llms.txt is a proposal from Answer.AI (published by Jeremy Howard) for a plain-text Markdown file at /llms.txt that gives an AI assistant a short, curated map of a site's most useful pages. The convention is not a web standard, and no browser or search engine is required to read it — it exists because some AI tools already fetch it when they encounter it, and more documentation platforms (several static-site generators and docs hosts) have started publishing one by default.
These three files answer different questions, and a site can have all three at once:
If you already publish a sitemap with hundreds of URLs, llms.txt is not a second copy of it — pasting your whole sitemap in verbatim (which this tool's sitemap mode makes easy to do by accident) defeats the point. The page-list tab is the better starting point for most sites: hand-pick the pages that matter.
# Site or Project Name
> One-sentence summary of what this is.
## Docs
- [Getting started](https://example.com/docs/start): what this page covers
- [API reference](https://example.com/docs/api)
## Optional
- [Changelog](https://example.com/changelog)An H1 title, an optional blockquote summary, then any number of "## Section" headings each holding Markdown link bullets. The convention's own suggestion is an "## Optional" section at the end for links that can be skipped first if an assistant has limited context to spend — this generator lets you name any section, including that one.
A docs site with a few hundred pages, several formats of the same content, or a lot of historical/versioned material is exactly the case the convention was written for.
An llms.txt that links to a page you deleted, or omits a major new doc, is worse than not having one — the same "a guard nobody runs is worth less than no guard" problem any generated artifact has. If you regenerate it from a sitemap or docs tree on every deploy (the CLI is built for exactly that — see the main page), it can't go stale the way a hand-maintained one will.
No. robots.txt controls crawler access, sitemap.xml lists every URL that should be indexed, and llms.txt is a curated, short map of the pages that actually explain a site or product — a different job from either, and a site can have all three at once.
It helps most on documentation sites, API references, and developer-facing tools where there is real depth to summarize and a real audience (including AI coding assistants) that benefits from a shortcut past the marketing pages. A five-page brochure site has little to gain.
Everything your sitemap already lists is the wrong scope — llms.txt is meant to be curated and short, not exhaustive. Leave out marketing pages, legal boilerplate, and anything not meant to inform an assistant answering a question about the product.