Does llms.txt Work? What the 2026 Data Actually Shows
Three independent studies and Google's own documentation now agree: the file marketers were told to rush out has almost no effect on AI search visibility. Here is what to do instead.
Out of 62,100 AI bot visits in one 90-day study, exactly 84 touched the llms.txt file. That is not a strategy. That is a rounding error.
Short answer: as of mid-2026, there is no credible evidence that llms.txt improves your visibility in AI search. Three independent studies and a fresh note in Google's own documentation all point the same way. If you published an llms.txt file hoping ChatGPT, Perplexity, or Google's AI Overviews would notice you more, the data says they almost certainly did not.
That is an uncomfortable conclusion, because llms.txt has been one of the most-hyped "AI SEO" tactics of the past year. So let us walk through what it is, what the evidence shows, why the idea spread anyway, and where the few real use cases actually live.
What is llms.txt?
llms.txt is a proposed standard: a single Markdown file placed at the root of your domain (yoursite.com/llms.txt) that gives large language models a clean, curated map of your most important content. Think of it as a robots.txt for the AI era — a deliberately simple file that says, in effect, "here is what matters on this site, and here are the links to read."
The proposal, floated in 2024, solved a real-sounding problem. AI models have limited context windows and struggle with bloated HTML, navigation menus, and JavaScript. A tidy Markdown index sounds like exactly the help they need. The pitch spread fast through the generative engine optimization community, and tens of thousands of sites added one.
The trouble is that a standard only works if the systems it targets actually read it.
Does llms.txt work for AI search? The 2026 evidence
Here is where the marketing pitch collides with measurement. Across three separate 2026 analyses, the verdict is remarkably consistent.
Ahrefs analyzed 137,000 domains. Among the sites that had published an llms.txt file, 97% received zero requests for it in May 2026. As Ahrefs put it: "No bots, no humans, nothing." Roughly 28% of studied websites had bothered to create the file — and almost none of them were getting it read.
OtterlyAI ran a 90-day controlled experiment. It published an llms.txt file and tracked every AI bot that hit the site: more than 62,100 visits in total. Exactly 84 of them — about 0.1% — touched the llms.txt file. Worse, the file drew roughly 3x fewer visits than an average page on the same site. AI crawlers treated it with no more priority than a stray PDF.
Google made it official. In a June 2026 update to its Search Central documentation, Google added a section titled "Clarifying guidance on llms.txt files," stating plainly that the file is not needed for Google Search and produces neither a positive nor a negative visibility effect. Google's John Mueller had already compared llms.txt to the long-dead keywords meta tag — a signal nobody reads.
No major model provider — OpenAI, Anthropic, Google, or Meta — has formally adopted llms.txt as part of its crawler. The standard exists. The infrastructure to honor it largely does not.
| Claim about llms.txt | What the 2026 data shows |
|---|---|
| "It boosts AI search citations" | No measured correlation (OtterlyAI, 90-day study) |
| "AI crawlers fetch it" | 97% of files got zero requests (Ahrefs, 137k domains) |
| "Google uses it for AI Overviews" | Google says it is not needed and has no effect |
| "Major LLMs read it" | No major provider has adopted it as of mid-2026 |
Why did everyone rush to publish one?
If the file does so little, why did it become a 2026 marketing checklist item? Three forces, none of them about the data.
- It is trivially easy. Writing a Markdown index and uploading it takes an afternoon. Low effort plus a plausible story is irresistible to anyone with a long to-do list and a fear of missing the AI wave.
- Plausibility masqueraded as proof. "Models have small context windows, so a clean file must help" sounds airtight. It is a hypothesis, not a finding — but in a fast-moving field, confident hypotheses travel faster than careful tests.
- Tool vendors amplified it. Plugins and platforms that auto-generate llms.txt had every incentive to frame it as essential. A checkbox is easy to sell.
This is a recurring pattern in AI marketing: a tactic that is cheap to do and hard to disprove spreads regardless of whether it works. The cure is the same as always — measure before you believe.
When llms.txt actually is useful
The honest nuance: llms.txt is not worthless. It is simply mis-sold. Its genuine use cases are about developer tooling, not search rankings:
- In-product AI assistants and docs. If you run an API or software product, tools like Cursor, Claude Code, GitHub Copilot, and Windsurf can fetch an llms.txt to help developers query your documentation. This is a real, growing pattern — but it is a documentation UX feature, not an SEO one.
- MCP servers and internal agents. Systems you control can be pointed at an llms.txt deliberately. The value comes from your own tooling reading it, not from the open web's crawlers discovering it.
- Low-risk hygiene. Publishing one costs little and carries minimal downside (the main caveat: it can make your content marginally easier for competitors to scrape). If you have docs and a spare hour, fine. Just do not expect citations.
In short: keep it if you ship developer tools; skip the hype if your goal is AI search visibility.
What to do instead for AI search visibility
If llms.txt is not the lever, what is? The same fundamentals that Google, OpenAI, and every serious GEO study keep pointing back to:
- Be crawlable and fast. AI answer engines read the same web your users do. Clean HTML, server-rendered content, and a working sitemap beat any side-file. If a crawler cannot reach your page, no manifest will save it.
- Write genuinely extractable content. Lead with a direct answer, use descriptive question-led headings, and structure information in short paragraphs, bullets, and tables that a model can lift cleanly. (We break this down in our guide to fixing AI-sounding content.)
- Build entity and brand authority. Models cite sources they "trust" — sites with consistent naming, citations, and a recognizable point of view. A well-trained, consistent brand voice does more for citation than any config file.
- Earn real references. Mentions and links from credible sites remain among the strongest signals that you are worth quoting.
None of this is as satisfying as uploading a file and declaring victory. But it is what the evidence supports — and it compounds. This is exactly the kind of research-then-execute loop an autonomous marketing system like ivon is built to run continuously, so your team spends its time on judgment instead of chasing every unproven tactic.
The takeaway
llms.txt is a tidy solution to a problem the major AI platforms have not agreed to solve. Until OpenAI, Anthropic, or Google announce they read it — and they have not — treat it as optional developer hygiene, not a visibility strategy. Spend the hour you would have spent agonizing over the file on content a model can actually find, parse, and trust. For a fuller playbook, see our guide to generative engine optimization.
Frequently asked questions
Does llms.txt help SEO or AI search rankings?
No. As of mid-2026, three independent studies and Google's own documentation found no measurable benefit. Ahrefs found 97% of published llms.txt files got zero requests, and Google states the file is not needed for Search and has no positive or negative effect on visibility.
Do ChatGPT, Perplexity, or Google read llms.txt?
Not at scale. No major model provider — OpenAI, Anthropic, Google, or Meta — has formally adopted llms.txt as part of its crawler. In OtterlyAI's 90-day test, only 0.1% of AI bot visits touched the file, and Google has explicitly said it does not use it.
Should I still create an llms.txt file?
Only if you ship developer tools or documentation. IDE assistants like Cursor and Claude Code can fetch it to help developers. For AI search visibility, it offers no demonstrated benefit, so prioritize crawlability and extractable content instead.
What actually improves visibility in AI answer engines?
The fundamentals: fast, crawlable pages; content that leads with a direct answer and uses clear headings, bullets, and tables; consistent brand and entity signals; and credible references from other sites. These are the factors every major 2026 study and Google itself keep emphasizing.
Sources
- Ahrefs — What Is llms.txt?
- OtterlyAI — The llms.txt Experiment
- Contentful — Do llms.txt files actually improve AI search visibility?
- Google Search Central — Clarifying guidance on llms.txt files (June 2026)
- SE Ranking — LLMs.txt: Why Brands Rely On It and Why It Doesn't Work
- Wix Studio — Debunking llms.txt Myths