Why a borrowed baseline gives you the wrong answer
Telling you how AI-flavoured your writing reads means comparing it to real writing in the same language. Comparing it to the wrong language does not give you a smaller signal, it gives you a meaningless one dressed up as a real measurement.
NeverPrompted measured its English baseline from contemporary English prose, and every document in that corpus was verified to be English before it was used.
The rewrite itself works the same way: it targets the phrasing habits that read as AI-generated in that specific language, not a translated list of English ones.
If the language cannot be worked out
When the engine cannot confidently tell what language a document is in, it says so rather than guessing and quietly measuring against the nearest match. A wrong-language result looks exactly like a right one until you notice everything you write scores the same, which is the sign something upstream is broken.
You can also set the language explicitly before running the check, if you know it and the engine is unsure.
Wherever this page describes a result: a detected mark is not proof of authorship, and an absent mark is not proof of human authorship. NeverPrompted's on-device rewrite can reduce detectable evidence but cannot guarantee defeating a vendor's undisclosed watermark, on any tier.