Knowledge · Ira Berg, 12 November 2026
Recognising AI use in class without surveillance
In short: AI detection software for texts is unreliable and feels like surveillance to young people. What works better are task formats that require thinking and tools that show the students themselves how much of their own thinking is in their work.
Why text detection alone is not enough
In recent years many schools have tried software that claims to recognise AI texts by their style. The problem: these programs are often wrong in both directions. They flag texts that students wrote themselves as AI generated and miss real AI texts that were slightly reworded. If you rely on them alone, you get false suspicions and a damaged relationship with your class, but no reliable answer.
What works instead
- Assess the process, not just the result: drafts, interim stages and short oral questions reveal more than a finished text.
- Tasks AI cannot simply solve: links to the class discussion, to specific sources from the lesson or to the students' own lives make plain copying harder.
- Oral follow up: a short question about their own text quickly shows whether a student understood the content.
- Build in reflection: a question like “What did you take from the AI and what did you add yourself?” makes AI use transparent without accusations.
The other way: let students see it themselves
Monkey Score takes a different route than control from outside. The extension makes students' own share of thinking visible while they work with AI, as a mirror for them. Teachers see nothing, there is no dashboard and no data transfer. The effect comes from students seeing their own behaviour, and they often start asking better questions on their own. More in Using Monkey Score in class.
A clear position in advance helps
A lot of conflict arises because rules on AI use are only discussed after a suspicion. It is better to agree with the class at the start of the school year when AI is allowed, when it is not and how its use is declared, for example with a short note at the end of a piece of work. The checklist for AI answers in class and the template AI rules for the classroom help with this.
FAQ
Can AI detection software reliably recognise AI texts?
No. These tools are often wrong, both with real AI texts and with student texts that are wrongly flagged as AI generated. They are not reliable.
Is it allowed to run student texts through an AI detector?
That depends on your country, region and school, and often on data protection rules, because texts are sent to external providers. If in doubt, ask your school leadership or data protection officer.
Prompt to try
Copy this straight into ChatGPT, Gemini or Claude:
“Write five oral follow up questions I can use to check in two minutes whether a student has really understood the content of their homework rather than just copied it.”