Why texts by non-native writers score worse with detectors
Detectors measure, among other things, how predictable the words are and how much sentence length varies. Someone writing in a language that is not their first picks commoner words and simpler structures — which is exactly what machine writing looks like. It is not the author's fault and not an intent, but a property of the measurement that has to be taken into account before a number becomes a conclusion.
Why it comes out that way
A non-native writer reaches for words they are sure of and structures they trust. The resulting text is even and predictable — two properties detectors associate with models.
Writing to a template, translating, and academic prose written strictly to a method all have the same effect.
Who it affects
Students writing in a foreign language, colleagues from another country, authors still learning to write academically. Often the people with the least room to defend themselves.
The risks stack: a foreign language plus an academic text plus a short word count gives a high score even for honest work.
How to work with it
Do not decide on one number. Read the flagged sentences and ask whether they are the passages where the author should be visible.
A comparison with the same author's earlier texts helps. If they always write like this, it is not a change — it is their style.
FAQ
Is this a flaw in the detector?
More a consequence of what it measures. Predictability cannot tell whether a model or a cautious author is behind it.
Can an author avoid it?
Not really, and rewriting a text to move a number is pointless. The fix sits with whoever reads the result.
Does it apply beyond English?
Yes, to foreigners writing Czech just as much as to Czechs writing English. A language you are not at home in shows in the text.
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