How do professors actually go about checking for AI use beyond running a detector score?
I’m asking specifically about the human judgment side, since I think the tool is only part of the picture. Curious what specific things experienced professors look for, whether that’s in the writing itself or in how a student talks about their own work when asked directly.
From teaching for several years now: argument coherence is the biggest one for me personally. AI-assisted writing sometimes states a thesis cleanly but doesn’t actually build a sustained argument toward it, the paragraphs feel more like separate points than a developing case.
Specificity is the other big one. Generic examples and vague claims that could apply to almost any topic in the field are a tell, regardless of what a detector score says. Genuine engagement with the material tends to produce more particular, less swappable content.
Asking a student to walk through their own argument verbally is genuinely effective and underused. It’s hard to fake explaining reasoning you didn’t actually do, even if the written product reads fine on its own.
As a TA I’ve found inconsistency between a student’s in-class writing and their submitted essays the most reliable single signal, more reliable honestly than any tool score I’ve seen used.
Worth noting this human-judgment approach has its own false positive risk too, a nervous student explaining their work poorly in an unscripted conversation isn’t proof of AI use either. It’s a useful signal, not a definitive one.