Why Students Using AI on Graded Tests Is a Problem We Can’t Ignore
The essay was about the symbolism in The Great Gatsby, and it was good. Almost too good.
I’d assigned the essay to my junior English class on a Tuesday. By Thursday, one of my quietest students, Marcus, had turned in a paper that read like something from a literary journal. The sentences were polished. The vocabulary was precise. The analysis was more sophisticated than anything I’d seen from him in four months of class.
So I did what I’ve started doing this year. I pulled Marcus aside after class and asked him to sit down with me and tell me, in his own words, what he thought the green light meant.
He couldn’t.
He stammered for a moment, then said, “I just thought it represented, like, his goals and stuff.”
That was it. Four months of essays, and this was the student I had been grading. Or rather, it wasn’t.
This is the moment that changed how I think about artificial intelligence in my classroom. Not in a theoretical, policy-debate way, but in a very practical “I can’t tell what my students actually know anymore” way. And it’s why I’ve become firmly convinced that AI should be banned on all graded tests and assessments.
Let me be clear about what I mean. I’m not talking about using AI to help a student brainstorm ideas or understand a confusing concept. I’m talking about the moments in a student’s education that are meant to measure what they can do alone. The graded essay. The unit test. The final exam. These assessments exist for one reason: to show us what a student has actually learned.
When AI writes the answer, that purpose disappears.
I’ve talked to other teachers in my building, and almost all of them have a Marcus story. The social studies teacher who had a student turn in a perfectly formatted DBQ essay but couldn’t name one cause of World War I in class the next day. The science teacher who received a lab report with statistical analysis that her student couldn’t explain. The math teacher who noticed that several homework assignments had the same unusual wording patterns.
We’re not Luddites. Most of us use technology constantly. We have students type essays on laptops, we use online platforms to administer quizzes, we encourage students to do research online. But there’s a difference between using tools to support learning and using tools to replace thinking. Graded tests are where we finally see what’s in a student’s head. If the AI does the work, that window slams shut.
When I first noticed this happening in my classroom, I wasn’t sure how to handle it. I tried a few things. I added honor code statements at the top of each assignment. I lectured about academic integrity. I even tried to explain why this mattered for their learning.
What I learned is that the problem isn’t laziness. Most students aren’t trying to cheat because they want to get away with something. They’re doing it because they’re overwhelmed, or they don’t know how to start, or they’ve convinced themselves that using AI is just “the way things are done now.”
One of my students told me, completely sincerely, “It’s like using a calculator for math. Why should I write the essay when I can generate one and then edit it?”
I understood her point. But a calculator doesn’t write the math problem for you. It doesn’t tell you which equations to use or help you understand why the answer is what it is. A calculator is a tool. AI, when used on a graded assessment, is a substitute.
The difference matters.
I’ve started requiring all in-class essays for major assessments. Students write on paper, in the classroom, with no devices. The first time I did this, I had students visibly anxious. One of them raised her hand and asked, “What if I don’t know what to write?”
“That’s the point,” I said. “I need to see that so I can help you.”
It was uncomfortable. It was slow. Some essays were rough and messy and full of half-formed ideas. They were real. I finally know what my students can actually do.
I also started having brief one-on-one conferences after major graded assignments. Not long ones — five minutes during class while other students read silently. I ask each student to walk me through one of their answers. Tell me what they were thinking, why they made the choices they made. It doesn’t work perfectly. There’s always a handful of students who still slip through. But the conversation changes things.
I’m not going to pretend this is easy. There are real challenges to banning AI on graded tests. Students with anxiety struggle with timed in-class writing and perform differently than they do at home. Students with learning differences sometimes rely on assistive technology that can look like AI but isn’t. There are kids who genuinely benefit from using grammar tools to correct spelling and syntax issues that would otherwise obscure their ideas.
These are problems we need to solve. But throwing out all assessments because they’re complicated isn’t the answer. The answer is figuring out what we’re actually trying to measure and then creating conditions where that measurement is real.
For parents who are reading this, I want you to know something. When your child’s teacher says an assignment must be done without AI, it’s not because we don’t trust your kid. It’s because we want to see your kid. We want to see their actual voice on the page, their actual thought process in the margin. We want to see the stumbles and the breakthroughs.
There’s an argument I keep hearing that goes something like “AI is the future, so students should learn to use it.” I agree that students should learn to use AI. They should learn how to ask it good questions and how to evaluate its answers. That can happen in classroom projects, in group work, in research assignments where the process matters more than the final product.
But a graded test is not the place for that. A graded test is where we ask a student to show us what they know, by themselves, without supports. The day we make graded tests open to AI is the day we stop assessing students and start assessing the technology.
I don’t have a perfect system yet. I still catch essays that don’t sound like the student who wrote them. I still have moments where I suspect an assignment was heavily AI-assisted and I can’t prove it. I still second-guess myself and wonder if I’m being too strict or not strict enough.
What I know is that when I sat down with Marcus and asked him about the green light, I saw a kid who didn’t remember what the green light was. He knew it was in the book, but he didn’t know why. In that moment, I wasn’t grading his essay. I was seeing where he actually was as a learner. That’s the only useful information a teacher can get. Without it, I’m just guessing.
So the ban on AI for graded tests stays. Not because I hate technology, and not because I want to make life harder for students. But because I need to know what’s real. My students deserve that too.
Please indicate: Thinking In Educating » Why Students Using AI on Graded Tests Is a Problem We Can’t Ignore