An llms.txt that passes the audit
Most generators crawl your site and dump every URL they find. Lighthouse then fails the result as not well-formed. This one builds a curated file and checks it against all eleven spec rules before you publish.
Nothing is uploaded. Parsing happens in your browser.
Checks whether GPTBot, ClaudeBot, PerplexityBot and seven others can actually reach your site.
Publish it at /llms.txt
Download and upload to your site root, served as plain text. Lighthouse looks for it there and nowhere else.
Your llms.txt
Nothing to build yet
Paste your sitemap on the left, or switch to validate and paste a file you already have.
Nothing is uploaded. No language model sees your site structure.
Next, check how the pages themselves present
An llms.txt points agents at your pages. These check what they find when they get there.
llms.txt Generator and Validator
Start here: llms.txt does nothing for your Google rankings
Google's own AI optimisation guidance says the file is not needed for Google Search, and John Mueller has compared it to the retired keywords meta tag. No major AI provider has publicly confirmed that it influences how their systems source or cite content. Anyone selling you llms.txt as an SEO win is selling you something.
So why does this tool exist? Because in May 2026 Chrome shipped Lighthouse 13.3, which moved a new Agentic Browsing category into the default configuration. Everyone running a standard Lighthouse report — and by extension PageSpeed Insights — now sees an agent-readiness section, and two of its audits are about this file: llms-txt-present checks it is reachable as plain text, and llms-txt-well-formed checks it parses against the llmstxt.org reference.
That category deliberately has no 0–100 weighted score, because as Google puts it, the standards for the agentic web are still emerging. It is a signal, not a ranking factor. But it is a signal your clients can now see in a tool they already run, which is why the question stopped being academic.
The llms.txt format, in one table
File structure
| Element | Rule | Required |
|---|---|---|
| H1 | The site or project name, exactly one, first | Yes |
| Blockquote | One or two sentences of summary | Recommended |
| Free-form text | Context, but no headings | Optional |
| H2 sections | Each holds a Markdown list of links | Recommended |
| List items | - [name](url): description | Fixed format |
| ## Optional | Links here may be skipped for shorter context | Optional |
Publishing checklist
Upload the file to your site root so it resolves at yoursite.com/llms.txt, served as text/plain with UTF-8 encoding. Lighthouse looks there and nowhere else.
BEFORE YOU PUBLISH
- Keep it to 10–20 links across 3–5 sections.
- Give every link a one-line description.
- Move legal pages and appendices into
## Optional. - Use absolute https:// URLs only.
- Check robots.txt is not blocking the agents you want.
Running this across a client portfolio? Agent readiness is four audits, not one file — accessibility tree, layout stability, WebMCP and llms.txt. We audit and fix all four.
Questions people ask
No. Google's guidance states it is not needed for Google Search, and no major AI provider has confirmed it affects how their systems source or cite content. Lighthouse audits for it under Agentic Browsing, but that category explicitly does not evaluate ranking.
Two audits. llms-txt-present confirms the file is reachable as plain text at your root. llms-txt-well-formed confirms it parses against the llmstxt.org reference structure. If the file returns a 404 the audit is marked not applicable, since publishing it remains optional — but if your server returns an error, that is flagged.
An H1 with your site name, then an optional blockquote summary, then optional Markdown without headings, then H2 sections each containing a list of - [name](url): description items. The H1 is the only required element.
Ten to twenty across three to five sections suits most sites. The file exists to express judgement about what matters. Listing everything defeats it and makes the file too long to be useful as context.
An H2 literally named Optional has defined meaning in the spec: parsers may skip those links when they need a shorter context. Use it for appendices, legal pages and secondary references.
No. Parsing, generation and validation all run in your browser. Nothing is uploaded, stored or logged, and no language model sees your site structure.
At your site root, so it resolves at yoursite.com/llms.txt, served as text/plain with UTF-8 encoding. Lighthouse looks there and nowhere else.