Ahrefs pulled logs from 137,000 domains in May 2026 and found that 97% of llms.txt files received zero requests. Not low traffic. Zero. No bots, no humans, nothing. For a file format that's been pitched as the next must-have for AI visibility, that number should stop you mid-sprint.
The audience that showed up isn't the one you were sold
About 28% of domains in the Ahrefs sample published an llms.txt file. Since Ahrefs' customer base skews technical, real-world adoption across the broader web is probably lower. Of the roughly 38,000 domains with valid files, only around 1,100 drew any traffic at all.
Among those files that were fetched, 96% of requests came from bots. That sounds promising until you look at which bots. SEO audit tools accounted for 21% of requests. Unidentified bots took 14%. Web crawlers like Googlebot grabbed 13%. Tech profiling tools like BuiltWith, 11%.
AI retrieval bots linked to ChatGPT and Perplexity? About 1.1% of total requests.
Coding agents (Claude-Code, GPTBot) sent 10% of requests. Training crawlers sent 5%. Assistants, 2%. The bots most directly tied to live AI search and citation barely touched these files. Slackbot fetched llms.txt more often than PerplexityBot did.
An industry studying its own reflection
Here's the detail that should make you uncomfortable: 12% of all llms.txt requests came from tools that audit, scan, or study the file rather than use it. GEO and AEO readiness tools sent 5%. Dedicated scanners and validators sent 3%, which is more than AI retrieval bots and assistants combined. The largest research crawler identified itself as a prompt injection survey.
An entire ecosystem has formed around scoring and cataloging a file format before any meaningful consumer of that format exists. We've built the measurement apparatus for a signal nobody's reading.
Google's John Mueller called llms.txt a "temporary crutch, perhaps to save some tokens" for AI coding tools. Gary Illyes has indicated Google doesn't support it and isn't planning to. SE Ranking's separate analysis of 300,000 domains found no connection between having llms.txt and AI citation frequency. The data and the public statements from major providers point the same direction.
Where llms.txt might still earn its keep
The file isn't worthless everywhere. For documentation-heavy SaaS properties with complex help centers, integration pages, pricing structures, and trust/security content, a curated llms.txt can help coding agents and internal AI tooling summarize vendor info more accurately. That's a real, if narrow, use case.
The key word is curated. Dumping your sitemap into llms.txt defeats the purpose. If you ship one, treat it like a reading list for an agent doing first-pass vendor diligence: pricing, integrations, security posture, proof points. Update it when those pages change. And measure whether anything actually fetches it (your server logs will tell you in about ten minutes).
But for brochure-style SaaS sites hoping llms.txt will unlock AI-driven acquisition? The Ahrefs data says the payoff is negligible right now.
What to prioritize instead
Major AI systems still rely on HTML and crawled documents. The levers that move the needle for AI visibility are the same ones that have always mattered for search: strong content architecture, clean internal linking, structured data, and pages that answer real questions with specifics. llms.txt is a packaging layer on top of that foundation. Without the foundation, the packaging is irrelevant.
If your team is debating whether to invest cycles in llms.txt, here's a quick diagnostic. Pull your server logs. Check if any named AI agent has fetched your existing pages in the last 30 days. If you're getting crawled by GPTBot or ClaudeBot on your HTML content, your pages are already in the pipeline. If you're not getting crawled at all, llms.txt won't fix that problem.
One more thing worth watching: Ahrefs flagged a crawler studying llms.txt files as a prompt injection vector, since agents tend to trust ingested content. If your CMS auto-generates these files, review what's in them. A file nobody reads is harmless. A file an agent trusts implicitly is a different risk profile.
Every figure in the Ahrefs report is a ceiling, not a floor. The study measured requests, not whether bots acted on what they fetched. The real utilization rate could be even lower. For a file format that's consumed marketing ops bandwidth across thousands of SaaS teams this year, that's the number worth sitting with.
