt as Google gets. If you were planning to hand-craft an llms.txt to win AI Overviews, save yourself the afternoon.

So why does any of this matter? Because Google is one consumer of these files and not the one Shopify built them for. The agents.md content Shopify generates is aimed at shopping agents walking a commerce flow, and the UCP profile underneath it is the thing Shopify and Google co-designed for exactly that job. Whether the markdown layer ends up mattering independently is genuinely unsettled. Nobody has published usage data on it. We would not build a quarter around it either.

What we would do is make sure the file that exists says something true about your store, because it exists regardless of what you think of the standard. Free option, five minutes, no downside. If you want the fuller comparison of how these files differ from the ones you already had, we wrote that up in llms.txt vs robots.txt vs sitemap.xml.

Ten minutes of checking, in order

Open these in a browser tab. You do not need a developer.

• yourstore.com/llms.txt. Does something load? Is it Shopify's generated version or yours? If you used a URL redirect or a third-party llms.txt app before May, assume yours is gone until you confirm otherwise.
• yourstore.com/agents.md. Read it. This is the file with priority now. Check that the store name, contact details and currency it pulls from your admin settings are actually correct, because they come straight from fields plenty of merchants filled in once in 2019.
• yourstore.com/.well-known/ucp. It should return JSON. If it returns a 404 or an error page, that is worth investigating.
• yourstore.com/sitemap.xml. Look at the top of the index for the agentic discovery entry.
• yourstore.com/robots.txt. More on this below, and it is the one most likely to be actively hurting you.

The failure mode we see most is not "the file is bad." It is "the merchant has three llms.txt implementations and no idea which one wins." An app installed last year, a redirect set up by a freelancer, and now Shopify's native file, all fighting. One of them is being served. Usually not the one anybody chose.

Taking it back

Shopify's override mechanism is the same pattern used for robots.txt.liquid, so any theme developer will recognize it immediately. In Online Store, Themes, Edit code, add a file under Templates:

• templates/agents.md.liquid controls /agents.md
• templates/llms.txt.liquid controls /llms.txt only
• templates/llms-full.txt.liquid controls /llms-full.txt only

The fallback order, per the changelog: if there is no template for a given path, it falls back to your agents.md template, then to the Shopify default.

One detail that trips people up. Your template replaces Shopify's output. It does not merge with it. If you write a beautiful curated file that drops the agent discovery links Shopify was emitting, you have made your store harder to transact with, not easier. Read the default first, keep the structural parts, then add what a machine cannot infer: your category structure, your sizing or compatibility guidance, your returns terms, the two or three pages that actually answer buying questions.

And leave out the things that do not belong in a public file. Discount codes. Internal fulfillment rules. Claims you cannot back up.

The file that is more likely to be costing you money

While you are in there: check robots.txt.

Shopify's default is fine. The problem is the custom robots.txt.liquid some stores added in 2023, back when the advice going around was to block AI crawlers on principle. Those snippets are still live on a lot of storefronts, and they now block the crawlers that feed AI answers.

The distinction people miss is that OpenAI runs separate crawlers for separate jobs. GPTBot collects training data. OAI-SearchBot is the one that feeds ChatGPT's search results. Blocking the first does not remove you from ChatGPT answers. Blocking the second does, completely. If someone pasted a blanket block-everything rule in three years ago, you may be invisible in AI answers for a reason nobody on your team remembers.

That is a bigger problem than a generic llms.txt, and it takes about the same amount of time to fix. If you want the wider picture on what agents can and cannot see, our post on agent readiness covers the rest of the surface.

If you would rather not check five URLs by hand, Seokai's site audit tells you exactly what your storefront is serving at each of them, whether your old llms.txt has been shadowed, and whether your robots.txt is blocking anything you would not want blocked: see what your store is actually serving.