Not sure where to start? Take a guided tour.
AI Site Grader
AI Query Scanner
llms.txt Generator
In short
Four findings, strongest first- 13.3%
of 37,894 AI-cited domains publish llms.txt. Among the 50 most-cited, it’s 6%.
Figure 2 - p = 0.85
for the top 5,000 cited domains: no significant citation difference between sites with and without the file.
Figure 3 - 3 and 3
median citations, with and without llms.txt. The means are 6.8 and 6.7.
Figure 3 - 24.1%
of SaaS and developer tool domains have one, against 1.5% of government and academic sites.
Figure 4
The most-cited domains mostly skip it.
13.3% of the 37,894 AI-cited domains we scanned publish an llms.txt file. Among the 50 most-cited sites, only 6% do.
Move down the citation rankings and adoption rises, peaking at 16.1% in the top 5,000 before easing off across the long tail. That pattern doesn’t explain why sites adopt the file, or show any effect on citations.
The most-cited domains are the least likely to publish llms.txt
Share of domains with an llms.txt file, by citation rank
| Group | Value |
|---|---|
| Top 50 (3 of 50) | 6% |
| Top 100 (7 of 100) | 7% |
| Top 250 (34 of 250) | 13.6% |
| Top 500 (72 of 500) | 14.4% |
| Top 1000 (153 of 1,000) | 15.3% |
| Top 2500 (398 of 2,500) | 15.9% |
| Top 5000 (803 of 5,000) | 16.1% |
| Top 10000 (1,571 of 10,000) | 15.7% |
| Top 25000 (3,424 of 25,000) | 13.7% |
| All scanned (5,035 of 37,894) | 13.3% |
The two citation tests disagree.
The published snapshot reports averages of 6.8 citations with llms.txt and 6.7 without it, and the same median, 3. The top 5,000 most-cited domains return p = 0.85 in the Mann-Whitney U test, which isn’t statistically significant at the 0.05 level.
The full 37,894-domain sample returns p < 0.001, with a reported effect size of r = −0.065. That test is significant, while the effect is small. It must not be described as the same null result.
The same median, and two tests that disagree
Published citation medians for domains with and without the file, then the two cohort tests.
| Group | median citations |
|---|---|
| With llms.txt | 3 |
| Without llms.txt | 3 |
- Top 5,000 cited domainsMann-Whitney U. No statistically significant difference in this subset.
- p = 0.85
- Full sample, 37,894 domainsSignificant, with a small reported effect (r = −0.065). The JSON rounds this p-value to zero.
- p < 0.001
Adoption follows the tech sector.
llms.txt adoption is led by SaaS and developer tools, at 24.1%. Government and academic sites sit at just 1.5%, while review sites and reference wikis are at 0%.
Uneven adoption leaves room for selection bias. The study doesn’t separate sector, site size, content quality or existing citation exposure from the decision to publish llms.txt.
Developer tools lead; reference and review sites have none
Share of scanned domains in each category with an llms.txt file.
| Group | Value |
|---|---|
| SaaS and Developer Tools (97 of 403 domains) | 24.1% |
| E-commerce (10 of 55 domains) | 18.2% |
| News and Media (52 of 332 domains) | 15.7% |
| Social Platforms (84 of 536 domains) | 15.7% |
| Government and Academic (9 of 581 domains) | 1.5% |
| Reference and Wiki (0 of 36 domains) | 0% |
| Review Sites (0 of 39 domains) | 0% |
Visibility scores look alike too.
Using Trakkr’s visibility score, which combines presence, rank, mentions and sentiment across AI models, brands with llms.txt have a median of 23.15 against 23.55 without it.
The medians are close. The comparison has no confidence intervals or adjusted causal estimate, so it can’t show that the difference is noise, or that llms.txt has no effect on recommendations.
Brands with and without the file score about the same
Trakkr visibility score, 0 to 100, for brands with both an audit and visibility reports.
Median visibilityWith llms.txt 23.1 · without 23.6
Average visibilityWith llms.txt 27.8 · without 26.3
What this means.
Read these results as a dated observational snapshot, with separate tests for the full sample and its most-cited subset.
Separate the two cohorts
The top-5,000 test is not significant. The full-sample test is significant with a small reported effect. Reporting either result alone loses useful context.
Association does not establish cause
Sites choose whether to publish llms.txt. Sector, size, content and existing citation exposure may differ between groups. This study can’t isolate the effect of adding the file.
The scan does not measure model behaviour
Finding a file on a domain doesn’t tell us whether an AI system fetched or used it. Citation counts alone can’t separate training, search or retrieval.
Measure the workflow you care about
If you publish llms.txt for documentation or retrieval, test that workflow directly. A controlled or longitudinal study would be needed to estimate changes in citations.
