Short answer: not for AI search visibility, not on any evidence published so far. As of mid 2026, no major AI provider has committed to reading llms.txt in production, Ahrefs' analysis of 137,210 domains found 97% of published llms.txt files receive zero requests, and SE Ranking's 300,000-domain study found no statistically significant correlation between having the file and being cited by AI. Google has said, on the record, that it does not use it.
And yet the file has one genuinely legitimate use case, costs about ten minutes, and does no harm when done properly. So the honest verdict is more interesting than either camp's version. Here is the full picture, including why we built a generator for it anyway.
What is llms.txt?
llms.txt is a proposed convention, published by Jeremy Howard of Answer.AI in September 2024. It is a Markdown file at your site root, at /llms.txt, that gives AI systems a curated index of your most important pages with one-line descriptions, often with a short authoritative summary of who you are.
Two things it is not. It is not robots.txt: it blocks nothing and controls nothing, it only suggests. And it is not a standard: there is no W3C or IETF backing, no version number, and no governance body. It is a community proposal that became widely known, which is exactly why the only honest way to judge it is by adoption and traffic data. Our evergreen reference on the spec and setup lives in the LLM Optimization hub; this piece is about the evidence.
Who actually reads it?
This is where the story falls apart, and the data is unusually consistent.
The providers have not signed up. As of 2026, none of OpenAI, Google, Anthropic, Meta, or Mistral has publicly committed to reading or acting on llms.txt in their production search or answer systems. Google is the most explicit: Gary Illyes confirmed in July 2025 that Google does not support the file and is not planning to, John Mueller compared it to the long-dead keywords meta tag, and by June 2026 Mueller was still calling the whole question purely speculative, noting the file has existed for years without any AI system using it. Google's own generative AI documentation lists it among unnecessary tactics.
The crawlers barely fetch it. Ahrefs read the server logs of 137,210 domains in May 2026. About 28% of them published an llms.txt, and 97% of those files received zero requests. Of the small fraction that were fetched, roughly a fifth of requests came from named AI tools, and about 12% came from the industry studying itself: GEO tools, llms.txt checkers, and researchers. Best of all, AI bots recorded zero requests for llms.txt on domains that do not have one. A crawler that wanted the file would probe for it and collect 404s. They do not ask. Limy's separate monitoring of more than 500 million AI bot events over 90 days counted only a few hundred requests targeting the file.
Having it does not correlate with being cited. SE Ranking tested roughly 300,000 domains and found no statistically significant relationship between llms.txt and AI citation frequency. When they removed the file as a variable from their prediction model, the model got more accurate: the file was noise, not signal. Among the 50 most-cited domains in AI answers, exactly one had the file.
One methodological trap worth knowing, because vendors fall into it constantly: adopters do show slightly higher raw citation rates in some panels. That is selection, not causation. The kind of site that ships an llms.txt is the kind of site already doing everything else right.
So why does it persist?
Three real reasons, one of which actually matters.
Adoption kept growing on momentum: SE Ranking measured 10.13% of domains, Ahrefs' traffic-weighted panel found 28%, and even 7.4% of the Fortune 500 had shipped one by March 2026. Platforms like Yoast, Webflow, and Mintlify made it a one-click default, so the cost of joining fell to nearly nothing.
The reason that matters: coding agents and developer tooling genuinely read these files. In Ahrefs' log data, the top fetchers were GPTBot and Claude Code, and the ecosystem around IDE assistants and agent tooling is the one consumer that demonstrably uses curated Markdown indexes today. If you run developer documentation, llms.txt has a real, current audience. That is the narrow case where it earns its keep.
For everyone else, it is a cheap option on a possible future. Mueller's practical test is the most useful sentence in the whole debate: create the file when an AI platform that actually brings you customers asks for it.
Should you add one anyway?
Yes, with correctly sized expectations. Our position, as a company whose product includes an llms.txt generator: ship it, spend ten minutes on it, and do not confuse it with a strategy.
The honest case for shipping it is threefold. It costs almost nothing. It has a real audience in agent and coding tools, which matters more every quarter as agentic browsing grows. And it is one more place where your brand facts are stated cleanly in your own words, which is worth having when the alternative is a model guessing. The honest case against overinvesting is everything in the section above.
Two rules if you do it. Keep it accurate and regenerate it when facts change, because a stale file that misstates your pricing is worse than no file. And never let it displace the work that actually moves visibility: teams reporting llms.txt as their AEO strategy is the real damage this file has done.
Diploria's generator builds the file from your brand profile, with your name, description, topics, permissions, and key pages, precisely so it takes minutes instead of a meeting. We tell every user the same thing this article says: it is hygiene, not leverage.
What moves AI visibility instead
The evidence-backed list is short and boring, which is why it works. Be crawlable by the AI agents you want (a robots.txt problem, not an llms.txt one, and one our AI Access tool inspects live). Earn brand mentions across the sources engines trust, since mentions predict AI citations roughly three times better than backlinks. Structure content so an engine can lift a clean answer from it. Keep key pages current, because cited content is measurably fresher than content that merely ranks. The full prioritized version is in our checklist for getting cited by ChatGPT, and the wider evidence base is in The State of AI Search.
If you want to know whether any of this is working for your brand, measure the output rather than the inputs: what the engines actually say about you. That is what Diploria tracks across 11 AI surfaces, and the free check takes under a minute.