Why AI Assistants Cannot Read JavaScript
The dedicated AI crawlers — GPTBot, ClaudeBot, PerplexityBot and their equivalents — do not run JavaScript. They fetch your URL, read the HTML the server returns, and stop there. If your page builds its content in the browser after load — with a React, Vue or Angular app that fetches data on mount — then the document those crawlers receive is an empty shell. The words a human sees never reach them. This tool exists to show you exactly what that shell contains.
Googlebot behaves differently. It may render a page in a second pass and index the result, which is why a site with excellent Google rankings can still be invisible to AI assistants. That gap is the reason this checker exists: it measures the first response, the one non-rendering crawlers use, because that response decides whether an assistant can quote you at all.
What Each Check Measures
The score is built from eleven checks, each weighted by how strongly it has been linked to being cited. The heaviest is content in the HTML (20 points): the number of words present in the served document. Below 50 words fails, 150 or more passes, and anything between is a warning. Heading structure (12 points) looks for an H1 that states the page's subject. Structured data (12 points) counts JSON-LD blocks. Title (10) and freshness (10) reward a descriptive title and a declared publish or modified date. Meta description (8), content-to-markup ratio (8), indexability (7), AI crawler access (5), canonical (5) and a declared language (3) make up the rest.
Every check is scored pass, warn or fail. A pass earns its full weight, a warning earns half rounded up, and a fail earns nothing. The points are shown beside each check so you can see where the score was won or lost rather than accepting a single number on faith.
The Research Behind Our Weights
The weights are not invented from thin air. They follow published correlation findings on which page properties actually appear in AI citations. Metadata and freshness correlate at r=0.68, semantic HTML at r=0.65, and structured data at r=0.63. Those three clusters carry the largest share of the score because they are the properties that most reliably separate pages assistants choose to quote from pages they ignore.
Read that honestly: correlation is not causation, and these figures describe patterns across many pages, not a guarantee for yours. A high score makes a page readable. It does not make it cited.
What This Tool Cannot Tell You
This checker reads one URL, once. It does not crawl your site, it does not test your sitemap, and it does not execute JavaScript.
That last limit matters most. Because we do not render the page, we cannot tell you what a real browser would see. A page can score badly here and still look perfect to a visitor — that is precisely the failure being measured. We also cannot tell you whether any AI assistant actually cites you. Being readable is a precondition for citation, not a promise of it. And we cannot see behind a login, past a bot challenge, or into any content that appears only after interaction.
Treat the score as a fast, honest diagnostic. It answers one question: is the text of this page present in the HTML that a non-rendering crawler receives, and are the signals that correlate with citation in place? If the answer is no, you now know what to fix.
Frequently asked questions
What does the AI Readability Checker actually check?
Eleven signals, all read from the HTML your server returns: how many words of body text are present in the HTML, whether an H1 states the page's subject, how many JSON-LD blocks it carries, its title and meta description, whether it declares a publish or modified date, the ratio of visible text to markup, whether a noindex tag is present, which AI crawlers robots.txt allows, whether a canonical URL is set, and whether the html element declares a language. Each is scored pass, warn or fail and shown with its own weight.
Why does it matter that AI crawlers do not run JavaScript?
ChatGPT, Perplexity, Claude and Bing's AI crawlers fetch a URL, read the HTML the server sends back and stop there — they never execute your scripts. If your page builds its content in the browser after load, the document they receive is an empty shell, and the words a visitor sees never reach them. Googlebot is different: it may render the page in a second pass, which is why a site can rank well in Google and still be invisible to assistants. This tool measures that first response.
Is this score a Google metric?
No. It is our own heuristic, not a Google metric, a ranking factor, or a figure any search engine or AI provider uses. The weights follow published correlation findings on which page properties appear in AI citations, but the model is ours. Treat it as a guide to what a non-rendering crawler can see, not as a measurement of how you rank.
Is blocking AI crawlers in my robots.txt a bad thing?
Not at all, and the tool does not treat it as a fault. Blocking GPTBot, ClaudeBot or others is a legitimate choice with real trade-offs — you may be protecting content or bandwidth, and you are also making it impossible for those assistants to read or cite the page. The tool simply reports which agent tokens are disallowed so you can make that trade-off deliberately rather than by accident.
Can it tell me whether an AI assistant cites my site?
No. It cannot see what any assistant says, what it retrieves, or whether it has ever quoted you. Being readable is a precondition for being cited, not a promise of it, and a high score is no guarantee of a single citation. The tool answers one narrower question: is the text of this page present in the HTML a non-rendering crawler receives, and are the signals that correlate with citation in place?