Last updated September 2026.
Two numbers, one file, a 6x gap
On Aug. 3, 2026, Digital Applied published a fixed-panel reading of llms.txt adoption at 51.8% (Digital Applied). Weeks earlier, Rankability’s monthly tracker had logged the same metric at 8.7% for June 2026 (Rankability).
Both numbers are real. Both come from disclosed methodology. They still land six times apart.
That gap is the story. Marketers quote whichever figure fits the point they want to make, and most never say which panel it came from. It also says something bigger: “AI readiness” benchmarking is still immature enough that two well-documented studies can land six times apart on the same file, in the same month.
The four criteria below decide which number generalizes, and for what claim.
llms.txt is a plain-text file, proposed in 2024, that sits at a site’s root and gives large language models a curated map of a site’s key pages. It answers one narrow question: did a domain publish the file? It says nothing about whether an AI engine reads it, or whether reading it changes what gets cited. We come back to that split further down.
The llms.txt evidence table
Four separate readings exist as of Sept. 2, 2026. Treat “as of” literally. Adoption moves month to month, and this table gets a new row whenever either tracker publishes again.
| Panel | Sample frame | As of | Adoption rate |
|---|---|---|---|
| Rankability, Tranco top 1,000 | Traffic-ranked, general web | June 2026 | 8.7% (87 of 1,000); 15.8% on the 549 reachable roots |
| Rankability, Tranco top 10,000 | Traffic-ranked, general web, wider | June 2026 | 5.6% (558 of 10,000) |
| Rankability, Tranco top 1,000, earlier reading | Same tracker and list, a year prior | June 2025 | 0.3% (3 of 1,000) |
| Digital Applied / llmtxt.info, fixed panel | Curated, 219 hosts weighted toward developer tools, SaaS, and AI companies | Aug. 3, 2026 | 51.8% (113 of 218 reachable) |
| SE Ranking, general domain study | Roughly 300,000 domains, broad | Nov. 7, 2025 | 10.13% |
Read down the first two rows and a trend is visible on its own. Adoption at the top of the traffic-ranked web grew from about 0.3% in June 2025 to 8.7% a year later. That is real growth, just measured against a much smaller base than the developer-panel figure below it.
That last row also breaks out by traffic tier, which matters for a claim later on. Low-traffic sites, 0 to 100 monthly visits, came in at 9.88%. Mid-traffic sites, 1,001 to 5,000 visits, came in at 10.54%. High-traffic sites, 100,001 or more visits, came in at 8.27% (SE Ranking).
Grading the two headline studies
Rankability and Digital Applied are the two studies driving the 6x headline. Here is how each one grades on the criteria that decide whether a number generalizes.
| Criterion | Rankability (Tranco top 1,000) | Digital Applied (219-host panel) |
|---|---|---|
| Sample frame | A | D |
| Reachability handling | A- | B |
| Traffic-tier weighting | B | F |
| Reproducibility | A | C |
Sample frame. Rankability draws from Tranco, a research-grade ranking that blends several traffic sources into one ordered list of the open web. That is close to the best available proxy for “the general web,” so a number pulled from it generalizes.
Digital Applied’s panel is openly curated. It is 219 hosts, hand-picked and weighted toward developer tools, SaaS, and AI companies, the exact cohort most likely to already publish the file. That makes it a fine sample for describing that cohort and a poor stand-in for the web.
Reachability handling. Rankability requested both /llms.txt and /llms-full.txt over HTTPS, retried failures, and still only reached 549 of 1,000 hosts. Rather than guess at the rest, it reports two numbers side by side: 8.7% across all 1,000 sites and 15.8% across the 549 that answered.
Digital Applied’s panel reached 218 of 219 hosts, a 99.5% rate, and reports that count cleanly. The grade sits a step lower only because near-total reachability is what you get from hand-picking known, live companies, not from solving the harder problem Rankability actually faced.
Traffic-tier weighting. Rankability shows real tier effects without isolating traffic size directly. Adoption drops from 8.7% at the top 1,000 to 5.6% at the top 10,000, and its category breakdown ranges from 36.4% in technology down to zero in government.
Digital Applied’s panel carries no traffic dimension at all. It is sorted by sector, not by visits, so a reader cannot tell whether 51.8% describes large companies, small ones, or an even mix of both.
Reproducibility. Rankability names its exact Tranco list, dates the crawl, publishes its pass and fail rules, and offers the dataset as a free API under a Creative Commons license. Anyone can rerun it and get the same answer.
Digital Applied’s 51.8% figure traces back to a fixed panel first published by llmtxt.info, which Digital Applied cites rather than re-runs on a disclosed schedule. The host list and sector split are named, which earns partial credit, but there is no open dataset and no confirmed monthly cadence behind it the way there is for Rankability.
Which number is defensible for which claim
Match the number to the question, not the other way around.
Asking how much of the open web has adopted llms.txt? Cite Rankability’s 8.7%, or 15.8% if you are counting only sites that answered. That is the traffic-representative reading, dated June 2026.
Asking whether llms.txt has become routine among developer tools and SaaS companies specifically? Cite Digital Applied’s 51.8%, dated Aug. 3, 2026. That is exactly the population its panel samples, and the figure holds up for it.
Asking whether the biggest companies are leading adoption? Neither headline number answers that. SE Ranking’s traffic-tier data does, and it says no: high-traffic sites adopted at 8.27%, below mid-traffic sites at 10.54%.
What you should never do is average the two headline numbers, or repeat one without naming its panel. “8.7%” and “51.8%” are both true. Neither one is the adoption rate for the web, because no single number like that exists to report.
The 6x headline hides a smaller, real gap
The 6x spread compares two different denominators. Rankability’s 8.7% divides by all 1,000 sites in its sample, including 451 that never resolved. Digital Applied’s 51.8% only ever divides by reachable hosts, because its panel barely has any unreachable ones.
Compare both studies on reachable hosts only, and the gap shrinks to 15.8% against 51.8%, a 3.3-fold difference instead of six. That is still a real gap. It is also the honest one.
The wider 6x figure is not wrong. It is what happens when a conservative, all-hosts denominator meets a nearly-all-reachable one in the same headline. Report the reachable-only comparison when the point you are making is about sample composition, not reachability.
The null finding both headlines skip
Adoption studies measure whether a file exists. They do not measure whether it does anything. The one study built to test that second question found nothing.
SE Ranking’s 300,000-domain analysis ran an XGBoost model to predict AI-citation frequency from a set of domain factors, including llms.txt presence. Removing the llms.txt feature improved the model’s accuracy. A factor that makes a prediction model worse when you include it is not a weak signal. It is noise (SE Ranking).
Neither the Rankability tracker nor the Digital Applied panel measured citations at all. Both report presence only. The honest read across all three studies is this: adoption is real and growing inside developer-heavy segments, and nothing published so far shows that publishing the file changes who gets cited.
Where visibility trackers fit in
None of the tools this lab tracks turn llms.txt presence into a scored visibility factor. Knowatoa, SE Visible, and Profound all track citations and mentions directly, and none bundles a full site audit into its own standalone product. SE Visible leaves that job to the wider SE Ranking suite, and Knowatoa and Profound both point to separate tools for it.
Temso is the one platform here that folds a site auditor into the same subscription that tracks citations. That is a workflow difference, not a scoring one. llms.txt presence still does not move any tool’s visibility number, on this roster or elsewhere.
Changelog
- Sept. 2, 2026: First publication. Logged Rankability’s June 2026 Tranco reading (8.7%), Digital Applied’s Aug. 3, 2026 panel reading (51.8%), and SE Ranking’s Nov. 7, 2025 traffic-tier study.
Check every adoption number against its sample before you repeat it. See how the tools on our full LLM visibility ranking measure citations directly, using the same method-first standard this table applies to a study.