Does llms.txt Actually Work? What the Data Shows

In short

llms.txt is a plain-text file at a site's root that lists a site's key pages for AI models to read. Semrush reported finding no correlation between having one and improved AI-search performance. It costs nothing to ship, but it will not replace real signals: static HTML facts, FAQPage schema, and third-party mentions.

• llms.txt is a simple index file, not a ranking signal with proven results.
• Semrush publicly reported no measured correlation between llms.txt and AI-search performance.
• The near-zero cost to add it is the real argument for shipping it, not a performance claim.

What is llms.txt and what does it claim to do?

llms.txt is a plain-text file placed at the root of a website, at a URL like example.com/llms.txt, that lists the site's most important pages in a short, structured format. The idea, proposed in 2024, was to give AI models a clean index of a site instead of making them crawl and parse full HTML pages.

The claim attached to it grew fast: add this file and AI models like ChatGPT, Claude, and Perplexity will understand your site better and cite it more often. That claim spread through SEO blogs and tool vendors long before anyone tested it against real data.

In practice, no major AI provider has confirmed that it reads llms.txt at all. GPTBot, ClaudeBot, PerplexityBot, and CCBot crawl and index the actual pages of a site the same way they always have. llms.txt sits next to that process, not inside it.

What does the evidence actually show?

Semrush, one of the larger SEO platforms, tested this directly and reported finding no correlation between having an llms.txt file and improved AI-search performance. That is a specific, testable claim from a company with the crawl and ranking data to check it, and it lands against the file, not for it.

This lines up with how the crawlers that matter actually work. GPTBot, ClaudeBot, PerplexityBot, and CCBot read the raw HTML response of a page and do not execute JavaScript. Gemini is the one exception, since it uses Google's renderer. None of these systems are documented as depending on an llms.txt file to decide what to cite. What they read is the page itself.

A separate claim worth naming and hedging: WRITER, a GEO analytics vendor, has stated that roughly 85 percent of what AI engines say about a brand comes from third-party platforms rather than the brand's own site. That number is vendor-reported, not independently verified, so treat it as a directional signal, not a fact to repeat as settled. Even hedged, it points the same direction as the Semrush finding: what happens off your own root directory matters more than a file that sits on it.

Is llms.txt worth shipping anyway?

Yes, and the reason has nothing to do with the evidence above. llms.txt costs almost nothing to add: one static file, a few minutes of work, no design, no content rewrite, no risk of breaking anything else on the site. When the cost of doing something rounds to zero, you do not need proof it helps to justify doing it. You only need proof it cannot hurt.

That is the honest way to frame llms.txt. It is not a growth lever. It is a low-effort, low-risk addition that keeps you aligned with an emerging standard in case adoption changes later, without pulling time away from the work that is actually shown to move AI citations.

What actually moves AI citations instead?

Four things have a real, observable connection to getting cited by AI engines, and none of them are a text file.

One more piece is worth naming on its own: ChatGPT search runs on Bing's index. If a page is not indexed in Bing, it is much harder for ChatGPT search to surface or cite it, regardless of whether that page ships an llms.txt file or not. Submitting a sitemap to Bing Webmaster Tools matters more than most sites realize.

What mistake do most businesses make with llms.txt?

The mistake is not adding the file. The mistake is treating it as a substitute for the harder work. A business that ships llms.txt and stops there has done the cheapest, least effective item on the list and skipped the four that actually correlate with citations.

The honest sequence is the reverse of how most content presents it: fix static HTML rendering first, add FAQPage and other schema second, source your statistics third, build third-party mentions fourth, and add llms.txt whenever it is convenient, because it takes minutes and changes nothing else about your priorities.