If you run a site, you have probably been told to add an llms.txt file — a plain-text, Markdown “map” that tells large language models which pages matter and how your content is structured. The pitch is tidy: help the AIs understand you, and get cited more. After eighteen months of that pitch, the data is in, and it is unkind. For AI search, llms.txt does close to nothing. For developer tooling, it quietly does something real. Knowing the difference saves you time.

What it is supposed to do

Proposed in 2024 (by Answer.AI’s Jeremy Howard), llms.txt lives at your domain root, like robots.txt, but aimed at language models: a curated, Markdown index of your key URLs and summaries so a model doesn’t have to wade through your HTML and navigation. Reasonable idea. The question was always whether anyone on the consuming side would actually use it.

The adoption numbers tell the story

Two figures, side by side, settle most of the debate:

  • Adoption is real-ish. An SE Ranking study of 300,000 domains found a 10.13% adoption rate — roughly one in ten sites now ship an llms.txt.
  • Usage is not. In one analysis of over 500 million AI-bot visits across 90 days, only 408 requests targeted llms.txt directly. That is on the order of eight ten-thousandths of one percent of AI-crawler traffic — statistically zero. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended overwhelmingly skip the file and crawl your HTML like always.

One in ten sites publish it; about one in a million bot visits read it. That gap is the whole point.

Google — and everyone else — said no

This is not ambiguity. In July 2025, Google’s Gary Illyes said Google does not support llms.txt and has no plans to; John Mueller publicly compared it to the long-discredited keywords meta tag (Search Engine Roundtable). Google noted the file even appeared to be supported only because an internal CMS had added it and some teams never removed it. As of mid-2026, having an llms.txt does not measurably improve your odds of being cited by ChatGPT, Claude, Gemini, or Perplexity in their answer surfaces. There is no standard, no enforcement, and no adoption from OpenAI, Google, Anthropic, Meta, or Mistral on the search side.

Where it actually helps: developer tooling

Here is the nuance the “it’s dead” takes miss. Agentic developer tools do fetch it. Cursor, Claude Code, GitHub Copilot, Windsurf, MCP servers, and a growing set of in-product AI assistants pull llms.txt to orient themselves in a codebase or a documentation site. If you run docs, an API, or a developer product, an llms.txt (and an llms-full.txt) genuinely helps coding agents and doc assistants use you correctly. That is a real, narrow, non-SEO benefit.

Our take

Full disclosure: we publish one at mrtd.net/llms.txt. We keep it because it is cheap, it is honest documentation of our structure, and the developer-tooling use is legitimate — not because we expect it to win us AI citations. If your goal is to be cited by AI search, your time is far better spent on the things that demonstrably move the needle: clean, extractable HTML; clear structured data; verifiable facts with sources; and being a primary reference others link to.

Bottom line: llms.txt is not a scam and not a ranking hack. It is useful documentation for agents and a no-op for AI search. Ship one if you serve developers; don’t expect it to do anything for your citations.

We cover SEO and GEO with sources and skepticism. Disagree, or have fresher data? Reach us via @mrtdnet on Telegram.