GenAI in the lecture hall: rules students actually use

Ask ten lecturers what is allowed with ChatGPT in their course and you will get ten different answers. That is not unwillingness — universities genuinely struggle with policies that are hard to find, unclear, or contradictory. The bill lands with the student, who has to guess the rules course by course.

In our research at Leiden University we flipped the perspective: start with the student, not with the rules. Through seventeen interviews and three workshops — with students and teachers — we designed guidelines for responsible GenAI use in computer science education, iteration by iteration.

What turned out to work

• Scales instead of one rule. No two courses are alike. The guidelines use three settings: prohibited, allowed under conditions, and free use — always with a duty to verify the output.
• Do's and don'ts with examples. Using GenAI to learn or to raise your productivity: fine. Copying output straight into your submission: not.
• Justification with reflection. Students record which model they used, what for, and — crucially — what they made of the output. That reflection keeps alive the critical thinking teachers fear losing.
• A why. Rules without a rationale get circumvented. Rules that explain what is at stake get carried.

The most striking finding: the guidelines students proposed themselves proved, with minor adjustments, acceptable to teachers. Start with the user and you have far less to enforce.

Why this matters beyond the university

Every organisation working with GenAI faces the same question: how do you make rules people actually follow? The answer from our research keeps coming back — start with the people who have to work with them, let the rules scale per context, and build in reflection rather than control.

The full paper was published by IEEE: Designing Guidelines for Responsible GenAI Usage in Computer Science Education.