A Markdown file placed at the root of a site, at /llms.txt, which introduces the site and points to its main pages so an AI assistant can read them directly. The format asks for a level-1 heading and leaves the rest free.

What exactly does this index measure?

The share of .fr sites serving an llms.txt at their root, the median size of those files, and the share of the sites publishing one that also turn an AI crawler away in their robots.txt.

Does a site answering 200 without serving a file count?

It is recorded, and counted apart. Many sites return their home page at any unknown address: the body is read, and a body that does not open on a Markdown heading says the site has no llms.txt.

llms.txt adoption in France

How many .fr sites publish the file AI assistants come to read, and what they tell the AI crawlers next to it.

At the reading of 07/09/2026, 7.9% of the .fr sites that answer serve an llms.txt at their root.

Publish an llms.txt

7.9 %

Median file size

3,878 bytes

Publish an llms.txt and turn away an AI crawler

2.8 %

The llms.txt and the robots.txt of the same site

2.8% of the sites publishing an llms.txt also forbid the root to at least one AI crawler in their robots.txt.

Both files are read on the same domain, in the same conversation, the same morning: that is what allows what a site writes for the assistants to be set beside what it writes for the crawlers that feed them.

The agent-by-agent detail is on the sibling index: AI crawler access to French websites

Week after week

See the figures as a table

Date	Publish an llms.txt
2026-09-07	7.9 %
2026-09-04	7.9 %

Method

The panel: The same panel as the two other indices of the .fr parc: 15,000 domains drawn at random from Afnic's open data file, which lists the 4,590,553 active .fr domains. The draw orders every domain by the sha256 fingerprint of its name, and the published seed is enough to rebuild it.
The cadence: The /llms.txt is asked for every Monday, in the conversation that already opens every domain of the panel. The panel stays the same from one week to the next: a movement in the series is a decision taken by the parc.
The reading: A body counts as an llms.txt when it opens on a level-1 Markdown heading, the one element the format requires. The published size is the median of the files served, in bytes.
The 200s that carry no file: 36.6% of sites answer 200 at this address while serving something else, most often their home page. Those answers are recorded as such and counted apart: they say the site has no llms.txt.
The base: The figures are read on the sites that answer the question: those that serve a file, those that serve something else, and those that answer they have no such file. That is 98% of the panel sites that serve a page.
The API: GET https://api.stileex.xyz/v1/series/fr/llms-txt-adoption/latest

Measured on the same panel, in the same weekly sweep: AI crawler access to French websites