0 of 0 case studies shown
Field survey · attempts, tagged by evidence
Every entry here is a strategy an educator or institution actually ran to reduce student AI use — not a proposal, and not just an opinion that it probably helped.
Entries are tagged by whether there's concrete measured evidence it worked, so evidenced results and untested attempts stay clearly distinguished — use the Evidence filter to see just one or the other.
Putting a student in a supervised room — a proctored written exam or a live oral assessment — changes the opportunity to use AI, not necessarily the desire to. It's an escalating, imperfect arms race (students have been caught using hidden earpieces and smartwatches even under supervision), and it carries real surveillance and equity costs of its own. Several sources make this case directly: "In-Person Classes Aren't Safe From AI Cheating Boom" (Inside Higher Ed), "The AI-resistant classroom is a myth" (eSchool News), the Canadian Journal of Learning and Technology's piece on the academic-integrity "arms race", and a broader ethics literature on algorithmic proctoring.
The gold standard for this catalog is evidence that students changed their own AI use by choice — because a redesigned assessment made reaching for it pointless or unappealing — not evidence that they were simply prevented from trying. Entries below marked with an asterisk (*) work mainly through supervision rather than through changing that underlying desire, and are kept as documented data points, not as an endorsement that supervision is the fix.
Entries marked with a dagger (†) are tests of AI-content detection tools, not strategies that change the task or the incentive, and none of them reports an effect on student AI use. They are kept as documented data points on why detection is a weak basis for any strategy: detectors vary in accuracy, tend to do worse on newer models, and can wrongly flag human writing — in one study, more than half of essays by non-native English writers. Because they aren't interventions, they carry a "Detector study" tag instead of an evidence tag.