/ llmtxt.info

llms.txt Adoption, Who Uses It?

A dated record of public llms.txt publication, with a fixed-panel tracker, direct file links, and explicit limits on what publication can prove.

Last updated:

Adoption snapshot

The relevant observable fact is narrower than a market-adoption claim: whether a public host returns a qualifying llms.txt response on a given date. This page records that signal for a fixed panel and links back to the source files. It does not infer discovery, consumption, citations, traffic or ranking impact from publication alone.

Measured adoption (our tracker)

Claims like “tens of thousands of sites use llms.txt” circulate without a source. So we measure it ourselves: a fixed panel of 219 notable hosts across eight sectors, fetched and parsed every week by an automated checker. Latest run: .

51.8%
of 218 tested hosts serve llms.txt
113
verified live files in the directory
14.0 KB
median file size
Sector Tested Serving Rate
AI & ML 36 19 52.8%
Developer tools 61 42 68.9%
Docs platforms 7 1 14.3%
E-commerce 13 7 53.8%
Fintech 16 8 50%
Media 14 0 0%
Other 23 4 17.4%
SaaS 48 32 66.7%

Methodology and limits

The panel is a fixed, curated list of 219 hosts (published at /adoption-panel.json); membership does not change between runs except by documented addition. For each host the checker requests https://<host>/llms.txt, follows redirects, and counts the host as serving only if the response is HTTP 200, plain text (not an HTML app shell), and starts with a valid Markdown H1. Unreachable hosts (1 in the latest run) are excluded from the base. Raw results: /adoption-stats.json. The 113 positive entries are listed in the verified directory.

Limits: the panel skews toward developer-facing companies, so the headline rate should not be read as an all-of-the-web figure. A file being served is evidence of publication only, not of automatic discovery, AI use, citations, traffic, or SEO effect. See the full methodology and reproducible Observatory release.

Publication patterns

The files in this directory show several publication patterns. A root file, a documentation subpath and a Markdown alternate are all observable publishing choices. They do not by themselves demonstrate the behavior of a named client:

  • Root files are the most common checked location in the fixed panel.
  • Subpath files can scope a map to a documentation area rather than an entire origin.
  • Markdown alternates expose a clean page representation independently of a site map.
  • HTTP response properties such as redirects and content type need checking instead of being inferred from a URL.
  • Generated files still need periodic review: a syntactically valid file can point to stale or irrelevant pages.

Verified directory

The directory lists checked public files and links to the original response. Inclusion means a dated file observation only. It does not mean that a company endorses this site, uses every optional feature, or has a commercial relationship with llmtxt.info.

  • Source link: every directory record links to the file that was checked.
  • Observation date: directory entries must be read with their check date.
  • Scope: a file response says nothing about a domain's crawl rules or search performance.

Client claims: evidence and limits

A published file does not establish that a tool reads it automatically. Client support must be labeled from provider documentation or a narrow reproducible test. The current matrix uses these categories instead of an unqualified native-support list:

  • Provider documented: the vendor describes a specific workflow and its trigger.
  • User-triggered workflow: a user or developer chooses to provide a URL or file.
  • Community implementation: a library parses the convention without representing all clients.
  • Unknown: no support claim is made where the behavior is not documented or reproducibly tested.

How to read the trend

The tracker is useful because the cohort stays fixed. Read a future increase carefully:

  1. It is a panel change only. The rate can move without representing the whole web.
  2. It is publication evidence. It does not turn into evidence of client consumption.
  3. It needs method continuity. The panel, qualification rules and measurement date make releases comparable.

Check a public file with the live verifier, then use the validator for full local diagnostics.

Related pages

Sources