Deter AI Cheating Through Support, Not Detection (opinion)

Last fall, at 11:42 p.m., a student emailed me in a panic asking for an extension on his paper. I gave him another day. The next afternoon, as I was packing up my office, he walked in smiling and handed me the finished essay. “I was freaking out last night,” he said. “I almost had Chat write it. Glad I didn’t.”

He was not confessing so much as sharing a moment of temptation—and relief. It helped me see more clearly that, in the AI era, we need to prevent cheating not by increasing student surveillance, but by increasing student support.

Many colleges have done the opposite. Feeling besieged by AI-assisted cheating, they have hardened the perimeter: stricter policies, zero-tolerance language, AI detectors, locked browsers, keystroke tracking, oral defenses, blue-book exams.

The response is understandable. Surveys have long suggested that academic misconduct is not rare: More than 60 percent of university students admit cheating in some form, and roughly one in five undergraduates admits cheating repeatedly. And now, with generative AI, cheating is not only easier but harder to detect.

So, should colleges double down on detection? Even if AI detectors were reliable, they would still be counterproductive, because they broadcast a message to students: We assume you are cheating.

This surveillance makes the educational relationship adversarial. It frames the classroom as a game of cat and mouse: You try to cheat, we try to catch you. Actually doing the work, at that point, becomes the losing position, where the student simply failed to find a viable shortcut.

In the classroom, that accusatory ambience breeds anxiety. A recent YouGov survey commissioned by Studiosity and reported in Inside Higher Ed found that 75 percent of students who use AI reported stress about being wrongly flagged for plagiarism.

If student well-being is one of higher education’s basic concerns, as it should be, then colleges should worry not only about deterring misconduct, but also about policies that intensify anxiety and alienation. Research on student persistence suggests that emotional stress and mental health strain are major reasons students consider leaving college. Other studies link student burnout and disengagement to weaker academic performance and greater intention of dropping out. And research suggests that when students are uncertain about whether they belong in an academic environment, ordinary setbacks can become evidence, in their minds, that they do not belong in college at all.

An approach centered on support rather than surveillance sends an affirmational message. It tells students: Yes, shortcuts exist, but they are bad bargains. We trust you, we want you to learn and we have designed this course to help you succeed.

In my own classes, I do not presume today’s students are aspiring fraudsters. In frank conversations with them, I have found that they genuinely care about their education. This aligns with the YouGov/Studiosity survey’s finding that only 21 percent of students said they would rely entirely on AI to write their papers even if permitted. Most students are not looking to outsource their thinking to a machine.

To address student cheating, then, instead of trying to deter it externally through threat and suspicion, we should weaken the conditions that tempt students to cheat in the first place. Psychology helps identify the levers that increase or decrease cheating.

One cheating catalyst is pluralistic ignorance: the phenomenon in which students who value integrity become more tempted to cheat when they suspect cheating has become the norm, lest they fall behind their peers.

So a course that emphasizes surveillance does not merely fail to deter cheating; it may de facto encourage it. When pedagogy tilts toward detection over development, it cultivates a classroom ethos of suspicion rather than community.

Course redesign, by contrast, works against pluralistic ignorance by signaling the opposite: that most students are expected to engage honestly, and that the conditions for doing so are actually in place.

Research suggests that cheating is often less a matter of student character than of student circumstance. Scholars such as Eric Anderman and David Rettinger have shown that academic misconduct is strongly shaped by context: pressure, competition, peer norms, students’ motivation and whether a course emphasizes grades over learning. Students are not always “cheaters” as a stable identity; rather, many are sincere learners who make bad decisions under stress, pressure or poorly designed incentives.

Behavioral research helps clarify why those weak moments arise. Temptation to cheat often rests on three conditions: incentive, rationalization and opportunity. Remove or weaken any one of them and the stool begins to wobble. Course redesign can do exactly that.

Incentives to cheat, the first condition, usually stem from desperation—as in the case of my student. When it is late at night and a student faces the choice between submitting an AI-generated paper or receiving a zero, cheating can seem like the rational option. As one professor recently put it in Inside Higher Ed, “It’s easy to blame students, but when it’s 9:45 p.m. and you have an assignment due in 15 minutes and you just finished a shift at your job and you’re exhausted, it’s just too easy and too tempting to take that question and feed it into AI.”

One solution is to build in pressure valves: two no-questions-asked extension tickets, one substantive redo per semester or a grace window after the formal deadline during which students can revise and resubmit. Students are less likely to cheat when they can see paths for recovery, and they appreciate it when teachers show enough care to give them room to breathe.

Rationalization, the second condition, is the story students tell themselves to justify cheating. If students perceive assignments as busywork—disconnected from their lives, their goals or any skill that matters—they will be more likely to offload the work to AI. Why not cheat when the task feels like just another pointless hoop?

To prevent that perception, faculty need to answer three questions up front: What skill is this building? How does it apply beyond this course? Why does it matter?

Opportunity, the third condition, can be reduced by shifting pedagogy from product to process. If a course grade hinges on a few heavily weighted papers, the easy button of AI pulses like an open invitation. But when the messy cognitive work of learning is highlighted and valued—through annotated outlines, revision histories, conferences, drafts and written reflections on changes—opportunity becomes diffuse, authorship becomes layered and students become more invested in their own work.

This shift toward process also alleviates pressure on students who are already struggling: those with ADHD who grapple with time management, those on the spectrum who may need more structure than open-ended prompts provide and multilingual students who need iterative feedback rather than a single high-stakes evaluation. Scaffolding a major assignment into a proposal, draft and revision—where each component counts for less—distributes pressure across the semester, leaves room for feedback and cultivates pride in authorship.

This is also where AI can be enlisted as an ally. A course-specific chatbot, introduced early and trained on course materials, can provide 24-7 support. Students can ask questions without fear of judgment, clarify directions, better understand the material and even receive initial low-level feedback. Encouraging students to use AI in this way reinforces the course’s implicit message: Support is available, shortcuts are not necessary and the point is your development.

None of this eliminates cheating. Some students will cheat no matter what you do. But fortifying a course to catch cheaters might catch a few, while rebuilding that same course to support students will reach the majority: those who want to learn with integrity and purpose.

Academic integrity will not be secured by better detectors, but by better design. We need to replace interrogation tools with support conditions that encourage honest effort. A small extension gave my student the breathing room to complete work that felt worth doing. That combination is rarer than it should be, and more powerful than any detector.