llms.txt Generator for Any Website
Paste a site and get a draft llms.txt: the site name, a one-line summary, and every page from the sitemap grouped by section with its title and description. Edit, copy, download.
AI assistants are reading your site whether you have prepared it or not. When a customer asks ChatGPT or Perplexity whether your product does something, the answer comes from whichever pages the crawler happened to fetch and understand — not necessarily the pricing page, the returns policy or the help article that would have settled it. llms.txt is a proposed convention for fixing that: a Markdown file at /llms.txt that tells an assistant what the site is and which pages matter.
Paste your site's address. The tool reads its sitemap — or, if there is none, crawls up to 40 pages — and takes each page's title and meta description. From those it drafts the file: an H1 with the site name, a blockquote summary taken from the home page's description, then a section per top-level path (/help, /pricing, /blog) with a - [Title](URL): description line per page. Edit it in the box, then copy or download. It is a draft from your own metadata, not a model's guess: where a page has no description, its line will be thin, and that tells you which pages need one. Nothing is stored beyond a short cache.
This ran once, on one input. The product runs on your whole site, every conversation.
Draft an agent from the same pages — start freeWhat each input means
- 01
Site URL
The home page, with the scheme. The tool needs the sitemap for a full draft, so if the sitemap finder shows none, expect the 40-page crawl and a shorter file. Run it on the host where the file will live.
- 02
The heading and summary
The H1 is the site's name, taken from the home page's title. The blockquote under it is the home page's meta description. If either reads like a slogan rather than a sentence about what you do, rewrite it in the box; this is the line an assistant reads first.
- 03
Sections by path
Pages are grouped by their first path segment: everything under /help in one section, /pricing in another. Each line is the page title, the URL and its meta description. Delete sections an assistant doesn't need (tag archives, author pages) and move the important ones up.
- 04
Edit, then copy or download
The box is editable. Trim it to the pages that answer questions, fix thin descriptions, then copy or download and publish it at /llms.txt on the same host, served as text/markdown or text/plain.
Questions about this tool
A proposed convention, first published in 2024, for a Markdown file at /llms.txt that describes a site to AI assistants: what it is, in one line, and a curated list of the pages that matter, each with a short description. It is to language models what a sitemap is to a search engine, except that it is meant to be read as prose and to be short.