AI detection for teachers and examiners

Exams, essays and take-home work

In-person exams with locked devices are a design problem, not a detector problem. Take-home essays are where style tools show up — and where false positives hurt most.

Run the detector. Open sentence-level highlights. Compare flagged spans to earlier work from the same student. Sudden fluency jumps matter more than a single percentage.

How to open the conversation

Ask about process: drafts, notes, time spent, sources. Ask the student to explain a flagged paragraph in their own words. A score is a reason to talk; it is not the talk.

Do not present the tool as omniscient. Students already know detectors err. Credibility comes from showing the passages and the limits.

Policy before tooling

Clear course rules beat surprise enforcement. Say whether drafting with AI is allowed, restricted or banned, and what disclosure looks like. Then the detector supports the policy instead of replacing it.

For integrity procedures, keep the report, the extract and your notes. Hybrid tiers ("Humanized AI", "Too close to call") are not "AI-generated" and should not be written up as such.

Questions

Can an AI detector prove a student used ChatGPT?

No. It estimates how much the writing resembles generated prose. That is not proof of a specific tool or of misconduct.

Are ESL students more likely to be flagged?

Yes, on average. Second-language writing often has more regular cadence and narrower vocabulary. Build review time for that bias into your process.

Should I use a detector on exam scripts?

Only if the exam conditions make outside AI plausible. For sealed in-person exams, spend energy on exam design, not on post-hoc scoring.

Continue on IA Checker

• AI Detector
• IA Checker vs GPTZero
• For recruiters
• For students
• Extract text from PDF