Does llms.txt Work? What the Evidence Actually Shows

llms.txt is a proposed text file placed at a domain's root, meant to tell AI systems which pages matter and how to describe the brand — modeled on robots.txt and sitemap.xml. As of mid-2026, the evidence is that it does not measurably improve AI citation: large-scale studies across hundreds of thousands of domains find no correlation between having the file and being cited, and adoption sits near 10% of sites. Google has stated plainly that its AI systems don't use it; other platforms have neither adopted nor endorsed it as a citation signal. It's cheap and harmless to ship, and there are early, limited signs of use in the agentic / developer-tooling layer — but the honest position is: ship it for agent-readiness if you want; don't expect, or sell, a citation lift.

TL;DR. For getting cited in AI search, no — the evidence says llms.txt does nothing measurable. Two large studies found no link between having the file and being cited, Google says its AI doesn't use it, and most files are never read. It's cheap and harmless, and it has one real use in the agentic/developer-tooling layer. Ship it if you want agent-readiness; don't pay anyone who sells it as a ranking or citation lift.

Everyone tells you what it is. Almost nobody tells you if it works.

Search does llms.txt work and look at what ranks. A stack of "what is llms.txt" explainers — gitbook, mintlify, AIOSEO, Semrush — all walking you through the format. Then, near the top, two things that actually answer the question you asked: a Reddit thread where someone bluntly asks whether these files actually help websites appear in AI, and SE Ranking, with a study that set out to measure it.

That's the gap. The category is drowning in explanations of the file and thin on evidence about whether it does anything. So here's the evidence, because the honest answer to "does it work" is more useful than another walkthrough of the syntax.

What the evidence says

It doesn't work for citation. That's not an opinion — it's what the largest measurements available have found.

SE Ranking analyzed 300,000 domains and found no correlation between having an llms.txt file and how often a domain gets cited by AI models. In their machine-learning model, removing the llms.txt variable actually improved accuracy — the file was adding noise, not signal. Adoption was about one in ten sites, far from the universal uptake of robots.txt or sitemaps.

Ahrefs went at it from the other side, looking at server logs across 137,000 sites, and found that 97% of llms.txt files received no requests at all — nothing fetched them. Of the small fraction that were fetched, a meaningful chunk came from SEO tools and the industry studying itself, not from the AI search bots people install the file to impress.

And the platforms themselves are explicit. Google's John Mueller has said plainly that its AI systems don't use llms.txt and described it as, at most, a temporary convenience for AI coding tools — not something that affects search or AI answers. No major AI platform has committed to reading it as a citation signal.

So why does everyone ship it?

Because it's cheap, harmless, and it feels like doing something. The file is trivial to create, it builds on familiar conventions like robots.txt, and in mid-2024 a wave of speculation framed it as the missing link between websites and AI. The speculation drove adoption; the confirmation never arrived. None of that makes it useful for citation — it makes it a low-cost bet that hasn't paid off.

The honest framing matters here, because it's exactly where overclaiming happens. llms.txt is not a scam and it's not harmful. It just doesn't do the specific thing most people install it for.

Where it actually does something: the agentic layer

There is one real use, and it's worth being precise about, because precision is the difference between honest advice and dismissal.

llms.txt shows genuine, if early and limited, traction in the agentic and developer-tooling layer — the part of the AI stack where agents fetch context to complete tasks rather than where search engines pick citations. In Ahrefs' logs, the AI bots that did fetch the file most were coding-agent and agentic-infrastructure bots, not the retrieval bots behind ChatGPT Search or Perplexity. That it ranks alongside developer-doc platforms like gitbook and mintlify in the results is the same signal: this file is plumbing for agents reading documentation, not a visibility lever for marketing pages.

So if your customers point coding agents at your docs, llms.txt has a real, if modest, job to do. If your goal is showing up in AI answers, it doesn't.

The honest verdict

Ship it if you want agent-readiness — it's cheap, and the day a major answer engine starts using it (if that ever comes), you're already there. Treat it as version-controlled like any file, since agents are built to trust it and a stale or tampered file misleads every agent that reads it.

But do not expect a citation lift, and do not pay anyone who sells you one. An agency or tool promising that llms.txt will get you cited in ChatGPT is selling a result the data doesn't support — which is its own kind of tell about whether to trust the rest of what they measure. (Why a claim has to trace to a real signal.)

FAQ

Does llms.txt work for AI search visibility? No, not for citation. Large studies across hundreds of thousands of domains found no link between having the file and being cited by AI, and Google says its AI doesn't use it. Most llms.txt files are never even fetched.

Will llms.txt help my SEO or AI citations? There's no measurable evidence that it does. Ranking and AI citation are won by content, authority, and extractability — not by a manifest file at your domain root.

Should I create an llms.txt file? Only if you want agent-readiness for coding agents and dev tooling, where it has a real but modest use. It's cheap to ship and harmless. Just don't expect it to affect whether you show up in AI answers.

Do AI engines actually read llms.txt? Mostly no. The bots that fetch it most are coding agents and the SEO industry's own tools — not the retrieval bots behind ChatGPT Search, Perplexity, or AI Overviews. The retrieval engines barely touch it.