Original research · September 2026
9.3% of the top 1,000 domains serve a confirmed llms.txt or llms-full.txt file. We checked the top 10,000 to see how far adoption extends.
Checked September 18, 2026 · Tranco list V3YPN · All domains stay in the denominator
How many domains have adopted llms.txt?
Confirmed adoption means we found at least one of the two files. A blocked or inconclusive check stays unknown.
Top 1,000
Tranco domains93 confirmed adopters / 1,000 domains
- Confirmed
- 93
- No file found
- 433
- Unknown
- 474
Top 10,000
Tranco domains830 confirmed adopters / 10,000 domains
- Confirmed
- 830
- No file found
- 4,959
- Unknown
- 4,211
Which files do sites publish?
The short index and full-content file are measured separately. “Both files” is included in each file count, so these bars should not be added together.
Bar lengths compare file counts within this sample.
Bar lengths compare file counts within this sample.
Adoption over time
Three published observations of the top 1,000. The September scan uses stricter validation and a new ranking list.
Separate observations, not a continuous monthly series. Changes in the sample and method prevent a like-for-like growth calculation. View the earlier dataset.
Adoption by category
Directional estimates for the top 10,000. Categories come from domain-name rules and a curated domain list; they are not supplied by Tranco or manually verified for every site.
Shown: the eight highest rates among named categories with at least 30 domains. Bars use a 0–100% scale. The download includes every category.
Examples of confirmed adopters
Highest-ranked domains where the scan confirmed at least one file. Rank refers to this study’s Tranco list.
How we ran the study
- Fix the sample. Download the top 10,000 domains from Tranco list V3YPN, dated 2026-09-17. The top 1,000 are a subset of the same scan.
- Check both files. Request
/llms.txtand/llms-full.txtover HTTPS with an identified Rankability crawler. Follow redirects, try apex and www hosts, and retry transient failures once. - Validate the response. Require HTTP 200, a text-compatible response and recognizable Markdown structure. Exclude HTML, empty responses and obvious text errors. Ambiguous content remains unknown. Inspect up to 256 KiB of content, with bounded downloads for large files.
- Check apparent positives. Request a nonexistent text-file URL on each responding host. Catch-all Markdown responses and inconclusive controls are excluded from confirmed adoption and marked unknown.
- Keep uncertainty visible. Count a domain once if either file is confirmed. Divide by the entire sample. A domain is “no file found” only when both files are absent on both host variants; blocks, timeouts and inconclusive validation remain unknown.
What this study does—and does not—measure
This is a snapshot of root-level file availability, not a full specification audit or proof that an AI platform uses the file. Files hosted only on documentation subdomains or subpaths can be missed. Rankings include infrastructure domains as well as public websites. Network conditions, redirects and automated validation affect the result. Redirects count as availability; file names do not guarantee complete or distinct content.
The llms.txt proposal describes Markdown information for agents using a website. Publishing one does not establish search-ranking improvements, AI citations, or training permissions.
Earlier results are retained as historical benchmarks. Their looser content checks, host handling and ranking lists differ from this run. No month-over-month growth claim is made, and no automatic monthly refresh is implied.
Explore the study data
Inspect all 10,000 domain verdicts, file counts and sample totals. Cite Rankability and the scan date when reusing the dataset.
Dataset: CC BY 4.0 · Scan completed September 18, 2026