Does llms.txt work? What the evidence says

llms.txt has no good evidence behind it as a way to get cited by AI assistants. Google Search ignores the file, two large studies found no link with citations, and in a one-month log study 97% of the files got no requests at all. Its most plausible use is narrower: helping coding agents and developer tools find your documentation.

Quotable offers a free llms.txt generator and llms.txt checker, even though the evidence for any benefit is weak. They save typing and catch format errors. They won’t get you named in an answer, and nothing we’ve read suggests the file will. Quotable doesn’t publish an llms.txt of its own.

What is llms.txt?

llms.txt is a proposed Markdown file, served at the root of a website, that gives AI tools a short guide to a site’s most useful pages. Jeremy Howard proposed it in September 2024. It isn’t an official standard. The proposal expects the file to be read on demand while an agent helps someone, more than during model training.

Only the title line is required. A minimal file for a developer product looks like this:

# Example Cloud

> Example Cloud is a log search service for engineering teams.

## Docs

- [Quickstart](https://example.com/docs/quickstart.md): Send your first logs
- [API reference](https://example.com/docs/api.md): Endpoints and limits

## Optional

- [Changelog](https://example.com/changelog.md)

llms.txt does a different job from robots.txt, which says what automated access is acceptable. Our robots.txt reference covers that side, bot by bot.

Does llms.txt get a site cited by AI assistants?

No study we could open shows that llms.txt gets a site cited, and two large ones found no link. All three studies below come from companies selling SEO or AI visibility tools, and none is a controlled test.

llms.txt is also awkward to test properly, in our view. A site has one file, so you can’t add it to half your pages and compare them with the other half. Any comparison has to be between websites, and websites differ in many other ways. So the absence of a link isn’t proof of no effect, but nobody has shown an effect either.

Do AI crawlers read llms.txt?

AI search crawlers rarely request llms.txt in the largest log data we found. Ahrefs looked at about 38,360 domains that publish the file, and counted about 22,000 requests to those files across May 2026. Requests were concentrated on about 1,100 domains.

Those groups are the ones most relevant here, so the chart leaves out the rest. AI retrieval bots are the group closest to AI search, and they made 233 of about 22,000 requests. Ahrefs also found no AI bot requests for llms.txt files that didn’t exist, so it saw no sign of AI systems probing for the file.

Ahrefs states its own limits. Its customers skew towards technical and SEO-minded teams. The data covers one month. It measured only the llms.txt file, not any Markdown pages the file links to. A request also doesn’t show that a system used what it fetched, so these counts are ceilings.

What do AI companies say about llms.txt?

Google is the only major operator we found with a direct statement on llms.txt. Google’s guide says Google Search doesn’t use these files, and that they neither help nor harm visibility there. A 15 June 2026 entry in its change log records that note, which also leaves you free to keep a file for other services.

The other crawler pages we opened on 6 November 2026 say nothing useful either way. OpenAI’s crawler page, Perplexity’s and Anthropic’s describe robots.txt and IP controls, and none lists llms.txt as an input to crawling. OpenAI’s and Perplexity’s documentation sites do point to llms.txt indexes of their own docs. That shows they offer the file to people using their docs, not that their crawlers read yours. Silence isn’t a denial, so we can’t rule out unpublished use.

When is llms.txt worth publishing?

llms.txt is worth publishing when your customers use coding agents on your documentation or API, and when upkeep costs almost nothing. This is the one use with some support. In the Ahrefs data, Claude Code’s agent fetched the files more often than any retrieval bot, assistant or training crawler except GPTBot. That’s a count of requests, with no measure of whether the agent did better work.

Our view, for a tech company or brand:

The file also carries a small security cost. Ahrefs flags prompt injection as a risk, because agents may act on what the file says. It advises treating the file like code, with version control, limited access, change alerts and plain links.

How can you check llms.txt in your own logs?

Your own access logs answer the question for your site, at no cost. This command counts requests for the file from the main AI crawlers:

grep '/llms.txt' access.log | grep -ciE 'GPTBot|OAI-SearchBot|ChatGPT-User|ClaudeBot|Claude-User|Claude-SearchBot|PerplexityBot|Perplexity-User'

Run it over at least a month of logs. If you use a CDN, check its logs as well, since cached requests may never reach your server. A user agent is only a claim, because anyone can send one, so treat the count as an upper limit. Our AI bot log analyser shows which AI crawlers visit overall and their top paths, in your browser.

What this can’t tell you

We can’t tell you what publishing llms.txt will do to your AI mentions, and neither can a before-and-after check. Answers move around from day to day for unrelated reasons, as our guide to AEO explains. Quotable sells a tool that tracks AI answers, and it couldn’t isolate this file’s effect on a single site either.

What would change this answer?

Three things would change our view of llms.txt, and we’d update this page if any appeared:

• An AI search operator stating that it reads the file and what it does with it.
• Log data showing retrieval bots requesting it regularly, across a wider sample than one analytics product’s customers.
• A controlled test across matched websites that finds a citation difference.

If you want to be named when buyers ask, start with our guide to AEO instead.