What is llms.txt and should you put one on your site?

Should : llms.txt is a standard introduced in September 2024 by Jeremy Howard (Answer.AI). It places at the root of your site a Markdown index designed to be read by Large Language Models. Short answer in 2026: yes, ship one. It is now the first reflex of any GEO optimisation.

What is llms.txt and should you put one on your site, in numbers?

The /llms.txt file was proposed by Jeremy Howard (Answer.AI) on September 3, 2024 and formalised on llmstxt.org. The logic is straightforward: LLM context windows are too narrow to ingest a full website (HTML, scripts, navigation, ads), and converting noisy pages into clean plain-text remains imprecise. llms.txt is a markdown file placed at the root of a domain that hands the model an orientation map: H1 title (mandatory per the spec), blockquote summary, then markdown sections of links to the most important resources.

The spec also encourages a second file: llms-full.txt, which contains the whole useful documentation flattened into a single markdown stream that a model can consume in one pass. Neither replaces robots.txt, they sit alongside it. robots.txt controls who is allowed to crawl; llms.txt controls how easy comprehension is for those already authorised.

Should you ship one? Yes, by default, for any editorial or premium B2B site. Three reasons. First, it signals machine-readable discipline. Serious publishers (Stripe, Vercel, Mintlify, Anthropic) shipped llms.txt across 2024-2025; absence is becoming a weak but real negative signal. Second, implementation cost is trivial, a static markdown file, generated once and version-controlled in CI. Third, the spec is still under open discussion on GitHub (the public repo accepts community contributions), so early adopters can shape the final version.

Two caveats. No answer engine (OpenAI, Anthropic, Google, Perplexity) has publicly confirmed it consumes llms.txt during crawl or retrieval as of 2026. It is a community standard, not a mandatory protocol. "The real value of llms.txt is indirect: an expertise signal, a cleaner editorial architecture and easier audits," notes Lorenzo Eeman, founder of PROEMA. Never sell it as a citation guarantee. The GEO Rocket portfolio, expertcafe.be, zeroproof.one, expertvin.be, ships the llms.txt + llms-full.txt pair as a baseline, but treats it as one of eighteen technical standards inside the PGSM v1.0 framework, not as the deliverable itself.