A generic prompt invites a generic answer, which is exactly what a language model produces best. The fix is to make the task specific and applied. Ask students to use course concepts on a particular case, a local problem, a data set you provided, a reading discussed in class, or their own field placement. Ask them to apply, decide, and justify rather than to summarize what is already known.

When the assignment depends on details that only someone in your course would have, a tool with no access to that context cannot complete it well. The student has to engage with the material to do the work. As a bonus, applied tasks are usually more engaging and more aligned with the outcomes you care about.