llms.txt adoption grew 8.8x in a year. Server-log data shows almost none of it is being read by the AI systems it's meant for. Here's what the file actually does, and why we wouldn't charge you to build one.
llms.txt is a small text file, placed at yoursite.com/llms.txt, that's supposed to give AI systems a clean, curated summary of what your site is and where the important pages are. It was proposed in 2024 by data scientist Jeremy Howard as a way to help language models navigate a site without having to read the whole thing. Since then it's become a small industry: agencies and plugins offering to generate one for you, sometimes pitched as an AI-visibility must-have.
The data says something different. Adoption of llms.txt is genuinely growing fast, and almost none of that growth is being read by the AI assistants it's meant for.
The headline numbers. Originality.ai's tracking study (covering more than 3 million websites) recorded 4,088 llms.txt files in June 2025. By May 2026 that had grown 8.8x to 36,120 (38,980 including the related llms-full.txt and ai.txt formats). Separately, Ahrefs pulled server-log data from 137,000 domains for the same period and found that 97% of llms.txt files received zero requests in May 2026. Where requests did happen, AI retrieval bots, the systems that actually answer user questions in ChatGPT, Perplexity and similar, accounted for just 1.1% of them. The single biggest requester of llms.txt files was SEO audit tools (21.7%), not AI.
| Metric | June 2025 | May 2026 |
|---|---|---|
| llms.txt instances tracked | 4,088 | 36,120 (8.8x growth) |
| llms-full.txt instances | 23 | 2,463 (107x growth) |
| ai.txt instances | 4 | 397 (99x growth) |
So the file is spreading. What it isn't doing is getting read by the AI systems it was built for. Of the requests logged against llms.txt files in May 2026: SEO audit tools made 21.7% of them, unidentified bots 14.9%, general web crawlers 13.1%. Named AI crawlers barely registered: GPTBot accounted for 4.51% of requests, ClaudeBot 0.80%, DeepseekBot 0.02%. Actual AI retrieval bots, the ones fetching pages live to answer a user's question inside a chat interface, made up just 1.1% of all requests.
Google's own AI search guidance states plainly that llms.txt "won't negatively or positively impact your visibility or rankings" in Search, a position it reiterated in a documentation update on 15 June 2026. OpenAI's crawler documentation doesn't mention llms.txt at all, directing site owners to the standard robots.txt instead. Anthropic takes the same position. John Mueller, former Senior Webmaster Trends Analyst at Google, put it bluntly on Bluesky in mid-2025: "no AI system currently uses llms.txt." All three major AI platforms publish an llms.txt file, but only for their own developer documentation, not as guidance for other sites to follow.
If someone offers to sell you an llms.txt file as a way to get your local business recommended by ChatGPT or Google's AI, the evidence doesn't support the pitch. The file has a real, narrower purpose: it can help AI coding agents or documentation-aware tools navigate a large site with substantial docs or APIs. That's a legitimate use case for a software company. It is not the lever that gets a dental practice or a mortgage broker cited when someone asks AI "who's the best near me."
What actually gets read, per the same body of research this article draws from, is the ordinary stuff: your Google Business Profile, the plain text on your website, and consistent listings across the directories AI already trusts. None of that requires a special AI-only file. It requires the same groundwork that's always mattered, just written and structured so a machine can parse it as easily as a person can.
The adoption figures come from Originality.ai's tracking of more than 3 million sites; the request-log data comes from Ahrefs' analysis of 137,000 domains, both third parties with a commercial interest in SEO tooling, though the methodology is publicly documented and the findings are consistent with earlier, independent reporting going back to 2025. Server logs can undercount requests from AI systems that route through third-party proxies or caching layers, so the 1.1% AI-bot figure could be a slight underestimate. And "no measurable use today" doesn't mean llms.txt could never matter, AI retrieval architectures are still changing, and a platform could start reading it tomorrow. Right now, though, there is no credible evidence it moves AI citations or rankings.
Skip the file du jour. Your free AI Visibility Audit scans ChatGPT and Google AI for your business and tells you the top three things worth fixing, based on what the evidence actually supports.
Get your free AI Visibility Audit