AgentBeacon

Is your site legible to AI agents

Search is being replaced by machines that read your pages for people. Paste a web address and get a score out of 100, a grade and a plain list of what to fix.

Free. No sign up. 50 audits an hour.

The problem

Same page. This is what the machine actually gets.

An agent does not run your site the way a browser does. It reads the HTML that comes back from the server. If your content only appears after JavaScript runs, the reader sees an empty room.

Server rendered site what the agent receives 
<h1>Ada Lovelace, mathematician</h1>
<p>The first person to publish an
algorithm intended for a machine.</p>
<a href="/notes">Read the notes</a>
Content that survives 95%

The content is right there in the first response. A reader gets the whole page.

Client rendered single page app what the agent receives 
<div id="root"></div>
<script src="/app.js"></script>

no readable content

Content that survives 0%

Everything is drawn later by JavaScript. Before that runs, there is nothing to read.

What gets checked

Twenty one checks across four categories

Each check carries a weight. Heavier checks move the score more. The number beside each one is its weight within the audit.

Discoverability

25% of score

Can an agent find your site and the files that describe it.

  • robots.txt present and parseable 3
  • Which AI agents your robots.txt allows reported
  • XML sitemap present 4
  • Sitemap has entries and looks maintained 2
  • RSS or Atom feed advertised in the page head 2

Structure

30% of score

Is the meaning of the page laid out in a way a machine can follow.

  • llms.txt present at the site root 8
  • llms-full.txt present 2
  • llms.txt follows the spec and its links resolve 5
  • Valid JSON-LD structured data 6
  • Structured data uses core schema.org types 4
  • One h1 and no skipped heading levels 5

Extractability

30% of score

Does the actual content survive without a browser running scripts.

  • Readable article content can be extracted 8
  • Extracted content is substantial 4
  • Extracted title matches the page title 3
  • Content exists without JavaScript 15

Identity

15% of score

Does the page say clearly what it is and who it belongs to.

  • Page has a title tag 3
  • Page has a meta description 3
  • Canonical URL declared 2
  • Open Graph title, description and image 3
  • Language declared on the html element 2
  • Contact or about route discoverable 2

How the score works

A number you can trust

No black box. Here is the whole method, plainly.

1

Each check earns credit

A pass is worth 1, a warning is worth 0.5, a fail is worth 0. Reported items are never scored.

2

Could not check is excluded

If a check could not be run it is left out of both the top and the bottom of the fraction, so it is never punished as a failure.

3

Categories are weighted

Each category is a weighted pass rate. The overall score is the weighted mean of the categories that actually ran.

Category weights

  • 25% Discoverability
  • 30% Structure
  • 30% Extractability
  • 15% Identity

Grades follow the score: A at 90 and above, B at 75, C at 60, D at 40, F below that. Every response carries the scoring version it was produced under, so a number always has its method attached.

Use it as an API

One endpoint, no key

Free, 50 audits an hour per address. CORS is open. One synchronous request, no polling. The JSON comes back in the shape below.

Request 
GET https://api.agentbeacon.alcun.dev/audit
    ?url=https://example.com
Response 200 OK 
{
  "url": "https://example.com/",
  "finalUrl": "https://example.com/",
  "scoringVersion": "1",
  "score": 91,
  "grade": "A",
  "categories": {
    "discoverability": 82,
    "structure": 93,
    "extractability": 100,
    "identity": 87
  },
  "unavailable": [],
  "checks": [ ... ]
}

Full parameter list and error codes are on the API docs page.

Questions

Before you ask

What does the score actually measure

How legible your site is to automated readers, the agents and assistants that increasingly sit between people and the web. It is not an SEO score and it is not about how the page looks to a person.

Do you store my site or the results

No. Each audit is a single request against the live page. Nothing is saved and there is no account to create. Results are cached briefly so a repeat audit of the same address is fast.

Why did a check come back as not checked

Sometimes a check cannot run, for example if a file could not be fetched. Rather than guess, we leave that check out of the maths entirely so it never counts against you.

Is llms.txt really worth adding

It is a small, cheap file that gives assistants a hand written map of your key pages. It carries real weight here because it is one of the clearest signals you can send that your site is meant to be read by machines.

Can I run this on a staging site

Yes, as long as the address is reachable from the public internet. Pages behind a login or a firewall cannot be read, by us or by any agent.