Does llms.txt do anything? We tested it
In a controlled test across 120 pages on 14 domains, adding an llms.txt file produced no measurable difference in time-to-first-citation or citation rate on any of six engines. It costs an hour to publish and may help agentic tools that look for it, but it is not a lever on AI visibility.
The claim
llms.txt is a proposed convention: a markdown file at your site root describing what the site is and which pages matter, intended to help AI systems understand and cite you correctly. It has been widely recommended as an AEO essential.
Widely recommended is not the same as tested, so we tested it.
The test
Within our 120-page publishing study, we randomly assigned half the participating domains to publish a well-formed llms.txt naming the study pages and half to publish nothing. Everything else was held constant: same content templates, same internal linking, same publishing cadence, same 60 days of daily sampling across six engines.
The result
| Metric | With llms.txt | Without |
|---|---|---|
| Median time to first citation | 9 days | 9 days |
| Cited within 60 days | 71% | 73% |
| Median citation rate once cited | 22% | 24% |
Every difference is comfortably inside the interval at this sample size. If there's an effect, it's smaller than we can detect with 120 pages, which also means it's smaller than anything that would change a client's outcome.
Why the enthusiasm exists anyway
It's legible. A file you can create, inspect and tick off is enormously more satisfying than "be accurately described on third-party sources over six months". The AEO market has a strong preference for actions that are cheap, visible and attributable, and llms.txt is all three.
Should you publish one?
What we'll keep watching
If a major engine publicly commits to reading llms.txt, this result becomes obsolete overnight and we'll re-run it. Until then, treat published guidance about it — including ours — as provisional.
See this for your brand
Nostimates shows you how your brand shows up across twelve AI engines and search — in a dashboard, or as data in your own tools. Tell us what you want to measure.