About
This site answers one question badly on purpose: does a web page look like it came out of the current generation of AI build tools, or like somebody sat down and made decisions? It is a joke with a working engine underneath.
There is no AI in it
Nothing here calls a language model. The scanner fetches a page once, server-side, and runs 86 hand-written detectors over the raw HTML. Each one is a pattern with a weight and a reason, and all of them are written down in public. That makes the tool fast, free, repeatable, and confidently wrong sometimes — which is the disclaimer and also the joke.
The score is not a percentage and not a probability. It is strength of evidence: the weighted tells found, damped by the craft found, mapped through a single logistic curve. 43/100 means roughly 26 points of net evidence, not “43% AI”.
Three rules keep it honest
Fingerprints only count from the machine surface. Asset hosts, meta tags, comments, data attributes. A blog post about v0 was not built by v0, and a customer logo named lovable.svg is not a fingerprint.
Craft damps the aesthetic score. Looks are the weakest evidence in the whole set — anyone can reach for a gradient deliberately. The more real craft a page shows, the less its visual defaults are allowed to say about it.
Absence checks are gated. A missing favicon on a page with no modern build signals is period-accurate, not a tell.
What it cannot see
One request, no JavaScript executed, capped at 2.5 MB. So it never sees how the page behaves, how it was actually built, who built it, whether the code behind it is any good, or whether the thing it sells works. A page can be excellent at all of that and still score high, and a page can score low and be terrible. Using a popular framework is not a crime.
We publish 28 worked examples with the full evidence for each, so the scoring can be argued with rather than trusted.
Scan this site
Fair is fair. Paste isitvibecoded.xyz into the box on the front page — the detectors do not know they live here, and they are not graded on politeness.
Contact
Corrections, false positives, and detectors we should add: a7madinquiries@gmail.com. If a detector fired wrongly on your page, say which one and send the URL — that is the most useful mail we get.