Paste in some text. We'll point out the bits that read like ChatGPT — em dashes used a bit too freely, the word "delve" turning up, that "it's not just X, it's Y" rhythm. The methodology comes from Wikipedia's editors, who've been sniffing this stuff out for years.
We show the evidence, not just a percentage
A telltale is the small thing that gives someone away — the smudge of lipstick, the tell at a poker table. That's what we look for in text: the small habits AI doesn't realise it has.
Wikipedia editors started noticing AI-written submissions piling up — articles that sounded authoritative but had odd habits in common. They wrote a guide to what they kept seeing.
Most AI detectors give you a percentage and leave you to trust it. We show you exactly which patterns we found and why they count. You can argue with the result — that's the point.
An AI model picks the most probable next word, again and again. The result reads smoothly but always seems to reach for the same handful of words, the same sentence shapes. Here's what we look for:
Plenty of human writers use these patterns. Someone might genuinely love em dashes. A journalist might write "in conclusion" out of habit. One pattern on its own means very little.
What gives AI away is when several of these show up together, often, and the writing has none of the things a human usually leaves behind: a specific anecdote, a moment of doubt, a slightly weird word choice, a voice that changes with the mood. So please don't use Telltale to accuse anyone of anything on its own. Treat it as one piece of evidence among others.
We've also built a free image metadata checker that uses the same honest approach — it finds AI generator signatures in image files without pretending to detect what it can't.