Detectors are noisy. They are biased against multilingual writers, they hallucinate confidence, and they reduce a complex teaching question to a single percent. They are still useful, but only as the third layer of a three-layer approach. The first two layers do most of the work.
Layer 1: design assignments AI cannot trivially do
The fastest way to stop chasing AI cheating is to write prompts the AI cannot answer well without a real student behind it. That means anchoring the writing in something only the student has access to.
- Reference specific in-class moments: a discussion thread, a lab result, a field trip observation
- Require process artifacts the student must produce over multiple days
- Ask for personal stance with evidence drawn from a shared class text
- Use timed in-class writing for high-stakes summative work
A practical test
If you can paste the prompt into a chatbot and get a B-grade answer in 20 seconds, the prompt needs rewriting.
Layer 2: make the writing process visible
TeacherAI's Research Paper Writer logs the full writing trail: outline, drafts, paste events, AI-assist calls, time on task. When the process is visible, you rarely need to guess about the product. Three signals tell you a lot:
- A finished paper with no draft history. That is the strongest signal there is.
- A single large paste event that matches the final text. Worth a conversation.
- A writing pace inconsistent with the student's prior work. Worth a conversation, not an accusation.
Layer 3: detection as a conversation starter
Use a detector the way a doctor uses a blood test: it raises questions, it does not deliver verdicts. TeacherAI's AI Detector returns sentence-level highlights and a confidence range, not a fake-precise score. When something flags, the move is the same: invite the student to walk you through how they wrote it.
What a useful detector looks like
- Sentence-level highlights, so you can see which parts feel off
- Confidence ranges, not single percentages
- Calibrated against multilingual student writing, not just polished native-English prose
- Treated as one signal among several, never the basis for an integrity charge on its own
When you do open the conversation
Lead with curiosity and the artifacts. Show the process log. Ask the student to talk you through their thesis. The goal is not to win a case; it is to find out what the student actually learned and to reset the relationship around the work.
Policy language you can borrow
A short, clear policy beats a long, fearful one. Three lines that work:
- AI is allowed for brainstorming, vocabulary, and feedback. It is not allowed to write paragraphs you submit.
- All major writing is done with process turned on. We can see the trail.
- If something is flagged, we will talk about it together before anything goes in your record.