International students and AI detection
Published evaluations keep finding elevated wrong-flag rates for writing in a second language. The mechanism is known, and so is what moves it.
If you write in English as a second language, the vendor's published average understates what you are likely to experience. This turns up repeatedly in published evaluations and it is not controversial.
Worth knowing the mechanism, because it is not about mistakes and it is not about vocabulary size.
Why learned English scores higher
Detectors read regularity. Two properties in particular: how predictable your word choice is, and how much your sentence length varies.
English learned formally tends to be applied more consistently than English acquired by immersion. The construction you were taught is the one you reach for, the register stays even, and idiom, which is the most irregular thing in any language, appears less often.
Consistency is exactly what these tools read as machine-like. So the more correctly and evenly you write, the closer you sit to the generated end of the scale.
Our corpus puts human academic writing at a within-document sentence-length standard deviation of 8.7 words against 6.4 for generated text. Second-language academic writing sits between them for reasons that have nothing to do with how it was produced.
Two things that make it worse
Language editing services. Widely recommended, often required, and they smooth prose by design. Smoothing removes variation, so the service improves your writing and raises your score at the same time. See does Grammarly get flagged as AI.
Formal register. International students are frequently taught to write more formally than native-speaker peers, because formality is safer when you are unsure. Formal is conventional, conventional is predictable, and predictable is the signal.
The scale of it
Turnitin publishes a wrong-flag rate on human writing of 4%. That average is across all writing. In a cohort where many people write in a second language, the rate you actually meet is higher than the published one.
Ours is 0.4%, which is 99.6% specificity, measured across 15,900 documents of real academic writing.
What moves it, without changing your English
None of these require you to write worse, and none of them touch what you are saying.
Vary sentence length deliberately. This is the strongest signal and the easiest thing to change. Find three sentences of similar length, split one, merge another pair. A four-word sentence next to a thirty-word one does more than any amount of word substitution.
Cut connective openings. "Moreover", "Furthermore", "Additionally". These are taught explicitly in academic English courses and they open 2.4% of human sentences against 5.8% of generated ones.
Keep the specific detail. Name the method, give the number. Generalising is the habit that reads as generated, and it is the habit people fall into when writing carefully in a second language.
Full list in what makes text read as machine written.
Checking before it matters
The only useful moment is before submission, and it takes a few minutes.
Ours is free and unlimited with no account, takes .pdf, .docx and .tex, has no word cap, scores every paragraph separately so you can see which ones sit highest, and never stores your text.