Skip to content
Free tool

Could an answer be lifted out of this page?

Paste one public URL. A language model reads the page against a fixed 28-check rubric and scores whether the content is shaped to be quoted and attributed, then names the checks it failed.

What comes back is an assessment, not a measurement. Nothing here observes an engine, so nothing here can tell you whether you are cited. It tells you whether the page is in a state that makes citation possible, which is the part you control.

One public page. Not a whole site, and not a page behind a login.

COST · FREE · NO ACCOUNT · NO EMAIL · THREE RUNS PER HOUR

What it checks28 checks · three dimensions · 56 points

Three things, in the order they can break.

Access comes first, because a page a crawler cannot read scores nothing on the other two. Then structure, then whether the page says who is speaking. Every check returns pass, partial or fail, worth two points, one, or none.

Content structure for extraction
Whether a self-contained answer sits in the first 300 words and under each heading, whether definitions are written as plain declarations, and whether frameworks, tables and lists are in elements a parser can lift rather than buried in prose.
DIMENSION E · 10 checks · 20 points
Entity and schema depth
Whether the page says who is speaking in a form a machine can resolve: one canonical name across schema, title and body, Organization and author markup with sameAs, JSON-LD rather than Microdata, and schema answers that match the visible copy.
DIMENSION F · 10 checks · 20 points
Access for AI crawlers
Whether the content is reachable before any of the rest matters: present in the initial HTML, not behind a click, not blocked in robots.txt for GPTBot, ClaudeBot or PerplexityBot, with valid schema and a visible date.
DIMENSION G · 8 checks · 16 points
How it scores

The Caldrin AEO readiness rubric, v1.

Twenty-eight checks, each scored pass (2), partial (1) or fail (0), for a raw score out of 56 that is reported as a percentage. The bands are 86 and above citation-ready, 68 to 85 near-ready, 50 to 67 significant gaps, and below 50 not citation-ready.

The rubric is ours and the bands are a judgement call, not an industry standard. They are drawn from what we see get extracted into generated answers: a direct answer in the opening, headings that describe rather than tease, frameworks and tables in real elements, and one consistent entity name that a model can resolve. Nobody has published a validated threshold for any of this, and neither have we, so read the failed checks rather than the number.

THE SCORE IS MODEL-GENERATED AGAINST THIS RUBRIC. IT IS NOT PRODUCED BY CAUL AND IT IS NOT A MEASUREMENT OF ANY ENGINE.

Stated in advance

What this check cannot see.

Five limits, and they apply to every run rather than to edge cases. A free tool that overstates itself costs more credibility than it earns, so these sit above the result rather than under it.

It does not watch any engine.
No prompt is issued and no answer is captured. The tool never learns whether ChatGPT, Perplexity, Gemini or AI Overviews cited this page, so nothing it returns is a citation share or a visibility figure.
It cannot see authority.
Being retrieved at all is a different failure from being retrieved and not chosen, and it is bought differently. This rubric only reads the second one. A page can score well here and still never be retrieved, because links, rankings and entity presence are outside what one page's HTML can show.
It reads the raw HTML, not the rendered page.
Content injected by JavaScript is invisible to the fetch, which is deliberate: that is roughly what a crawler that does not execute scripts sees. It does mean a page can be marked down for content a browser would display.
It is one model's reading, and it varies.
The same page run twice can score a point or two apart, because a language model is doing the scoring. Treat a score as a band, not a figure, and treat a two-point move as noise.
It checks one page.
Site-wide patterns, internal linking and whether this page is the one your category actually needs are all outside the frame.
Method

The disclosure that travels with the score.

instrument
A language model reading extracted HTML. Not Caul, and not a measurement instrument.
what was measured
Nothing is measured. One public page is scored against the Caldrin AEO readiness rubric v1: 28 checks, 56 points.
how
The page is fetched once as raw HTML with its robots.txt, reduced to headings, schema, meta, a text sample and structural counts, and scored by a language model against the fixed rubric. Nothing is stored.
over what window
The moment you press run. The score describes the page as fetched, and it goes stale the next time the page changes.
what this cannot tell you
It never observes an engine, so it cannot tell you whether you are cited. It cannot see authority, which is the other half of why pages are not retrieved. It reads raw HTML, so JavaScript-rendered content is invisible to it. Scores vary by a point or two between runs, because a model is doing the scoring.

This checked one page. The report measures your category.

The free AI visibility report runs a fixed prompt set across the engines and counts how often you are named against competitors you name. That is the measurement this tool cannot make, and it is free too, delivered on a call so the findings get explained.

Get your free auditWhat is in the report

Free. The $1,500 is what this sweep is priced at when it is sold, not a survey of anyone else’s rates.