How to tell if students used ChatGPT: a fair approach

If you set written homework, you've probably wondered whether a polished response was genuinely the student's own work. The instinct is to paste it into an "AI detector", but that's exactly the approach that's been getting schools and universities into trouble.

Why text detectors let you down

Detectors that score how "AI-like" prose reads rely on statistical patterns that vanish as models improve and that disproportionately flag simpler or non-native English. A Stanford study found seven detectors flagged non-native essays as AI up to 61% of the time. Several universities disabled their detectors for exactly this reason.

Worse, a single false accusation can do real harm: to a vulnerable student, and to your relationship with the class.

Look at how the work was written

A fairer, more defensible signal is the writing process. Genuine drafting looks like typing in bursts, pausing to think, deleting and revising, over a believable stretch of time. AI-assisted shortcuts often look different: a large block pasted in at once, an essay finished implausibly fast, or a robotically even typing rhythm from transcribing.

None of these prove anything on their own, but together they tell you which submissions deserve a conversation.

Make it a conversation, not an accusation

Whatever tool you use, the goal isn't a verdict. It's better context for your professional judgement. Lead with curiosity: ask the student to talk you through how they approached the task. Behavioural evidence gives that conversation a concrete, fair starting point.