I’ve read enough comparison posts to know that ‘it depends’ is not a useful answer when you’re trying to build a defensible review process for an Academic Integrity Committee. So let me be more specific about what I actually need.
The tool has to have a lower false positive rate on formal academic writing. I teach composition, which means my students write in a structured, precise, citation-heavy style. That style trips detectors. A tool calibrated against blog posts or casual web content is useless to me.
It needs to give me something other than a percentage. A percentage alone is not defensible evidence in an appeal. I need the tool to highlight specific passages, show me what it flagged, give me something I can actually reference in documentation.
And it needs to handle multilingual writers. A significant portion of my students are international. Any tool that disproportionately flags non-native English writing is a liability, not an asset.
I’m not asking what detector is most popular. I’m asking which one actually fits this use case. If the answer is that none of them fully do, that’s useful information too.
The honest answer to your question is that no current tool checks all three boxes you described. The ones with good passage-level detail tend to have higher false positives on formal writing. The ones calibrated for academic text tend to give you less granular output. You’re probably looking at a workflow that uses more than one tool for different parts of the assessment.
On the defensibility point: I’ve started treating detector output the way I treat a plagiarism flag. It’s a starting point for investigation, not a finding. The documentation I build is the conversation with the student, the comparison to their prior work, the in-class writing sample. The tool score appears nowhere in my formal report. That’s deliberate.
The passage-level highlighting requirement is the most important thing you said. A percentage score with no underlying explanation is not usable in any kind of procedurally fair review. For what you’re describing, you need a tool that shows you exactly which text segments it flagged and with what confidence. That narrows the field considerably.
The multilingual calibration problem is real and I don’t think any tool has fully solved it. What I’ve seen in practice is that tools trained primarily on native English academic writing systematically over-flag ESL writers. It’s not a calibration tweak, it’s a training data problem. Worth asking vendors directly how they’ve addressed this before you commit.
From a non-education standpoint, the due process framing you’re applying is exactly right. Any system used to make consequential decisions about people needs to produce auditable, explainable output. A black-box percentage score doesn’t meet that bar. The fact that most of these tools aren’t built to that standard is a real problem.