Why a translated text comes out as machine-made
Translation evens a text out: sentences settle into similar lengths, the vocabulary simplifies and idioms disappear. Detectors read that the same way they read machine writing, so even an honestly translated human text often scores high. With machine translation it goes double, because a model really did produce it.
What translation does to a text
A translator keeps the meaning, but rhythm and structure shift with the target language. The result is more even than the original.
Machine translation goes further still — it picks the most frequent option, so the output is predictable almost by definition.
Why a detector cannot tell
The tool does not know the text came from a translation. It sees a finished text and, in it, properties it knows from models.
Telling translation apart from generation would require the original, which the detector does not have.
How to judge such a text
Ask for the original. If it exists and matches, the question is answered regardless of the score.
With machine translation, the real question is whether it was allowed and whether anyone went over the output.
FAQ
Does running the original through a detector help?
Yes, that is the most useful step. The original's score says more than the translation's.
Is machine translation the same as writing with AI?
Technically both are a model. Rules often treat them differently, which is why it pays to have it written down.
Can it be avoided when translating?
Only by editing that gives the text its rhythm back. But rewriting a translation for the sake of a detector is a bad reason.
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