What the percentage from a detector means and what to do with it
The score is an estimate of how much the text looks like something a model wrote — not the probability that someone cheated. Above 50 it is worth going through the marked sentences, below 20 the text is usually fine, and in between lies a grey zone that no number resolves; reading the text does.
What the number measures and what it does not
The detector does not know who sat at the keyboard. It measures how much the text resembles what language models usually produce — even sentences, cautious phrasing, few surprises.
A text written by a person who writes precisely and tersely can come close to that description. And conversely, a generated text that was then rewritten by hand moves away from it.
Thresholds you can actually work with
Below 20 there is usually nothing to deal with. Above 50 it makes sense to look at the sentences the tool marked. In between, the number alone does not decide.
Set the thresholds according to what you are checking. In a bachelor's thesis the grey zone has different consequences than in a product description.
The second number: deviation from your own hand
Alongside the general score, DetekceGPT also shows how far the text differs from your older texts. This number does not say "this is AI" but "this is not how you usually write".
For your own texts this is often the more useful figure, because it compares against something concrete.
FAQ
Does 100 % mean AI wrote the text?
No. It means the text has all the traits the detector looks at. You cannot turn that into certainty.
Why does the score change when I submit the text again?
If you edited the text in the meantime, its measurable properties changed too. For an unchanged text the result stays the same.
What do I do with a result around fifty?
Read the marked sentences. More is decided there than on the scale.
Try it on your own text
Paste a text or upload a file and look at the score and at the specific sentences that came out suspicious.
Run a detection