Not sure where to start? Take a guided tour.
AI Site Grader
AI Query Scanner
llms.txt GeneratorDoes llms.txt increase AI citations?
This snapshot does not establish a causal increase. The top-5,000 test returns p=0.85, while the full 37,894-domain sample returns p<0.001 with reported r=-0.065.
Different cohorts, different results.
- Top 5,000 cited domains
- Subset of the top 5,000 cited domains; not the full-sample test.
- p = 0.85
- Full sample
- A small reported association (r = −0.065). Full 37,894-domain sample. The public JSON rounds this value to zero.
- p < 0.001
Mann-Whitney U tests · 14 March 2026.
The cohorts give different statistical results; neither measures what adding the file would cause.
The same median.
An open causal question.
Figure 01
Published citation medians
| Group | median citations |
|---|---|
| With llms.txt | 3 |
| Without llms.txt | 3 |
Published means are 6.8 with the file and 6.7 without it. Equal medians do not imply identical distributions. The public data does not specify the cohort behind these citation summaries.
Study aggregates Snapshot: 14 March 2026| Measure | Result |
|---|---|
| Median citations with llms.txt Median citations for domains with llms.txt. | 3 |
| Median citations without llms.txt Median citations for domains without llms.txt. | 3 |
| Average citations with llms.txt Mean citations for domains with llms.txt. | 6.8 |
| Average citations without llms.txt Mean citations for domains without llms.txt. | 6.7 |
| Top-5,000 Mann-Whitney p-value Subset of the top 5,000 cited domains; not the full-sample test. | 0.85 |
| Full-sample Mann-Whitney p-value Full 37,894-domain sample. The public JSON rounds this value to zero. | <0.001 |
| Full-sample reported effect size r Small reported magnitude; not an estimate of causal citation lift. | -0.065 |
Population and method
Built from HTTP scans of 37,894 AI-cited domains, linked to 337,362 citations and 882 citation snapshots in the Trakkr corpus.
Where the evidence stops
This is an observational study. It measures correlation with citation outcomes, not a controlled experiment.
Adoption is uneven by sector, so raw averages can hide category concentration in SaaS and developer tooling.
The top-5,000 test returns p=0.85. The full scanned sample returns p<0.001 with reported r=-0.065. Neither establishes a causal effect. The public aggregates do not include confidence intervals or document the cohort for citation means and medians well enough to recompute the tests.
What to do with this
- Treat llms.txt as an optional housekeeping file, not a primary citation-growth lever.
- Prioritize answer quality, source coverage, and page structure before spending disproportionate effort on llms.txt.
- If you do publish llms.txt, measure discovery and crawl behavior directly instead of assuming it improved citation performance.
This answer matters because it turns a study finding into an operating rule teams can use when they decide what to publish, refresh, or measure next.
Questions about this answer.
Cite this answer.
Source study: llms.txt Effect“This snapshot does not establish a causal increase. The top-5,000 test returns p=0.85, while the full 37,894-domain sample returns p<0.001 with reported r=-0.065.”
Mack Grenfell · Does llms.txt increase AI citations?
The llms.txt Effect · Trakkr Research · Updated 4 September 2026