AI detection for teachers
A practical guide — not a product pitch. Written for educators who need honest advice, not another "top 10 detectors" list.
Dr. Monica's full column on this topic is coming soon. This guide covers the essentials until then.
The golden rule
Never accuse a student of using AI based on a detector score alone.
Detectors produce false positives — especially for non-native English speakers, students who write formally, and anyone whose style is unusually consistent. A high AI score is a reason to have a conversation, not a reason to fail someone.
What detectors can do
- Flag text that might be AI-generated
- Help you prioritise which submissions deserve a closer look
- Start a conversation: "Can you walk me through how you wrote this?"
What detectors cannot do
- Prove cheating
- Distinguish "AI wrote it" from "AI helped edit it"
- Work reliably on short answers (<100 words)
- Stay ahead of humanizer tools indefinitely
Recommended tools for classrooms
- GPTZero — free tier, widely known, LMS integrations. Overconfident but useful as a first pass.
- Sapling — less hype, sensible scoring. Good if your school allows it.
- HumanProduced checker — free, unlimited, runs in browser. Rough estimate only — good for your own curiosity before escalating.
A better workflow than "run detector → accuse"
- Design assignments that are hard to fully AI-generate (personal reflection, in-class drafting, process portfolios)
- If something feels off, run a detector as one data point
- Talk to the student. Ask about their process. Request drafts or notes.
- Compare to their previous work — sudden style changes matter more than any score
- Only escalate to formal academic integrity proceedings with multiple evidence types
False positives hurt real students
This isn't theoretical. Students have been falsely accused because a detector flagged their legitimate work. The reputational damage — and the message it sends about trust in your classroom — is real. Use these tools carefully.