Last updated September 2026.
Two studies, one file format, two very different numbers
On August 3, 2026, Digital Applied published a fixed-panel measurement of llms.txt adoption at 51.8% (Digital Applied). Weeks earlier, Rankability’s monthly tracker had put the same metric at 8.7% for June 2026 (Rankability). That is a 6x spread on a number marketers were already quoting as if it were one fact.
Both numbers are real. Both come from disclosed methodology. Neither is wrong. The reason they land so far apart is the panel each study sampled, not the file format itself.
llms.txt is a plain-text file, proposed in 2024, that sits at a site’s root and gives large language models a curated, markdown-formatted map of a site’s key pages. It plays a role for AI crawlers similar to what a sitemap plays for search engines, though no major AI engine has confirmed it reads or requires the file. Whether publishing one does anything for citations is a separate, unresolved question. This piece covers adoption, not effect.
The two panels, side by side
| Rankability | Digital Applied | |
|---|---|---|
| Panel | Tranco top 1,000, ranked by traffic | Fixed 219-host list, developer tools, SaaS, and AI companies |
| Panel type | Broad, traffic-ranked, refreshed monthly | Curated, hand-picked, fixed |
| Measurement date | June 2026 (Tranco list dated June 23, 2026; published August 23, 2026) | August 3, 2026 |
| Reachable hosts | 549 of 1,000 (54.9%) | 218 of 219 (99.5%) |
| Headline adoption rate | 8.7% of the full 1,000 | 51.8% of 218 reachable hosts |
| Rate on reachable hosts only | 15.8% | 51.8% |
| Top-10,000 comparison | 5.6% | Not measured |
The reachability gap alone explains part of the spread. Rankability’s broad list includes domains that time out, block crawlers, or no longer resolve; only 54.9% of its 1,000 domains were reachable at all. Digital Applied’s curated list, picked from live developer and SaaS companies, resolved at 99.5%.
But reachability is not the whole story. Even scored only on reachable hosts, Rankability’s broad panel lands at 15.8% adoption, less than a third of Digital Applied’s 51.8%. The remaining gap is who each panel contains.
Why the panels disagree
Rankability samples the Tranco top 1,000, a traffic-ranked list dominated by retail, media, government, and consumer platforms. Most of those domains have no product team thinking about AI crawler files, and Rankability’s own category breakdown shows why: the technology sector leads the top 1,000 at 36.4% adoption, well above the 8.7% blended average, because non-technology categories pull the number down.
Digital Applied’s panel is not a traffic sample at all. It is a hand-picked list weighted toward developer tools, SaaS platforms, and AI companies, the exact segment already publishing changelogs, API docs, and markdown-first content. That is close to a best-case sample for llms.txt adoption, not a representative one.
Neither approach is flawed. A traffic-ranked panel answers “how much of the open web has adopted this,” and a curated panel answers “how far has adoption gone inside the segment most likely to care.” Those are different questions with different correct answers.
Even Rankability’s own numbers shrink as the sample widens. Adoption sits at 8.7% across the top 1,000 domains but drops to 5.6% across the top 10,000. A larger, more general sample pulls the rate down further, not up, because it adds more domains outside the technology sector that has driven most of the adoption seen so far.
What this means
If you see an llms.txt adoption number without a panel description attached, treat it as incomplete. “8.7%” and “51.8%” are both true statements about different populations, and neither one is “the” adoption rate for the web.
Before you repeat any adoption stat, ask three questions: How many domains were in the panel? Was the list traffic-ranked or hand-picked? What date was it measured? A number that cannot answer those three is not safe to cite as a general figure.
The bigger point sits underneath the number itself. Adoption studies measure whether a file exists, not whether it changes anything. Neither study checked whether pages with an llms.txt file got cited more often by ChatGPT, Perplexity, or Google AI Overviews than pages without one. That question stays open, and any claim that llms.txt adoption “helps AI citation” is getting ahead of what either study actually measured.
This is also why watching what gets cited matters more than watching what gets published. A file sitting at a site root tells you a team took a step. It does not tell you whether an AI engine ever read it, or whether a competitor’s plain page without one still won the citation. Among the platforms on our LLM visibility benchmark, neither Profound nor Temso publishes llms.txt presence as part of its scoring. Both track citations and mentions directly across ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot, which sidesteps the whole adoption-rate debate by measuring the outcome instead of the input.
An adoption rate tells you what a sample of sites did. A citation record tells you what an AI engine actually did with the page. See how our measurement protocol handles that second question, then check where your own domain stands in the full LLM visibility ranking.