We Publish an llms.txt. The 2026 Data Says Almost Nothing Reads It
This site serves an llms.txt and an llms-full.txt. We should therefore be straight about what the evidence says they do, which as of 2026 is: not much.
The numbers
An SE Ranking study of 300,000 domains put adoption at roughly 10.13%. That sounds like real traction until you look at where the files are. Among the fifty most-cited domains in AI answers, only one had the file at all. Adoption is concentrated among smaller sites that are optimizing hopefully, not among the sites that actually get cited.
Usage is worse than adoption. Monitoring across more than 500 million AI bot visits over a 90-day window found only 408 requests that targeted llms.txt directly. An Ahrefs study of 137,000 sites found that 97% of llms.txt files received zero traffic in May 2026.
And the vendors have not signed up. As of Q1 2026, no major AI company (OpenAI, Google, Anthropic, Meta or Mistral) has publicly committed to reading or acting on llms.txt in production. Google has been explicit in the other direction: John Mueller compared it to the keywords meta tag, which is about as unambiguous a dismissal as that organization issues.
Why the idea keeps selling anyway
The proposal is genuinely reasonable on its face. Markdown is cheaper to parse than a JavaScript-heavy page, a curated index is friendlier than a crawl, and robots.txt set a precedent for a plain-text file at a well-known path that everyone eventually honored. It should work.
The gap is that robots.txt solved a problem crawler operators had: bandwidth and legal exposure. llms.txt solves a problem publishers have. Nobody on the receiving end is motivated to implement it, and the engines already extract semantic structure during normal crawling without the detour through an extra file.
So why do we still publish one
Three honest reasons, none of which is "it improves visibility."
It is nearly free. Ours is generated from the same article data that builds the sitemap; maintaining it costs one incremented counter per publish. A file that costs nothing and might matter later is a defensible bet in a space where the standards are two years old at most.
It is a useful forcing function. Writing a one-paragraph summary of every article, in one place, surfaces which pages have no clear claim in them. That exercise has value regardless of who reads the output.
And it is directly fetchable. If someone points an agent at this domain and asks what we cover, the file is a clean answer. That is a real use, just a much smaller one than the marketing around it implies.
What to do instead, if you only have time for one thing
Make sure your pages render their content without JavaScript, that your robots.txt does not block the AI crawlers you actually want, and that your key claims are stated as specific, sourced sentences a model can lift. The evidence says topical authority and conventional crawlability drive citations far more than any file at a well-known path. Publish the llms.txt if it is cheap. Just do not put it on the roadmap ahead of anything on that list.