llms.txt vs robots.txt

robots.txt controls access and crawlers obey it. llms.txt suggests what to read and nothing enforces it. The two files answer different questions.

robots.txt controls access and crawlers obey it. llms.txt suggests what to read and nothing enforces it. The two files sit at the same place in a site's root and answer different questions.

robots.txtllms.txt
Agefolklore since 1994, standard since 2022 (RFC 9309)proposed 2024-09-03
Formatgroups of User-agent and Disallow rulesmarkdown: a title, a summary, then lists of links with notes
Question it answersmay you fetch this pathwhat should you read first, and where is the clean copy
Enforcementhonored by well behaved crawlers, ignored by othersnone. It is a hint.
Read by search enginesyesGoogle has said Search does not use it

What llms.txt actually is

A markdown file at /llms.txt. The proposal defines one required part, an H1 with the site name, then an optional blockquote summary, then H2 sections holding lists of links in the form [name](url): note. An Optional H2 has a defined meaning: those links may be skipped when a shorter context is needed.

The proposal's second half gets forgotten and it is the useful half: serve each page as clean markdown at the same URL with .md appended. Domduck does that at /docs/mcp.md and everywhere else under /docs.

What Domduck records

The same daily pass that reads robots.txt probes for llms.txt and stores whether it is present. Presence only, not the contents, so the reading is one boolean with a date on it and the cost is one conditional request.

On 2026-08-03, 4.76% of the 2,228 domains read that day served an llms.txt. The live figure is on the AI crawlers page.

Why it matters

Do not put access rules in llms.txt. Nothing enforces them, and a site that writes "do not train on this" there has published a preference, not a control. The control is robots.txt, and even that is advisory.

Do not expect llms.txt to move rankings either. Ahrefs measured 97% of 137,000 llms.txt files receiving zero requests. The realistic model is that the file gets read when an agent is pointed at your domain, which is worth something if your audience runs agents and worth nothing otherwise.

Primary source: llmstxt.org, the proposal