AI and cheating: what to do instead of banning it

[ INTEGRITY ]

What students and teachers report

59% of teens think AI cheating happens at least somewhat often at their school (Pew Research Center, Feb 2026). 41% of K-12 students and young adults in a Gallup and Walton Family Foundation survey said most or all of their classmates use AI when they are not supposed to. And 71% of teachers say student AI use adds to the burden of judging whose work it is (CDT, 2025).

What students do with AI answers varies. Of teens who use AI for schoolwork, 63% get answers from it: 35% fix or improve the answers, 31% rewrite them to sound like themselves and 25% use them as-is (Common Sense Media, Aug 2026).

The deeper problem is skipped practice

Cheating is a rule problem. Skipped practice is a learning problem, and it can happen even when a student breaks no rule. In a 30-month study of 26,811 Chinese students in grades 7-12, adopting AI raised homework scores 18% and cut monthly closed-book exam scores 20% within six months; students who kept spending normal time on homework lost little (a working paper; not randomized). Design for practice, and much of the cheating problem shrinks with it.

Make the rule clear for each assignment

Students cannot follow a rule they cannot see. Put the AI level on every assignment: no AI, AI for hints and feedback, AI with disclosure, or AI as part of the task. Howard County, Maryland made assignments AI-free unless a teacher says otherwise when it gave high schoolers Gemini. Our syllabus AI statement gives wording for each level.

Design assignments around the process

• Start drafts in class, so the first attempt is the student's.
• Ask for a short reflection: the question the student asked AI, one response they changed or rejected, and what they can now explain alone.
• Use short oral explanations: a student explaining their own work for two minutes shows understanding quickly.
• Tie tasks to class discussions, local sources or the student's own data.
• Grade revisions and reasoning, not only the final product.

Check what students can do alone

Short unaided checks a few days after practice do two jobs: they show what a student can do without help, and they strengthen learning. Classroom quizzing improves learning, g = 0.50 across 222 studies.

Use process evidence, not detectors

Seven AI-writing detectors flagged on average 61% of essays by non-native English writers as AI-written. A detector score is not evidence. Process evidence is: drafts, version history and, in Nexus, class chats that show how a student worked. Nexus does not detect cheating or AI-written work; students are told when a class chat can be reviewed.

Treat misuse as a teaching problem first

When work looks AI-written, start with a conversation: ask the student to explain the work or redo part of it in class. In 2023-24, 64% of teachers of grades 6-12 reported students disciplined over generative AI, up from 48%, and students with disabilities and English learners were more exposed. A consistent, evidence-based process protects students and teachers.