The llms.txt Guide: Help AI Engines Find Your Content
llms.txt is a plain-text file served at your domain root that gives AI engines a curated, readable map of your most important content. It does for answer engines what sitemap.xml does for crawlers — but in Markdown meant to be read, not parsed.
The format
The convention is simple: an H1 with your site name, a one-paragraph summary in a blockquote, optional key facts, then sections of links with one-line descriptions. Everything is standard Markdown.
- Start with `# Your Site Name`.
- Add a `>` blockquote stating what the site is in one or two sentences — this is the paragraph engines quote.
- List your best pages as `[title](url): description` bullets, best first.
- Serve it as text/plain at /llms.txt and keep it under a few kilobytes.
What to include (and what to skip)
Include the pages you would want quoted: definitive guides, product pages, methodology docs and pricing. Skip legal boilerplate, tag archives and anything you would not stand behind as an answer.
Keep descriptions factual and specific. “Complete guide to X with steps and a comparison table” beats “Read our blog”. Engines use these lines to decide whether the page is worth fetching.
Does it actually work?
Adoption is early but growing, and the cost is ten minutes. Several AI crawlers already fetch llms.txt when present, and the file doubles as a clean summary for any agent visiting your site. Maveriko scores its presence as part of AI retrievability — the same check this site passes with its own llms.txt.
An honest assessment of whether to bother
No major AI engine has publicly documented llms.txt as a ranking or retrieval input, and anyone telling you otherwise is guessing. That is the state of the evidence as of 2026, and it should shape how much effort you spend here.
The case for adding one anyway is that it costs about ten minutes, it cannot hurt, and the exercise of choosing which twenty pages you would most want an engine to quote is genuinely clarifying. Most teams that do it discover several of their nominated pages are client-rendered, undated or unquotable — which is a finding worth far more than the file.
The case against spending more than ten minutes on it is simply opportunity cost. If your pages are not server-rendered, or your robots.txt blocks the crawlers, or your sections do not answer their own headings, all of those matter demonstrably and llms.txt does not. Do it last, not first.
Frequently asked questions
Does llms.txt actually work?+
There is no published evidence that any major engine reads it, and no engine has confirmed support. Treat it as low-cost optionality — sensible to have, unwise to prioritise over crawler access, server rendering or schema.
Where does the file go?+
At the root of your domain, served as plain text at /llms.txt, exactly like robots.txt. Keep it under a few hundred lines and link only to pages that are genuinely worth quoting.
Is llms.txt the same as robots.txt?+
No, and they solve opposite problems. robots.txt is enforcement — it tells crawlers what they may not fetch, and the major AI crawlers honour it. llms.txt is a suggestion — it tells a model which pages you consider most useful, and nothing is obliged to read it.
Sources
Published · Last reviewed .