Translation work and AI detection
A detector reads rhythm and word choice as signals of the writer’s own voice. A translation’s rhythm and word choice partly belong to the original text, not the translator.
Why translated text is a genuinely different case
A translator is constrained by the source text’s meaning, structure, and sometimes rhythm in ways an original piece of writing isn’t. Some of what a detector reads as uniform or formulaic in a translation may reflect the source language’s own patterns carried across, not the translator’s writing style at all.
Literal versus adapted translation score differently for unrelated reasons
A more literal translation tends to preserve the source’s sentence structure closely, which can read as less varied to a detector regardless of how skilled the translation is. A freer, more adapted translation has more room to vary — that difference is a translation-approach choice, not a signal of anything else.
What to actually check for
The same two signals worth reading in any edit pass — stock transitions and repeated phrasing — are still useful here, since translators reaching for safe, literal phrase-for-phrase choices can produce exactly that pattern. Read those notes as translation-craft feedback, which is a fair use of them.
If a client raises a detector score on translated work
Explain the constraint plainly: translated text is bound to a source in ways original writing isn’t, and a mid-range score reflects that more often than it reflects anything about how the translation was produced.