An AI-readable map of your site. No API key, no crawl service, no signup — list your pages, get a spec-shaped file.
Add one row per important page. Output follows the llms.txt draft (H1, blockquote summary, link list).
Updates live. Save as llms.txt at your domain root.
Your llms.txt appears here…
This generator never crawls anything — you curate the page list by hand, which is exactly the point: llms.txt is a human-judgment map, not an auto-dump. Output updates live with every keystroke.
1 · Name the site. Fill Site name and the one-line summary. They become the file's # Title line and its > blockquote summary — the two lines every conforming file opens with. Write the summary for a stranger model: what the site is, who it serves, in one sentence.
2 · Add one row per important page. Each page needs a title, a full https:// URL, and a one-line description of what it covers. The tool validates as you go: titles are required, URLs must parse with http(s), and duplicates are rejected. List the pages an agent should reach first — docs, pricing, key guides — not your entire sitemap.
3 · Read the live output. Everything you add lands under a ## Main section as - [title](url): description lines. The page counter and byte counter under the preview track size — keep the file small enough to fit in model context (more on that in §04). Not sure where to start? Load sample prefills three sensible rows you can edit or clear.
4 · Copy or download, then deploy. Copy drops the text on your clipboard; Download saves it as llms.txt. Upload that file to your domain root so it serves at example.com/llms.txt, confirm it returns Markdown or plain text with a quick curl check (see §03), and you are done — no registration, no crawler to notify.
llms.txt is a community proposal — not a W3C or IETF standard — by Jeremy Howard of Answer.AI, published September 3, 2024, with the draft specification maintained at llmstxt.org and the reference implementation at AnswerDotAI/llms-txt. The premise: HTML pages wrap information in navigation, ads and scripts that are expensive to convert back into clean text, so sites publish one concise Markdown map that points agents at the good parts. This tool's output follows that draft exactly:
| Line | Meaning | Example |
|---|---|---|
| # Title | Site name, first line | # mosaic/lab |
| > Summary | One-line blockquote description | > Small AI tools. Real experiments. |
| ## Section | H2 grouping, e.g. Main | ## Main |
| - [name](url): desc | One link per line, description after colon | - [Toolbox](https://…/tools.html): Eight utilities. |
Variants you will meet. llms-full.txt inlines the linked pages' full content into one big file for pasting into a model; documentation projects also generate expanded context files (the reference tooling calls this llms_txt2ctx) that optionally fold in secondary links. Keep the root llms.txt itself lean — the spec's rule is that the file stays small enough to fit in context while detail lives behind the links.
Deploy in four moves. Save the output as llms.txt, upload it to the domain root, serve it as text/markdown or text/plain, and verify with curl -I https://example.com/llms.txt — expect HTTP 200 and a text content type. Sub-path files (example.com/docs/llms.txt) are legitimate for covering one section; root placement is the default discovery location.
The ranking truth, stated plainly. Google does not use llms.txt for ranking: Google's Gary Illyes has said Google will not crawl or use these files, and that normal SEO is what surfaces content in Search and AI Overviews (Search Central Deep Dive, Jul 2025; verified Sep 2026). Treat llms.txt as an advisory map for agents already visiting, never as a ranking lever. For contrast, robots.txt (RFC 9309) is the binding standard where Allow and Disallow directives actually constrain crawlers — llms.txt constrains nothing and permits nothing.
The six confusions around this format, answered against the draft spec and vendor statements current to Sep 2026.
No. Google's Gary Illyes stated at the Search Central Deep Dive in July 2025 that Google will not crawl or use llms.txt files, and that ordinary SEO is what gets content into Search and AI Overviews. John Mueller has likewise called the file speculative from Google's side. Publish llms.txt so helpful agents navigate your site better — never as a substitute for titles, content quality, and links. Anyone selling it as a ranking boost is selling fiction.
Three jobs, three files. robots.txt — a real standard, RFC 9309 — tells crawlers what they may and may not fetch, with Allow and Disallow rules crawlers honor. sitemap.xml enumerates URLs so crawlers discover them. llms.txt is neither: an advisory Markdown summary plus curated links that helps an agent already reading your site decide where to go next. Keep all three: robots.txt for control, sitemap for discovery, llms.txt for guidance.
Small enough to fit comfortably in model context — a title, a summary, and a curated link list, typically a few kilobytes. The spec's design rule is explicit: the file itself stays compact while detail lives behind the links, fetched only when needed. Do not paste article bodies, API dumps, or your whole sitemap into it. If the byte counter on this page climbs past a few thousand, you are cataloguing instead of curating — cut ruthlessly.
Links that are useful in some contexts but not core to understanding the site: archives, tangential posts, alternate formats, deep reference pages. Clients building a tight context window may skip Optional entirely, so essentials always live under Main. Note the honest UI detail: this generator currently files every page you add under ## Main, following the draft's Main-first convention — structure Optional by hand if your file needs the split.
Follow John Mueller's pragmatic rule: when an AI platform that brings you clients says it needs the file, make one. Documentation sites consumed by coding agents benefit most — several AI labs publish llms.txt for their own developer docs. A small brochure site with five pages gains little today; the cost is low but so is the payoff. Revisit when your server logs show agents actually requesting the file.
No. There is no fetcher, no API key, and no network call — you type or paste each page, the tool checks the title exists, the URL parses over http(s), and the URL is not already listed, then formats the file locally. That manual step is a feature: the file's value is your editorial judgment about which pages matter. For bulk generation from an existing sitemap, the reference llms_txt tooling linked in §03 is the right next step.
Proposal history and format claims were verified against the draft spec at llmstxt.org and the Answer.AI reference repository in September 2026 (Howard, Sep 3 2024; v2 updates noted where relevant). The tool-behavior claims — manual page list, https URL validation, duplicate rejection, # / > / ## Main output shape, live page and byte counters, copy and download actions — were re-read from this page's own JavaScript the same month. Everything runs client-side with no crawl, no key, and no signup. Limits to respect: llms.txt is a proposal no major search crawler requires, it confers no ranking benefit, and anything you list in it is public by definition — never link non-public URLs.