Free llms.txt Generator and Validator

The generator starts at your homepage. It follows any redirect to the address your site settles on, such as the www or https version, and reads the homepage's title, meta description and og:site_name tag. Then it looks for your sitemap: the Sitemap: lines in robots.txt first, then, if robots.txt names none, common addresses such as /sitemap.xml. When the sitemap is an index of other sitemaps, it reads up to 5 of them, one of each kind first, so a WordPress page sitemap is read even when the index lists it after many post sitemaps.

1. It keeps only pages on your own site, lists each page once (with or without www, with or without a trailing slash), and sets aside files such as PDFs and images without fetching them.
2. It picks up to 30 pages, homepage included: the shortest addresses first, taking one page from each part of the site (each first folder, such as /blog or /docs) before any part gets a second. From a very large sitemap it chooses among the 10,000 shortest addresses, and the result says when that limit, or the 5-sitemap limit, applied.
3. It fetches each chosen page once. A page that redirects is listed at the address it ends at, and only if that address is on your site and answers 200 with a web page.
4. It writes the file: your site's name as the heading, your homepage's meta description as the summary, and one section per folder, with each page listed under its own title and its own meta description.

If your site has no sitemap, or its sitemap lists none of your pages, the pages come from the links on your homepage instead. The result tells you which source was used.

Every line of the file comes from a page we fetched. A page with no meta description gets no description in the file, rather than one we made up. A page that fails to load, answers an error, redirects to another site or to your homepage, or refuses our checker is left out, and is listed under the file with the reason. One run reads at most 30 pages and stops after 60 seconds; if it stops early, the result says so and names the pages it did not reach.

llms.txt is a proposal, first published by Jeremy Howard in September 2024 at llmstxt.org, for a plain markdown file that tells AI tools what a website is and which of its pages matter. It lives at /llms.txt, next to robots.txt and sitemap.xml, but it does a different job.

The proposal requires only one thing: a first line with the site's name as an H1 heading. Everything else is optional: a blockquote summary under it, any notes, then H2 sections that list links, each written as a markdown link with an optional note after a colon.

Not in any way that has been confirmed. Google says plainly that Google Search does not use llms.txt: its guide to AI features in Search, last updated in July 2026, says such files are not needed to appear in Search, including its AI features, and will neither help nor harm your visibility or rankings.

As of October 2026, no major AI company has said that its crawler or its answer engine reads llms.txt. OpenAI, Anthropic and Perplexity document their crawlers and how they follow robots.txt, but none of them documents reading llms.txt. Some developer documentation sites publish one anyway, and a person can always hand the file to an AI assistant themselves.

Lighthouse's llms.txt audit treats the file as optional, and checks for server errors and basic structure: an H1, links and a minimum length. So treat llms.txt as a cheap convention for AI agents and developer tools that look for it, not as a way to rank. What decides whether AI answers mention you is what they find about you across the web.

The validator reads /llms.txt at the root of your site, or a file you paste, and checks it against the proposal's format. A missing file is reported as missing. If the address answers with a web page instead of a text file, which happens on sites that show the homepage for every unknown address, it says so and does not check the page as llms.txt.

The score starts at 100 and loses a fixed amount for each problem. It sums up the checks in one number; it is not a published standard, and a file can follow the format with a score below 100. The proposal also lets you place an llms.txt in a folder, such as /docs/llms.txt for the pages under /docs. The URL check reads only the file at the root, so paste any other file to check it.

This is the shape the generator writes for a small site with a blog, line by line:

Each section is named after its folder, so /case-studies becomes Case studies. Titles and descriptions are written as the pages give them, with characters such as [ and ] escaped, so no page title can break the file or add a link of its own.

A generated file is a starting point built from what your pages say about themselves, so read it before you publish it. Remove pages you would not send an AI tool to, such as a login page or a thin tag page. Rename a section if its folder name is unclear. Where a page has no description, the better fix is a meta description on the page itself, which also gives search results a better snippet; you can also write a note into the file by hand.

Publish the file as plain text at /llms.txt, so it answers 200 with a text or markdown content type, then check it with the validator above. Generate it again when you add or move important pages, because the file does not update itself.

The generator reads raw HTML and does not run JavaScript. If your site builds its pages in the browser, titles and descriptions added by scripts are not seen, and those pages may be listed by their address alone.