A study examining roughly 300,000 domains found no measurable relationship between having an llms.txt file and how often a domain gets cited in AI answers. Mainstream AI search systems do not appear to meaningfully rely on the file at this stage. Adding one is still low-risk, but there is no evidence it earns you visibility.
Key takeaways
- No measured effect at scale. Across ~300,000 domains, presence of llms.txt did not correlate with citation frequency.
- This is a null result, not an absence of research. Someone actually checked, and found nothing — that is different from nobody having looked.
- The mechanism is plausible; the evidence isn't there yet. A structured summary file for AI systems is a reasonable idea. Reasonable and proven are not the same claim.
- Low cost still justifies adding one — but as a hedge against future adoption, not a current optimisation with a demonstrated return.
- This site has one (/llms.txt), on that same "low-cost hedge" basis, not because it is expected to move anything today.
What llms.txt is supposed to do
llms.txt is a plain text file placed at a site's root, similar in position to robots.txt, intended to give AI systems a structured summary of what a site covers and which pages matter most. Where robots.txt states what crawlers may access, llms.txt is meant to state what is worth reading and how to interpret it.
The idea is reasonable on its face: large language models have limited ability to crawl an entire site the way a search engine does, so a short, structured pointer file could plausibly help a retrieval system prioritise the right pages. That reasoning is why the file spread quickly through SEO and AEO content in 2025 and 2026 — it sounds like it should work.
- Null result
- A study outcome where the measured effect is statistically indistinguishable from no effect at all. It is not the same as "nobody tested it" — a null result means the test happened and came back empty, which is a meaningfully different, and usually more informative, finding than silence.
What the study actually found
Examining roughly 300,000 domains, researchers found no relationship between the presence of an llms.txt file and how frequently a domain was cited in major AI-generated answers. The dataset suggests that even where a model retrieves the file, it does not appear to change citation behaviour at scale.
That is a large enough sample that "we just haven't found the effect yet" is a weaker explanation than "there currently isn't one to find." It does not prove AI systems never use the file — some may fetch it for narrower purposes not captured by a citation-frequency study — but it directly contradicts the specific claim most content repeats: that adding llms.txt improves how often you show up in AI answers.
| Common claim | Evidence status |
|---|---|
| "Adding llms.txt increases AI citations" | Contradicted — no correlation found across ~300k domains |
| "AI crawlers fetch llms.txt when present" | Partially observed — fetching happens; using it for citation decisions is unconfirmed |
| "It's a quick, low-risk file to add" | Accurate — genuinely low effort, no technical downside |
| "It will meaningfully improve AI visibility now" | Unsupported — the specific mechanism most articles imply is not demonstrated |
Why so much content recommends it anyway
A few incentives line up to keep this claim circulating regardless of the evidence.
- No downside to recommending it. Telling someone to add a free file costs the writer nothing if it turns out not to help.
- A plausible mechanism reads as a finding. "AI systems need a way to know what your site covers" sounds like a conclusion, but it is a hypothesis, not a measurement.
- Early-adopter narratives are more shareable than null results. "We tried this and got nothing" gets written far less often than "here's the next big AI SEO tactic," even when the former is the more useful thing to know.
- Nobody is checking the checkers. Most llms.txt content cites other llms.txt content, not primary data — a citation chain that never touches a measurement.
Should you add it anyway
Yes, on a low-cost-hedge basis, not as an optimisation. It takes minutes to write, carries no technical risk, and if AI systems do start weighting it meaningfully in the future, you are already covered. Just don't expect a visibility change from adding it today, and don't let it substitute for the things that are actually documented to matter.
The things worth prioritising instead, in order: server-rendered HTML that a simple crawler can actually read, self-contained answer paragraphs under question headings, and consistent author/entity signals — all covered in the technical SEO pillar guide. None of those are speculative; llms.txt, on current evidence, still is.