llms.txt is a proposed convention: a Markdown file at your site root that gives language models a curated map of your most important content, with short descriptions and links to clean, readable versions of each page. It was proposed by Jeremy Howard of Answer.AI in 2024 as a way to help models find signal on sites full of navigation, scripts, and boilerplate.
What goes in the file
The format is deliberately simple: an H1 with the site name, a blockquote summarizing what the site is, then sections of links with one-line descriptions. Some sites pair it with llms-full.txt, a single file containing the full text of key pages. Documentation-heavy products publish the most elaborate versions; a blog's file can be a dozen lines.
Who actually reads it
This is where honesty is due: as of early 2026, no major answer engine has confirmed that it uses llms.txt for retrieval or citation. Google has said publicly that it does not use the file. Adoption is real on the publishing side, with thousands of sites shipping one, but evidence that it changes crawling or citation behavior is anecdotal. Treat any vendor claiming llms.txt is an established ranking factor with suspicion.
Should you bother?
- It costs an hour and carries no risk: nothing in the file can hurt your existing crawling or ranking.
- It may help agent-style tools that fetch pages directly rather than through a search index.
- It is not a substitute for the fundamentals: a sitemap, clean robots.txt, and quotable content move citations; llms.txt currently does not, measurably.
Our position: ship one if the hour is spare, skip it if the same hour could improve an actual article. The mechanics that demonstrably matter are covered in Schema markup for answer engines and the AEO starter checklist.