llms.txt is a simple text file in a website's root (at /llms.txt) with which you tell AI systems and agents which content and endpoints are relevant to them — curated, in Markdown and without the ballast of navigation, ads and scripts. For AI it is roughly what robots.txt is for search-engine crawlers: a machine-readable wayfinder.
A short overview of the website, followed by curated links to the most important content — documentation, core pages, policies — each as a Markdown list with a brief description.
Optionally it points to structured data or an MCP server. Some sites additionally offer an llms-full.txt with the full content in one file.
AI models have limited context windows and stumble over HTML ballast. An llms.txt delivers the relevant content bundled and clean, so a model grasps the website faster and more accurately.
It is a cheap building block of an agent-ready website — alongside Schema.org, clean HTML and answer-first content.
llms.txt is a proposed standard, not a law — not every provider consumes it. Its usefulness grows with adoption.
Because the file is cheap to create and never hurts, it belongs in a GEO strategy: low risk, potentially growing return.
No. robots.txt controls what crawlers may do; llms.txt curates which content is relevant to AI models and how it is structured. They complement each other.
Yes, the proposal places the file at /llms.txt in the root, in Markdown. Sometimes there is an additional llms-full.txt with the full content.
Its usefulness grows with adoption. Because the file is cheap to create and part of the agent-ready basics, it is already worthwhile as part of a GEO strategy.