Enter your domain. We read your homepage title and description, the sitemaps declared in your robots.txt, and the URLs inside them, then produce a sectioned llms.txt draft in the llmstxt.org format. Treat it as a starting point — rewrite the titles and add descriptions before you publish.
What is llms.txt?
llms.txt is a plain-text Markdown file that sits at your site root. When a language model reaches your site, instead of wading through navigation menus, cookie banners and marketing blocks, it finds one file answering "what is this site, and where are the pages that matter". Jeremy Howard proposed the format in September 2024: a # heading (the site name), a > block (a one-sentence definition), then ## sections, each holding an annotated list of links. robots.txt grants access; llms.txt gives directions. Neither replaces the other.
Does llms.txt actually work?
The honest answer: the evidence is still thin, and we would rather say so. In May 2026 Google stated plainly that AI Overviews and AI Mode do not need llms.txt. OpenAI, Anthropic and Perplexity have not announced official support either. SE Ranking's crawl found the file on roughly 10% of the top 1,000 sites. So ignore anyone telling you that adding llms.txt gets you recommended by ChatGPT. Two concrete benefits remain. First, real agents do fetch the file — you will see the requests in your own server logs — and for those, a ready summary of your site beats nothing. Second, writing the file forces you to answer "which 40 pages of mine actually matter", which is the GEO work itself. Ten minutes of effort, zero risk.
What does a good llms.txt look like?
A bad llms.txt is a sitemap converted to Markdown: 800 links, no descriptions. A good one is a selection. Thirty to eighty links is plenty. Every line reads - [Title](https://...): one sentence saying what it covers, and that sentence does the real work — it lets a model tell which page answers a question without opening it. Name sections the way your visitors would ("Guides", "Pricing"), not with internal jargon. Put informational content at the top and commercial pages below; the file is read from the beginning. Write it in the language of your site — an English llms.txt on a German site sends the wrong signal.
llms.txt versus llms-full.txt
llms.txt is an index: headings and links. llms-full.txt is the content itself — your entire documentation concatenated into a single Markdown file. The second one makes sense mainly for documentation sites, where it lets a model learn your product in one request. It is not required for every site, and flattening a large site into one file produces megabytes of text. The rule is simple: if you have technical documentation explaining how your product is used, add llms-full.txt; if you run a corporate brochure site, you do not need it. This tool produces the llms.txt half.
Where it goes and how to verify it
The file must be served from exactly https://yourdomain.com/llms.txt — the root, not a subdirectory. Your server has to return it as text/plain; some configurations force extensionless or unknown files to download, or return 404 outright. After publishing, open it in a browser: seeing plain text rather than HTML means it is correct. Then watch your server logs for /llms.txt requests — that is the only way to know which agents genuinely fetch it, as opposed to guessing. Refresh the file as your content changes; a six-month-old page list is more misleading than no file at all.
Common mistakes
Three come up constantly. One: publishing the file while still blocking AI agents in robots.txt — that is handing out directions and locking the door; check your robots.txt first. Two: leaving links undescribed, which tells a model nothing it could not already derive from your sitemap. Three: treating llms.txt as authoritative — it is not a ranking signal, it does not index anything, and it is no substitute for the quality of your pages. Visibility in AI answers still comes down to what your pages say, how they are structured, and how you are referenced on sources you do not own. llms.txt makes that easier to read; it does not replace it.
