No engine that decides your AI visibility is documented to read llms.txt. Google says outright it does not use it, and server logs show the crawlers behind ChatGPT and Claude barely request the file at all. OpenAI, Anthropic and Perplexity each publish their own llms.txt only for their developer docs, a different job from getting your business cited.
That is also the honest, short answer to the myth-check on whether llms.txt helps AI find you at all. This article goes one level deeper: engine by engine, what each company has actually said or shown, not what a blog post assumes.
llms.txt is a Markdown file, published at a site's root, meant to give AI systems a short, curated map of a site's most important pages instead of making them read the whole thing. Jeremy Howard and the Answer.AI team proposed the format in September 2024 as a fix for large, navigation-heavy sites that are slow for a model to read directly. The idea was never in question; whether the engines that name businesses in AI answers actually consume it is.
Proof: the specification is published by its author at llmstxt.org.
OpenAI's own documentation for its crawlers does not name llms.txt as something GPTBot or OAI-SearchBot fetch or use. GPTBot crawls pages to help train future models; OAI-SearchBot fetches pages live to answer a specific question inside ChatGPT search. Both are described in detail in the Reflexa AI crawler guide, and neither description mentions a curated index file as an input.
Do this: before assuming llms.txt is the reason ChatGPT does or does not mention you, confirm GPTBot and OAI-SearchBot can even reach your site with the free AI Crawler Access Check.
Anthropic publishes its own llms.txt on its developer documentation site, but that is a statement about its docs, not about how ClaudeBot treats other people's websites. Anthropic runs three separate bots, ClaudeBot for training and Claude-User and Claude-SearchBot for live use inside Claude, and none of the public documentation for those bots names llms.txt as part of how Claude decides what to cite in an answer.
Rule of thumb: a company publishing its own llms.txt for developers is not the same claim as that company's crawler reading yours. Keep the two separate.
Perplexity, like OpenAI and Anthropic, publishes its own llms.txt for its documentation, and PerplexityBot's published crawler documentation describes how it indexes and respects robots.txt, not that it consumes a site's llms.txt as a citation input. Perplexity is the engine most worth checking directly if this changes, since it already shows its sources openly in every answer, the behaviour covered in how to get cited by Perplexity.
Do this: if you want to see exactly what Perplexity is citing instead of your llms.txt, run a real buyer question through the free check at dashboard.reflexatechnologies.com/audit and read the sources it names.
Mostly not, and that is now measured rather than guessed. Ahrefs analyzed server logs across roughly 137,000 domains and found 97% of published llms.txt files received zero requests in the period studied. Of the small share that were fetched, most hits came from SEO audit tools checking whether the file existed, not from the crawlers that actually retrieve content for AI answers.
| Engine / company | Publishes its own llms.txt (docs) | Documented to read third-party llms.txt for citations |
|---|---|---|
| OpenAI (ChatGPT) | Yes | Not documented |
| Anthropic (Claude) | Yes | Not documented |
| Perplexity | Yes | Not documented |
| Google (AI Overviews / AI Mode) | No | States directly: no |
Proof: Ahrefs — server-log study of llms.txt requests across 137,000 domains · Search Engine Land — Google says llms.txt won't be used.
Bottom line: adoption of reading the file, not adoption of writing one, is the actual gap. A file almost nobody's crawler requests cannot be deciding who gets cited.
Yes, for one narrow case: sites that are documentation-heavy and talk to coding agents or AI-assisted developer tools. That is exactly why OpenAI, Anthropic and Perplexity each maintain one for their own docs. It costs a few minutes to write, there is no real downside to having one, and it can make dense technical docs easier for an agent tool that is explicitly built to look for the file. For a general business chasing AI recommendations, that is not the same problem.
Do this: if you already have an llms.txt file, run the free llms.txt validator first. It checks the file against the llmstxt.org format and resolves its links, so at minimum the file you spent time on is not broken.
None of it has anything to do with a Markdown index file. As the what is GEO pillar covers in full, it comes down to five things an engine checks before naming a business: can it reach you, can it read you, can it identify you, does the web corroborate you, and does your content answer the question directly.
Do this: before writing or rewriting an llms.txt file, confirm none of the AI crawlers are blocked in robots.txt with the free AI Crawler Access Check — that single check affects citations far more than any curated index ever will.
None of these require buying anything, and each takes a few minutes:
Bottom line: spend the 10 minutes on access and identity first. An llms.txt file is the last five minutes of a GEO checklist, not the first.
There is no public evidence for that. Google states directly it does not use llms.txt, and server-log data shows the crawlers behind ChatGPT and Claude almost never request the file at all. It is not a citation lever.
OpenAI, Anthropic and Perplexity all publish their own llms.txt files on their developer documentation sites, mainly to help coding assistants navigate their docs. That is a different job from a general website getting cited in an AI answer.
No. Publishing one is harmless. The risk is time spent on it instead of the things that actually decide AI visibility: crawler access, entity clarity, and content that answers the buyer's question directly.
It looks similar but is not adopted the same way. Robots.txt and sitemap.xml are named directly in engine documentation and control real crawl behavior. No engine's own documentation instructs its crawlers to fetch or act on llms.txt during retrieval.
If your site is documentation-heavy and talks to coding agents or AI-assisted developer tools, a curated llms.txt can help those specific tools navigate your docs. For everyone chasing general AI recommendations, that 10 minutes is better spent on crawler access and entity clarity.
The free check reads your crawler access and entity clarity — 3 minutes, evidence included.