Aizii Learn · AI Search

[← Back to Learn](https://aizii.ai/learn)

# How to test if your llms.txt actually works

Testing an llms.txt file means answering three separate questions, and a pass on one does not prove the others. Is the file written in the right shape. Can an outside reader fetch it as plain text. And does what it says still match your store. Test all three, in that order, and fix what fails before you judge whether the file is helping.

The file is a community proposal, maintained at llmstxt.org. It is not a web standard, and the proposal does not claim that any particular engine reads it. The proposal does give one test of its own: ask an agent questions about your content, giving it only your llms.txt as a starting point. That test is useful for judging whether the file explains your store clearly. It does not prove that any engine reads the file on its own, so treat it as a usefulness check, not as proof of adoption.

## Layer one: the shape of the file

Start with the text itself, before you check anything live. Open the file you wrote and check it against the proposal, which defines the sections in a fixed order.

The first line that matters is the H1 with your site or brand name. That is the only required part, and there should be exactly one H1. A second H1, or a file that starts with a link list and no name at all, is a shape failure, even if the file loads.

Next, the blockquote summary should sit directly after the H1, one short paragraph with the key facts about your store. Then any extra detail, with no headings in that part. Then H2 sections of links, each item written as a link title, a URL, and a one line description. A section called Optional holds secondary links an agent can skip.

Check every link as you go. Each one should be a full URL, not a relative path, and each description should say what is behind the link and why it matters. A link with no description is a shape failure in practice, because the description is what makes the file useful. A link to a page that no longer exists is worse, because the file exists to be trusted.

## Layer two: can an outside reader fetch it

A file that looks right on your computer can still fail live. Fetch it the way an outside reader would, from a browser and from the command line, and compare what each one gets.

Open [https://\[yourdomain.com\]/llms.txt](https://[yourdomain.com]/llms.txt) in a browser, replacing yourdomain.com with your actual domain. You should see your Markdown served as plain text. A 404 means the file is in the wrong place or a redirect is misconfigured. A login screen means the file is behind authentication, and no outside reader can use it.

Then check the headers. Run this, replacing yourdomain.com with your actual domain:

```
curl -I https://[yourdomain.com]/llms.txt
```

Look for a 200 status and a text/plain content type. Then fetch the body and look at the first lines:

```
curl https://[yourdomain.com]/llms.txt
```

The body should start with your H1, a line starting with #. If the body starts with an HTML tag instead, your site is returning its HTML shell with a 200 status, and the file is not actually being served. That trap catches single-page sites in particular: the status looks healthy, and the body is junk to an agent. A browser will happily render the HTML fallback and hide the problem, so check the body with curl, not just the status.

Check for a redirect chain as well. A short redirect to the right place is survivable. A redirect to a login page, to the homepage, or to a different domain is a fail, because the reader asked for the file and did not get it.

Finally, check that robots.txt is not blocking the file. The llms.txt file should be accessible to all crawlers, including the AI-specific ones. If your robots.txt disallows the file, or your bot protection layer challenges outside readers while letting your browser through, the file is visible to you and invisible to the readers it was written for. Test it as a stranger, not as the site owner, and if the answers differ between a browser fetch and a command line fetch, decide whether that difference is what you meant.

## Layer three: does it still tell the truth

A file can pass the shape test and the fetch test and still be wrong, because what it says no longer matches your store. This is the layer no syntax check can catch for you.

Open every link in the file, starting with the highest value ones: the collections, the guides, and the policies. Confirm that each one resolves without a 404 or a server error, and that the page it lands on still says what the link description claims. A collection that was renamed, a guide that was rewritten, or a policy whose return window changed all make the file stale, and a stale file teaches a model last season's catalog.

Check the summary too. If you stopped shipping to a region, changed what you sell, or closed a physical shop, the blockquote at the top needs to change with it. The summary is the paragraph a model reads first, so a wrong fact there spreads further than a wrong fact in a link description.

Re-run the whole test after every change to the file, and again after every catalog change. There is no submission step, no search console, and no ping endpoint, so nothing will tell you the file went stale. You have to check it yourself.

## What a pass looks like

A passing file has one H1, a short summary directly after it, H2 sections of links with descriptions, and an Optional section for secondary links. It loads at [https://\[yourdomain.com\]/llms.txt](https://[yourdomain.com]/llms.txt) with a 200 status and a text/plain content type, and the body starts with your H1, not with HTML. Every link in it resolves, every description still matches the page it points to, and robots.txt does not block the file.

A failing file is usually one of these: a 404, a login screen, an HTML shell returned with a 200 status, a second H1, links with no descriptions, dead links, or a summary that no longer describes the store. Fix the failure at its layer. A shape failure is fixed in the text. A fetch failure is fixed in how the file is served. A truth failure is fixed by updating the file to match the store.

## A toy store that passed the status and failed the body

Picture a toy store, Brick and Bear Toys, that publishes an llms.txt file and checks it in a browser. The page loads, the Markdown looks right, and the status is 200. The store assumes the file works.

A command line fetch tells a different story. The body starts with an HTML tag, not with the H1. The site is a single-page app, and it returns its HTML shell for any path, with a 200 status. The browser rendered that shell and hid the problem. The file is not being served at all.

The fix is on the serving side, not in the text. The file needs to be served as plain text at the root, with a 200 status and a text/plain content type, and the body needs to start with the H1. Once that is fixed, the store re-runs the link checks and the summary check, because a file that was never served has never been tested for truth either.

## The agent test, used carefully

The proposal suggests asking an agent questions about your content, giving it only your llms.txt as a starting point. Do that, and judge whether the answers are accurate and complete. If the agent cannot answer a question the file should cover, the file is missing a section or a description is too thin.

Do not use that test as proof that an engine reads your file on its own. Asking a model a question about your brand does not validate that llms.txt is being consumed, because a model can answer from training data, from other pages it has read, or from live search. The only verifiable checks are the ones above: the file loads with a 200 status, the body is plain text, the links resolve, and, where your server logs are available, AI crawler requests for the file appear in them. A log entry showing a crawler fetching the file is evidence that the file was requested. It is not evidence that the crawler used it to answer a question.

## What this guide does not cover

Testing the file is one page of a larger job. Your catalog drifts, crawler behavior changes, and being readable is not the same as being recommended or bought from. That is what monitoring and the full audit exist for. For what goes in the file for a store, see our guide on what to put in llms.txt for an ecommerce store, and for the full format, see our [llms.txt guide](https://aizii.ai/learn/ai-search/how-to-write-a-good-llms-txt-file).

## Check your own brand now

Run the free scan. One click, no signup, and it shows exactly how readable your site is to AI tools today: aizii.ai/aeo-scan

[Run the free scan](https://aizii.ai/aeo-audit)

## Sources

-   llms.txt proposal (specific): [https://llmstxt.org](https://llmstxt.org/)

Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
