/llms.txt is a plain markdown file at the root of a website that tells AI agents which pages are canonical. The specification lives at llmstxt.org. The format is intentionally minimal: a title, an optional tagline and description, then a sequence of H2 sections containing markdown links. Agents like ChatGPT, Claude, and Perplexity request the file first when they want to understand what your site is about and which pages they should ground their answers in.
Why a separate file?
Two reasons. First, the content is curated. A sitemap lists every URL on a site, including pages that exist for SEO but aren’t the best citation. /llms.txt is a hand-picked subset: docs, pricing, support, policies, changelog. Second, the format is lossless for agents. Markdown is the lingua franca of LLM training data; an agent can parse a markdown link list with no ambiguity.
What goes in it?
A typical file has six sections, each with a few links:
- Docs. Quickstart, guides, conceptual docs.
- Pricing. Your plans page (and the FAQ if it lives elsewhere).
- API. Reference, SDKs, playground. Skip if you don’t ship an API.
- Support. Contact, status page, help center.
- Policies. Terms, privacy, security, DPA.
- Changelog. Release notes.
Where does it live?
Apex of the domain: https://your-domain.com/llms.txt. Same place as /robots.txt and /sitemap.xml. If your CMS doesn’t allow files at the root, route a Cloudflare Worker (the Tools page has copyable snippets for Webflow, Shopify, WordPress, Next.js, and static sites).
Does it help SEO?
Indirectly. /llms.txt is read by AI agents, not by Google. But agents that ground answers in your linked pages cite your site more accurately, and that drives traffic. Sites without a file get summarised from the homepage, which usually means agents quote marketing copy instead of the docs you wrote.