Does an llms.txt file help you show up in AI search?
No measurable effect has been found, across two large studies. If an agency sold you one as a package, that is worth knowing before you buy anything else from them.
llms.txt is a proposed text file you put at the root of your website, meant to tell AI models
what your site is about — roughly the idea of robots.txt, but aimed at language models. It has
been sold hard for the last year, often as the centrepiece of an "AI optimization" package.
Here is what the measurement says.
Study one: no relationship across 300,000 domains
A study covering roughly 300,000 domains looked for a correlation between having an llms.txt
file and being cited more often by AI assistants. It found no measurable relationship. Not a
weak one — none that could be distinguished from noise.
Study two: almost nothing reads them
Ahrefs analysed 137,000 sites and looked at actual traffic to the file itself:
- 97% of
llms.txtfiles received zero traffic at all - of the remaining 3% that got any hits, 96% of that traffic was non-AI bots — crawlers, scanners, monitoring tools
So the file mostly is not fetched, and when it is fetched it is mostly not being fetched by the thing it was written for.
And the search engine says it isn't needed
Google has stated publicly that llms.txt is not required and is not something its systems
depend on.
Should you delete it?
No. It costs nothing to leave in place, it is not harmful, and the proposal may yet get adopted. This is not a warning about a dangerous file.
It is a warning about what it means when somebody sold it to you. If an agency's AI-search
package leads with llms.txt setup, they have sold you the one deliverable in this field that is
cheap to produce, easy to demonstrate, and demonstrably without effect. It is worth asking what
else in that package was chosen the same way.
What the same research says does matter
The measurement that does hold up points somewhere else entirely. Whitespark's study of 540 local queries found that about 60% of the citations inside AI answers point at third-party sites — directories, roundups, local press, forum threads — rather than at the business's own website.
That is the uncomfortable finding for anyone selling on-site fixes, because it means the majority of the signal is not on a page you control. Getting accurately represented on the sources that actually get cited is slower, less demonstrable, and considerably more useful than adding a file to your root directory.
Sources: 300,000-domain analysis of llms.txt and AI citation frequency; Ahrefs analysis of 137,000 sites; Whitespark, "AI Overviews in Local Search" (540 queries). Google's position is stated publicly by its Search team.