# llms.txt

A hand-written index of a site's most useful pages, written for language models.

**[Discoverability](/docs/discoverability)** · watched, not scored

The idea is sound: point an agent at the pages that matter instead of making it crawl everything. Nothing consumes it, so publishing one currently changes nothing an agent does.

## Why we do not score it

We will score it when a major AI vendor states that a production surface reads llms.txt. Until then the files are written and never fetched, and scoring them would reward work that has no effect.

## Adoption

**Early production** (L2) — widely published, and almost never fetched by an AI crawler

Several independent organisations run it in production, though it is still niche on the open web. Direction of travel: ► flat, as of 2026-07-28.

Files are up 8.8x year on year to roughly 36,000 in a 300,000-domain sample, but Ahrefs found 97% of 137,000 files never receive an AI-bot request, and Google has said it does not and will not support it.

| Adopter | Depth | Evidence |
|---|---|---|
| Mintlify (auto-generates for all hosted docs) | default-on | [mintlify.com](https://www.mintlify.com/blog/context-for-agents) |
| Cursor, Windsurf (docs-lookup consumers) | production | [organikpi.com](https://organikpi.com/blog/distribution/llms-txt-adoption-impact/) |

## Specifications

| Document | Revision | Kind |
|---|---|---|
| [llms.txt proposal](https://llmstxt.org/) | living document | specification |

Last re-read against the published documents: 2026-07-28.

## Further reading

- [Ahrefs: the llms.txt study](https://ahrefs.com/blog/llmstxt-study/)
- [SE Ranking data via ppc.land](https://ppc.land/llms-txt-adoption-rises-8-8x-but-97-of-files-get-zero-ai-requests/)