LLMs-full.txt

llms-full.txt is a single plain-text file, usually served at /llms-full.txt, that contains the complete Markdown text of a website's documentation or key pages so a developer or AI tool can load all of it in one request. Where llms.txt is a short index of links, llms-full.txt is the content itself, concatenated page by page.

The convention came from Mintlify, which built it in late 2024 with its customer Anthropic and then generated it automatically for every documentation site it hosts. It is not defined in the llms.txt proposal. Version 1 of that proposal described similar generated files for the FastHTML docs (llms-ctx.txt and llms-ctx-full.txt, produced by the llms_txt2ctx tool), and v2, published August 10, 2026, removed the context-expansion tooling and does not mention llms-full.txt. In practice, llms-full.txt is a widely copied habit of docs platforms rather than a standard.

A typical file, as Mintlify generates it, repeats the same block for each page in navigation order:

• The page title as a heading
• The source URL
• The page description
• The full Markdown body of the page

Mintlify also serves the file at /.well-known/llms-full.txt and advertises it in an HTTP Link header, and plugins for Docusaurus and VitePress generate the same file for self-hosted docs. For the shorter index file, a free llms.txt generator crawls a few key pages from your URL and drafts one. Size is the obvious constraint. A large docs site can run to several megabytes, more than many context windows hold, which is one reason the llms.txt v2 proposal moved toward agents reading an index and fetching individual pages on demand.

The main audience is developers and coding tools. A developer can paste one URL into Claude, ChatGPT or an IDE assistant and give it the entire API reference, or point an internal retrieval pipeline at a single file instead of crawling hundreds of pages. That is a real convenience for software documentation, and it is why the format spread through docs platforms.

There is little evidence that it affects AI search visibility. Google's guide to generative AI features in Search says Google Search does not use AI text files or Markdown versions of pages, and that publishing them neither helps nor harms rankings. Promptwatch's own citation data found that .md files made up 0.05% of citations across ChatGPT, Claude, Perplexity and Google AI Overviews in a February 2026 sample, with HTML pages taking the rest; see our study on markdown in AI search. AI search engines retrieve and cite the rendered HTML pages that AI crawlers fetch, so a page that is hard to parse loses citations whether or not a Markdown copy exists elsewhere.

Publish llms-full.txt if you run developer documentation and your docs platform can generate it automatically on every deploy. A stale file is worse than none, because tools that load it will repeat outdated instructions. Exclude private, gated or deprecated pages, and keep the file consistent with the canonical HTML. For marketing sites and blogs, the effort is better spent on server-rendered HTML, clear structured content and accurate schema markup, which is what AI search engines actually read. If you already publish llms.txt, our guide to llms.txt v2 covers the newer approach of linking per-page Markdown files from the index.