llms.txt generator

A plain text file at the root of a site, /llms.txt, written in Markdown, that tells a language model what the site is and which pages to read. The proposal is simple on purpose: an H1 with the name, a blockquote with one sentence, optional paragraphs, then sections of links with a line each. A companion file, /llms-full.txt, can carry the full text of the pages for a reader that wants everything at once.

It exists because an assistant reading a site does not get a home page the way a person does. It gets HTML, often built by scripts it does not run, with a menu and a footer repeated on every page. A short file in its own format is easier to read than the site, and it is where you say what you would say to a new hire in ten lines.

1. 1

   Read the site

   Paste the address and the tool reads the home page for the name and the summary, then up to forty pages from the sitemap, spread across the site's folders, with the title and the description of each. They arrive grouped by folder as sections.

2. 2

   Cut, then order

   An assistant with thirty seconds does not need the privacy policy first. Keep the pages that say what the product is, what it costs, how it works and what it is for; move the rest to Optional or delete the line.

3. 3

   Write the summary

   The blockquote is the sentence most readers take first: what it is, for whom, and what makes it different, in one line. The description of the home page is a start, rarely the finish.

4. 4

   Name the sections

   A line that starts with two hashes opens a section: Docs, Pricing, Guides. Each page line is label, address, and one line on what the page holds, separated by vertical bars.

5. 5

   Serve it and check it

   Copy or download the file, serve it at /llms.txt as plain text, then paste the site into the checker below: it reads the file back, fetches its links, and says what is missing.

• Found and in shape means an H1 first, a blockquote summary, sections of links, and every link fetched answers 200. Nothing to do but keep it current.
• Things to fix are listed in sentences: no H1 first, two H1 lines, no summary, links outside any section, bare URLs without a label, links without a description, links that answer 404 or redirect, HTML where Markdown should be.
• Sections shows each H2 with the number of links under it, and whether an Optional section exists (it is optional in itself: it marks what a reader short on room may skip).
• Links checked fetches up to thirty of the links; the table lists the ones that do not answer 200 directly, with the section each sits in.
• llms-full.txt says whether the companion file exists. It is optional: the full text of the pages as Markdown, for a reader that wants it all in one fetch.

Nothing about your Google rankings. Google has said nothing about llms.txt, and there is no evidence it reads the file. What it changes is how well an assistant that does read it understands your site, and which of your pages it opens first. Several assistants and crawlers fetch it today; more will.

It also does not open a door that robots.txt has closed. An assistant's crawler blocked in robots.txt does not read llms.txt either. Check the crawler access tool below before you spend time on the file.

The generator reads titles and descriptions; the words are yours, and the words are what matters. It reads up to forty pages from the sitemap and the home page; a site without a sitemap gets the home page alone, and the rest is typed. The check reads the shape of the file and fetches thirty of its links, not all; a file of three hundred links is judged on its first thirty. Neither says whether any assistant reads the file: adoption is on the vendors, and as of 2026 none has committed to it.