llms.txt

The llms.txt proposal places a markdown file at yoursite.com/llms.txt containing a structured summary of the site: what it is, plus curated links to key documentation, product pages and guides, optionally with companion markdown versions of important pages. The idea is that LLMs working within a limited context window get a clean, token-efficient map instead of parsing navigation-heavy HTML. Think of it as a sitemap for reasoning engines rather than crawlers: prioritized, annotated and human-readable.

As of mid-2026, llms.txt remains a community convention, not a ratified standard. Thousands of sites, particularly developer tool companies, have published one, and some AI coding assistants and agents read them. But OpenAI, Anthropic, Google and Perplexity have not committed to honoring llms.txt in their crawlers or search products, and Google representatives have publicly downplayed it. Treat claimed visibility gains skeptically: there is no verified evidence that publishing llms.txt improves citations in major AI engines today. It does not replace robots.txt, which remains the binding access-control mechanism.

The cost is an hour of curation; the downside is essentially zero. For documentation-heavy sites, llms.txt has immediate utility for AI agents and developer tools that already consume it, and it positions you well if major engines adopt the convention. Keep it short, link only canonical resources, and update it when key pages change. Just sequence it honestly in your GEO strategy: crawlability, server-rendered content and structured data are proven levers; llms.txt is a cheap, speculative complement, not a substitute.