llms.txt won't save you: what the 2026 studies actually found.
short answer: no measurable evidence that llms.txt improves your AI citation rate, as of mid-2026. If someone is selling you an "llms.txt setup" as the fix for your AI visibility problem, you're paying for a file that the two companies who'd actually use it, OpenAI and Anthropic, have never confirmed they check.
llms.txt is defined as a proposed markdown file at a site's root that lists its most important pages in a form AI systems can read quickly. It was proposed in 2024 as an AI-readable equivalent of a sitemap. As of mid-2026 no major AI company has confirmed using it.
what llms.txt is supposed to do
Proposed by Jeremy Howard in 2024, llms.txt is a markdown file at your site's root meant to act like a curated, AI-readable summary of your most important pages, solving the problem that a language model's context window is too small to read an entire website at once. It's a reasonable idea on paper. The question is whether any AI system actually uses it. (For the neutral explainer of the format itself, see what is llms.txt.)
what the actual tests found
Semrush ran llms.txt on one of its own sister properties, Search Engine Land, specifically to test whether it produced any meaningful lift in AI visibility or traffic. It found no correlation between implementing the file and improved AI-result performance.
SE Ranking went further, running a structured, multi-step analysis combining traditional statistics with machine learning across a broader dataset, testing whether llms.txt adoption predicted how often a domain gets cited by LLMs. Both the statistical analysis and the machine learning model showed no effect.
Removing llms.txt as a variable from the model actually improved its predictive accuracy, meaning the file wasn't just neutral, it was adding noise.
And it isn't just third-party testing. Google's John Mueller has said directly that llms.txt is "not done for search," calling it a temporary crutch, at most useful for AI coding tools parsing developer documentation, not something most sites need to worry about. He's compared it to the old keywords meta tag: something a site owner claims about their own content, with no confirmation any AI system is actually reading it. He's also pointed out that server logs on the sites he's checked show AI crawlers don't even request the file.
| source | what they found |
|---|---|
| Semrush (Search Engine Land test) | no correlation between llms.txt and AI visibility |
| SE Ranking (statistical + ML analysis) | no effect on citation frequency; removing it improved model accuracy |
| Google (John Mueller) | AI services haven't confirmed using it; compares it to the keywords meta tag |
so why does everyone keep talking about it
Because it's an easy thing to sell. It's a single file, it takes an afternoon to write, and it feels like doing something concrete about a confusing problem. None of that means it works. The llms.txt directory that tracks adoption shows plenty of companies have added the file, adoption isn't the same as effect.
There's a version of this that's genuinely useful: writing a clear, accurate summary of what your business does is good practice regardless of whether any crawler reads the specific file it's saved in. But that's true of your homepage copy too, and your homepage copy is something Google and every AI crawler unambiguously does read, assuming they can reach it at all.
what to actually spend that afternoon on instead
If you were about to spend an afternoon writing an llms.txt file, spend it here instead, in order of actual, tested impact:
- Confirm AI crawlers can reach your site at all. This is the fifteen-minute audit that matters more than anything else in this post, and it's covered in full in the crawler audit.
- Add real structured data. Organization schema on your homepage, FAQ schema on question-style pages, Product schema on offers. Unlike llms.txt, this is a format search engines and AI systems have publicly confirmed they use.
- Write the one clear sentence about who you are and repeat it everywhere, site, bio, guest posts, podcast intros. This is the actual foundation the how chatgpt decides what to recommend framework is built on, and it's tested, not speculative.
None of this is about avoiding new tools. It's about spending limited time on the thing that's been shown to move the number, not the thing that's easy to sell. Where this fits in the bigger picture, and how it differs from classic search work, is covered in geo vs seo.
want the plain-text setup that's actually been shown to matter?
structured context files a model reads reliably, not a speculative sitemap. the ai builder toolkit is the working set of skills and context files behind everything published here, built to be installed and used the same afternoon you were about to spend on llms.txt.
get the toolkitfrequently asked questions
Does llms.txt improve AI search visibility?
There is no published evidence that it does. Semrush tested it on a site it owns and found no correlation with AI visibility. SE Ranking ran a statistical and machine-learning analysis and found no effect on citation frequency, noting that removing the variable improved model accuracy.
Do AI companies actually read llms.txt?
None have confirmed that they do. Google's John Mueller has said publicly that llms.txt is not used for search and compared it to the old keywords meta tag, noting that server logs show AI crawlers do not request the file.
Should I create an llms.txt file anyway?
Only if it costs you nothing. Several major developer platforms publish one as a routing layer for AI coding agents, which is a legitimate use. Paying someone to create one as an AI-visibility service is not supported by the available evidence.
What should I do instead of llms.txt?
Three things, in order: confirm AI crawlers can reach your site at all, add real structured data such as Organization and FAQPage schema, and state one clear positioning sentence identically everywhere. All three are formats search engines and AI systems have confirmed they use.
Is llms.txt the same as robots.txt?
No. Robots.txt controls crawler access and is honored by compliant crawlers. Llms.txt contains no directives at all, it is a suggested reading list, and it neither permits nor prevents anything.
written by elisabeth hitz, certified in anthropic's ai fluency program (framework & foundations, and ai capabilities & limitations), plus claude 101 and claude cowork. sources: semrush, llms.txt testing on search engine land · se ranking, llms.txt statistical and machine-learning citation analysis · google's john mueller, public statements on llms.txt via bluesky and reddit, as reported by ahrefs, semrush, bluehost, and se ranking · jeremy howard's 2024 llms.txt proposal.