The llms.txt proposal, written by Jeremy Howard in September 2024 and revised in August 2026, defines one file: a markdown document at /llms.txt with an H1, a short summary in a blockquote, and H2 sections of links, plus an "Optional" section an agent can skip. It says nothing about llms-full.txt.
llms-full.txt is a convention from documentation hosting. Where llms.txt is a table of contents, llms-full.txt is the whole book in one file: every page's text, concatenated, so an agent can load a product's documentation in one request. Mintlify serves one for its own docs at 1.6 MB, and Anthropic's documentation serves one at 46 MB; we read both on 2026-10-21.
Does anyone read it
The proposal notes that OpenAI, Anthropic and Google publish llms.txt files for their own developer documentation. It does not claim their assistants read the files when answering, and none of the three has said so. In our 25-question check on 2026-10-21, the only citation of llmstxt.org was ChatGPT citing the specification itself when asked what llms.txt is. Our llms.txt guide has the longer version of that story.
Do you need one
If you run a documentation site with hundreds of pages, llms-full.txt is cheap to generate and costs nothing to serve. A developer who points an agent at it gets your whole manual in context.
For a company website with twenty or fifty pages, llms.txt alone does the job the file was proposed for: it tells an agent what the site is and where the important pages are. The pages Seen42 hosts ship an llms.txt for that reason and no llms-full.txt. Spend the effort on pages with direct answers, which the assistants demonstrably read.
