AI text detector
Paste a text, or a page address, and Uthentic tells you how likely it is that a language model wrote it — and above all, why. The report doesn't stop at a percentage: it shows you the passages that weighed on the verdict.
The three signals we read
Word predictability
A language model almost always picks the most probable continuation. A text where every word is exactly what you'd expect, line after line, is the first signal: human writing strays from the easiest path far more often.
Uniform rhythm
People writing naturally mix long and short sentences, asides, changes of pace. Generated text tends to keep a steady rhythm and sentences of similar length: very tidy, not very lived-in.
Stock phrasing and recurring formulas
Polite openings, transitions that repeat themselves, conclusions that summarise without adding anything new. These are typical model tics, and a good report flags each of them individually.
Why more than one engine
A single text detector fails in a systematic way: it penalises people who naturally write in a very linear style and rewards someone who generates text and then lightly rewrites it. Uthentic queries several independent engines and compares their readings; when they disagree, the report says so instead of picking a number at random.
One thing worth keeping in mind: no AI text detector is proof of anything on its own. It exists to give you a well-founded question about a text, with the evidence in front of you. That's why text analysis stays separate from image analysis, with its own engines and its own report.