Paste a site or docs URL. We find its llms.txt, check it against format v2, test a sample of its links, rule out a catch-all 200 reply with a random control path, and show which AI crawlers its robots.txt lets read the file. Free, no signup.
Takes about 3 to 10 seconds. Free: 30 checks a day.
We ran this checker once, on 5 November 2026, on the 128 hosts that our Tranco top 1,000 census found serving a file, and could read 123 of them. (The census headline of 122 leaves out hosts whose home page it couldn't read.) Links were limited to the first 12 on the site's own host that its robots.txt allowed.
| Open with the required H1 | 120 of 123 |
| Have a blockquote summary | 101 of 123 |
| List at least one link in the format | 108 of 123 |
| Also serve llms-full.txt | 23 of 123 |
Link to the file with rel="describedby" (new in v2) | 8 of 121 |
| Block GPTBot or ClaudeBot from the file in robots.txt | 4 of 121; none block a search or browsing agent |
| Links that answered 200 with something other than HTML | 254 of 1046 (24%) |
| Sites where most tested links aren't HTML | 24 of 98 |
| Sites with at least one tested link answering 404 or 410 | 10 of 123 |
Of 1189 links tested: 1046 OK, 30 answered 404 or 410, 11 gave a server or connection error, 86 blocked our checker and 16 gave no clear answer. These are the answers our checker got under its own user agent; a site may answer a browser differently, so some of the 404s may not be what people see. One run from one French IP. Not read: 5 of 128 files (3 bot walls, 1 HTTP 406 that returned 200 when retried later, 1 connection failure).
The format is plain Markdown in a set order. Only the H1 is required. A file can sit at the site root or in any folder, where it covers the pages under that folder (/docs/llms.txt covers /docs/), and the most specific file wins. Links should lead to LLM-friendly content, such as a Markdown copy of the page.
# Your site or product
> One or two sentences: what this is, who it's for, and the fact an agent most needs.
Anything an agent should know before following the links. Paragraphs and lists are fine here, headings aren't.
## Docs
- [Quick start](https://your-site.com/docs/quickstart.md): Install and first request
- [API reference](https://your-site.com/docs/api.md): Every endpoint and parameter
## Optional
- [Changelog](https://your-site.com/changelog.md)| H1 | The name of the site or project. Required, and the first thing in the file. |
| Blockquote | A short summary with the key facts needed to read the rest. Optional, but it's the line that tells an agent what this is. |
| Details | Paragraphs or lists with no headings. |
| H2 sections | File lists: - [name](url): notes. The name and URL are required, the notes aren't. |
| Optional section | A convention for secondary links. Since v2 (August 2026) it no longer has a mechanical meaning. |
rel="describedby" | New in v2: a Link header or <link> element that points a page at its llms.txt, so agents don't have to guess: Link: </llms.txt>; rel="describedby". |
A 200 status proves little. When we measured the Tranco top 1,000, 46 of 466 readable sites answered /llms.txt with a 200 HTML page, usually their own home or error page, and 122 served a real file. So this checker reads the body and asks for a random path next to the file as a control.
A missing file isn't an error. Per Chrome's docs, Lighthouse flags a server error on /llms.txt and marks a 404 as not applicable, since the file is optional.
Coding agents and documentation tools do use these files, according to the proposal's author. Our own logs are one small site, so treat them as an anecdote: our /llms.txt and /llms-full.txt drew 35 fetches between 17 October and 5 November 2026, none from ChatGPT-User, which read our home page 24 times in the last 7 of those days. If ChatGPT matters to you, make the pages themselves readable as plain text too. The robots.txt checker shows whether AI crawlers may fetch them at all, and the website screenshot tool shows what a browser gets.
The same report as JSON, free (0 credits, 500 checks a day per key). Fail a build when verdict is problems:
curl "https://shotmint.creasework.com/v1/check?type=llms&url=your-site.com" \
-H "X-API-Key: YOUR_KEY"verdict is valid, warnings, problems, missing, soft_404, walled or server_error. Keys take one request (curl -X POST https://shotmint.creasework.com/v1/keys/free), with no signup and no card. The same key renders screenshots and PDFs of any page: 25 a month free, then €5 for 2,000, and credits never expire. Get a free API key
robots.txt checker · TXT to PDF · Redirect checker · Broken link checker · SSL checker · HTTP header checker · SPF checker · DMARC checker · Website technology checker · Structured data checker · Favicon checker · Open Graph checker · Diff checker · all free tools