AI Content Detector

Honestly, it depends on the text. On blog posts and articles over 300 words, it does a solid job. On short texts or very technical writing, take the results with a grain of salt. No detector, ours included, gets it right 100% of the time.

Ten things at the document level: Zipf's law conformity, repeated starters, punctuation spacing patterns, sentence length distribution shape, hapax ratio, paragraph uniformity, perplexity, burstiness, vocabulary richness, and word length variation. On top of that, eighteen per-sentence checks looking at dashes, transition words, formulaic structure, and other patterns.

Zipf's law is a pattern in language: the most common word appears roughly twice as often as the second most common, three times more than the third, and so on. AI text follows this curve almost perfectly because it samples from probability distributions. Human writing is messier; we get stuck on words, go on tangents, repeat things we probably shouldn't.

Some websites load their content with JavaScript after the page loads (client-side rendering). Our fetcher grabs the initial HTML, so if the text isn't in there, we can't see it. For those sites, just copy the text and paste it in the Text tab.

Nope. Everything runs in your browser. The text you paste never leaves your device. If you use URL mode, the fetch happens from your browser too (or through a lightweight proxy if CORS blocks it, but the analysis still happens locally).

Yes. Free, no account needed, no limits on how many texts you scan. There's no premium tier or hidden upsell. The URL proxy has a rate limit of 5 requests per minute, but the text mode has no limits at all.

At minimum 30 words and 3 sentences, but you'll get much better results with 300+ words. The statistical patterns need enough data to be meaningful; a two-sentence paragraph doesn't give the algorithm much to work with.

That's the hardest case for any detector. If a human rewrites a third of the sentences, adds personal anecdotes, and removes the obvious AI patterns, detection gets unreliable. The mixed percentage in the results often points to this kind of hybrid content.