What is an llms.txt file?

A short markdown file at /llms.txt that summarises a site for language models: what it is, which pages matter, and where the machine-readable surfaces live. It is a convention rather than a ratified standard, but it is cheap to publish and it is the only place you get to write your own description.

Does being agent-readable help SEO?

It overlaps heavily but is not the same goal. Classic SEO optimises for a ranked list of links; agent readability optimises for being quoted correctly inside an answer. Sitemaps, JSON-LD, and canonical URLs serve both. llms.txt and machine-readable twins serve the second.

Should I block AI crawlers?

That is a business decision, not a technical one. Blocking them protects your content from being trained on and summarised; it also removes you from the answers people increasingly read instead of search results. You cannot have both, and this tool only reports which one you have chosen.

An instrument for agent-native publishing

Is your site legible to an agent?

Twelve checks against the paths and markup an AI crawler actually looks for. Free, no signup, nothing stored. It will also tell you how to fix whatever it finds.

Public sites only. We fetch six paths and read the homepage head.

The four that matter most

The full check is twelve items, but these four carry most of the weight — and all four are a single file or a single tag.

/llms.txt

A short markdown brief at a predictable path. The one file that most reliably changes how a model describes your site, because it is the only place you get to write that description yourself.

# Your Site
> One sentence on what this is.

## Key pages
- [What we do](https://example.com/about): the short version
- [The product](https://example.com/product): what it does and for whom

## Machine surfaces
- [JSON feed](https://example.com/feed.json)
- [Sitemap](https://example.com/sitemap-index.xml)

Sitemap in robots.txt

The cheapest discovery win available, and the one most often skipped. A crawler that finds your sitemap in the first file it fetches indexes deeper and faster.

User-agent: *
Allow: /

Sitemap: https://example.com/sitemap-index.xml

Machine-readable twins

Every human page gets a JSON sibling, linked from the page itself. An agent that finds the twin stops guessing at your HTML and reads structured data instead — which means it quotes you correctly.

<link rel="alternate"
      type="application/json"
      href="/almanac/el-segundo.json"
      title="Machine-readable version" />

JSON-LD identity

Says who is behind the site in a form an entity graph can absorb. Without it you are a string; with it you are a thing that can be linked to other things.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "name": "Your Site",
  "url": "https://example.com",
  "publisher": { "@type": "Organization", "name": "Your Org" }
}
</script>

The register

Sites that publish the pattern. Every score below is fetched live when this page loads — no hand-maintained list to go stale. To add yours, run the check and send the result through the wire.

PointCast The reference implementation — every human surface has a JSON twin.
checking…

Built because the essay needed an instrument. The checker itself is a Cloudflare Pages Function — it fetches only public hosts, follows a fixed and small number of requests, and never echoes back the content of the site it checked.