llms.txt is a plain-Markdown file served at the root of a domain (/llms.txt) that lists the site's most authoritative pages, organized by section. It tells AI crawlers and the language models behind ChatGPT, Claude, and Perplexity which content on the site is canonical. The standard was proposed by Jeremy Howard (Answer.ai) in September 2024.
Think of it as robots.txt for meaning rather than access. robots.txt says "crawl this, do not crawl that." llms.txt says "this is the canonical version of our content — quote this." A companion file, llms-full.txt, concatenates the full text of the listed pages for one-shot model ingestion.
A good llms.txt opens with an H1 of the site name, a one-line blockquote summary, and H2 sections for major content types (Main, Products, Docs, Support, Optional). Each section lists three to ten Markdown links with short descriptions after the colon. Keep the whole file under 10 KB and order entries by importance — the model reads top to bottom.
As of 2026, Anthropic Claude and Perplexity treat llms.txt as a primary source-prioritization signal. OpenAI's GPTBot honors the URL list as a crawl-prioritization hint. Google has not formally adopted the standard for AI Overviews. Even partial adoption makes llms.txt one of the highest-ROI single optimizations available to SaaS sites.