Professor’s Chart: How Common Is AI Cheating In College Exams And What It Means For Parents
Last Saturday, I met an old college friend Tom, who now teaches computer science at our state’s flagship university, at the glazed donut shop halfway between our houses. The espresso machine was broken, so we drank bitter black drip, split a cinnamon cruller that was still warm from the morning batch, and I asked how my oldest son, enrolled in Tom’s intro CS 101 this semester, was keeping up. He sighed, wiped crumbs off his flannel shirt, and spun his beat-up laptop around on the sticky Formica table. He pulled up a color-coded bar chart he’d built after grading last month’s midterm.
The x-axis was percentage AI similarity score, the y-axis was number of students. I leaned in. Only 38% of the 182 students in the lecture hall had responses that scored below 10% similarity to AI-generated content. Almost 25% scored over 70%—meaning nearly a quarter of the class had turned in work that was almost entirely written by a chatbot. I whistled. I’d heard my son mention offhand that half his class used AI on exams, but I’d written it off as kid exaggeration, the same way my friends and I talked about “everyone” cheating on chem quizzes back when I was in college. This chart made it real.
Tom said he didn’t want to believe it either, until a student turned in a midterm response that ended with a random line: “This functionality is not currently available in the OpenAI API, please upgrade your plan to access this feature.” That’s when he ran every single submission through a detection tool, sorted the results, and made the chart. He sent it to the whole department email list not to call out students, but to expose how widespread the issue had become, because most senior professors still insisted AI cheating was a rare problem, limited to a handful of lazy kids.
We sat there for another 20 minutes talking about the kids he’d pulled into his office after the midterm. None of them tried to deny it when he showed them the score. Almost all said the same thing: tuition went up 14% this year, they’re working 20 to 25 hours a week to cover rent and books, they had three midterms and a lab report all due that same week, and cutting corners on this one 100-point exam felt like the only way to not fail the two that mattered for their major. One girl told him she couldn’t afford to retake the class if she failed, so using AI was a less bad option than dropping out.
That conversation stuck with me, because I’d always framed AI cheating as a moral problem—bad kids making bad choices. But after seeing that chart, and talking to Tom, it’s clear it’s just as much a problem of overload. As a parent, I’ve changed a few small things already, and they’ve made a big difference in how my son talks to me about school.
First, I stopped leading with “what grade did you get?” when we talk on the phone. Now I start with “how many hours of work and studying do you have this week?” It sounds small, but it opens the door for him to admit he’s overloaded, instead of hiding it to make me happy. A week after that coffee, I asked that question, and he admitted he’d used AI to outline his last 20-point quiz response, because his gym shift ran two hours late and he couldn’t focus on writing it from scratch. He never would have told me that if I’d started by asking if he was cheating.
Second, I told him if he’s carrying 15 credits and working 15 hours a week and it’s too much, he can drop one class. I grew up with the rule that you never drop a class, you tough it out. But that rule doesn’t make sense now, when the alternative is cheating and getting suspended, or flunking out entirely. He dropped a gen ed geography class two weeks ago that he didn’t need for his major anyway, and his stress level dropped so much he called me to say he actually has time to study for the classes that matter now.
For teachers, Tom’s example is pretty straightforward, too. After he published the chart, he changed his exam structure for the rest of the semester. Now he splits every major test into two parts: one 50-minute in-person, handwritten section with no devices allowed to check basic knowledge, and one take-home project that allows AI, but requires students to turn in a one-paragraph breakdown of how they used it and what they changed to make it their own. He told me the number of high AI similarity scores on the last exam dropped to 8%, and most of those were from students who forgot to add their disclosure paragraph, not people trying to hide it. He didn’t have to fail a dozen kids, and he still gets to see who actually knows the material.
I still don’t know exactly where my son draws the line between using AI as a study tool and using it to cheat on graded work. Tom’s chart is still circulating around his department, and there’s no consensus on what to do next—some professors want harsher penalties, some want to adjust assignments like Tom did, and most are still waiting for someone else to fix it. It’s a messy problem, and there’s no easy fix. But I know that pretending only a handful of bad kids are doing it doesn’t help anyone. The chart exposed how big the problem is, and the next step isn’t just more punishments—it’s asking why so many kids feel like they have to cheat to get by.
Please indicate: Thinking In Educating » Professor’s Chart: How Common Is AI Cheating In College Exams And What It Means For Parents