How five llms.txt trackers reported five different adoption rates

Ask five people what share of websites publish an llms.txt file and you get five numbers: 7.4%, 8.7%, 10.13%, 15.8%, 28%. None of them is wrong. Each one answers a different question - a different population, crawled with a different definition of “has the file” - and treating any one of them as the adoption rate is the actual error.

Five counts, five populations#

Tracker	Population	What counted as adoption	Result
Rankability	Top 1,000 sites by Tranco (traffic-based ranking), June 2026 list	HTTP 200 with plain-text content on /llms.txt; soft-404 HTML and empty bodies excluded	8.7% (87/1,000); 15.8% among the 549 domains it could reach a verdict on
SE Ranking	Roughly 300,000 domains, selection criteria not published	File detected present; validity criteria not published	10.13%
Ahrefs	137,210 domains running Ahrefs Web Analytics or Bot Analytics	HTTP 200, confirmed via server logs as real Markdown rather than HTML	28% (about 38,360 domains)
Chris Humphrey	Top 10,000 domains in the Majestic Million, June 2026 crawl	HTTP 200 and a genuine Markdown body; 313 of 1,050 200-responses were soft-404 HTML pages and excluded	7.4% (737 sites); only 55% of those met the full recommended structure
Originality.ai	Continuous monitor of 3+ million sites	Not published in detail	Grew 8.8x, from 4,088 files in June 2025 to 36,120 in May 2026

Why these numbers cannot be averaged together#

The populations barely overlap. A traffic-ranked top 1,000 (Rankability), a link-graph-ranked top 10,000 (Chris Humphrey), a 300,000-domain sample of unstated selection (SE Ranking), and one vendor’s own customer list (Ahrefs) are four different slices of the web, and only one of them - Ahrefs - even claims to include mid-size sites that never heard of the spec. Ahrefs says so itself: “Ahrefs Web Analytics customers skew more technical and SEO-aware than the web at large,” and its own report tells readers to treat the 28% figure as an upper bound, not a web-wide rate. That is the most honest sentence in any of these five studies, and it is the reason Ahrefs’ number is the highest of the five by a wide margin.

“Has llms.txt” is not one definition. Rankability and Chris Humphrey both publish the detail that matters most here: a plain HTTP 200 overcounts, because a large share of sites answer any unknown path with a styled 404 page under a 200 status. Chris Humphrey’s crawl hit exactly that - 1,050 domains returned 200 for /llms.txt, and 313 of them were soft 404s, not files. Only the 737 that remained after excluding those counted toward his 7.4%. SE Ranking and Originality.ai report a number without publishing that filter, so their figures may or may not have the same correction applied - there is no way to check from what they’ve published.

The dates do not line up either. Rankability’s figure is a June 2026 Tranco list processed in August 2026. Chris Humphrey’s crawl ran in June 2026. Ahrefs’ server-log analysis covers May 2026. SE Ranking’s study reads as late 2025. Originality.ai’s 8.8x growth figure is the only one that holds population and method constant across a full year, which is what makes it a trend line rather than a single snapshot - the other four are four different snapshots of four different things, not four points on the same curve.

Why we care#

A single adoption percentage, quoted without its population and its definition, is not a citable fact. If a pitch or a deck says “X% of websites now have llms.txt,” the next question is always which tracker, which population, and which month - and, per this report, the honest answer sometimes has to be “we don’t know, because that tracker didn’t publish it.”

The two most reusable numbers here are the two with published methodology you could rerun yourself. Rankability’s soft-404 exclusion and reachable-subset breakdown, and Chris Humphrey’s Markdown-validation and quality-structure check, are the kind of disclosure that lets a reader verify the claim instead of trusting the headline - which is the same bar this site holds its own reports to.

None of these five trackers measured whether llms.txt does anything. Adoption is a count of files on servers, not a signal that any answer engine reads. For that question, see our companion piece on who actually publishes llms.txt and why - the file’s audience problem is a separate finding from its adoption rate, and conflating the two is its own common error in AI search coverage this year.