Why I Tell My Students AI Is Not Allowed on Graded Tests
The question came from the back row on a Tuesday morning. “Can we use AI on the essay test?” Marcus asked. He wasn’t trying to hide anything. His hand was up, and his Chromebook was open to a chatbot that had become as normal to his generation as a calculator. A dozen kids turned to hear my answer.
I’ve taught high school English for eleven years, and I’ve learned not to make quick rules about technology. But this one is clear to me: no AI on graded tests. Not because every AI use is cheating. It isn’t. But because a graded test is one of the few moments when a student shows me, without support, what they can do alone. If AI is allowed, the test stops showing me the student. It shows me a combination of the student and the tool.
The week before that Tuesday, I had used a chatbot with the same class. We were working on thesis statements for an essay about Of Mice and Men. I typed a generic prompt into the AI, and it gave us a list of thesis statements. We read them aloud and talked about why some were too broad, why some were too obvious, and which ones had enough tension to support an argument. It was a productive class period. Students saw possibilities and then went off to strengthen their own ideas. Some told me later they used a chatbot at home to check their own thesis statements before starting their homework. I was fine with that. That’s practice.
The essay test that Tuesday was different. It asked them to read a short passage from a writer we hadn’t talked about in class. No notes, no internet, no discussion. The question was simple: “What does the passage reveal about the narrator, and how does the author show it?” They had forty-five minutes. They could write on paper, ask me to pronounce a word, or raise their hand if they needed a new copy of the passage. That was the whole thing.
Marcus’s question made sense, though. I had just spent a class period encouraging them to use AI. Now I was asking them to put it away. It must have felt contradictory. I tried to explain.
“In class, use every tool you can find,” I said. “On the test, I need to see what you can do without the tool. That’s how I know where to help you.” Marcus nodded, but he had a follow-up. “So if I use AI on the test and it makes my thesis better, the grade doesn’t mean anything?” I told him the grade wasn’t fake. The issue was the information. If I read a thesis that came from AI, I would think he was further along than he actually was. Then I’d move on to the next skill and leave him behind. The test isn’t just a grade. It’s data for me.
He didn’t argue. I’m not sure he agreed. But he put his Chromebook aside and started reading the passage. When I collected the essays later, he paused at the door. “I didn’t use it,” he said. “Yeah, I know,” I said. “That’s the point.”
I didn’t get into the bigger conversation with Marcus that day, but the bigger conversation is what I actually think about. Schools are scrambling to decide where AI belongs. Some teachers allow it. Some try to police it. Some have given up. I’ve landed on a middle ground that makes sense to me: AI gets to be a practice partner, not a test-taker. It can brainstorm, coach, explain, argue, and suggest. It cannot take the final exam.
There are practical reasons for this. A test is the most honest signal a classroom has. Homework can be collaborative, and honest students will tell you they sometimes get help. But a timed, in-class test happens in a controlled moment. It tells me whether the student can hold an argument in their head and put it on paper. If I take that away, I’m teaching blind.
I also think tests need to be designed with this rule in mind. I don’t use AI-detection software. I’ve seen too many false positives, and I don’t want to accuse a student who didn’t do anything wrong. Instead, I try to make the test itself something a chatbot would be useless for. The passage is new. The question asks for a response to specific words on the page. Any student can argue anything, as long as they can point to the text. A chatbot might give a polished essay, but it can’t read the student’s mind. The only way to get the student’s mind is to ask the student directly.
I’m also honest with parents. At back-to-school night, I say it plainly: “AI is allowed on homework in my class, but not on graded tests. If you see your child using AI on a test question, that’s not laziness. It’s usually a sign they don’t trust their own thinking or don’t feel ready. Please tell me. That’s helpful information.” A few parents have thanked me for saying it that way. Others still worry their kids will fall behind. I understand. The technology is powerful. But the students who have practiced writing without it are the ones who will be able to use it well later.
Last spring, a student named Maya showed me why this rule matters more than I knew. She was a junior, quiet, always completed her work. She wrote an essay for a final exam that was, to be blunt, too smooth. Every sentence sounded like it had been through a grammar checker, and the vocabulary was not hers. In class, Maya wrote short, direct sentences. The essay said “in the aggregate.” I had never heard her say “in the aggregate” once.
I didn’t call it cheating. I asked her to come in at lunch. She answered before I could finish the question. “I ran it through AI to make it sound better,” she said. “I didn’t want you to think it was bad.” She wasn’t trying to deceive me. She was trying to make her own words seem good enough. We sat down and she pulled up a paragraph from the essay. I asked her to write the same idea in her own voice, using words she would actually use. She did. It was shorter, a little rough, and completely hers. It was better.
That moment made me see the rule as more than a policy. Tests can be imperfect and stressful. But they are also one of the few times we tell students, “I want to hear what you have to say.” AI is a fantastic assistant, but it has a habit of turning their voices into someone else’s. A test that requires them to answer alone is a way of saying their voice, as it is, counts.
I don’t pretend that settles everything. Graded tests aren’t perfect measures of learning. Some students freeze even when they know the material. And the line I draw—practice with AI, tests without it—will probably shift again as the tools improve. Maybe one day there will be a smarter way to assess what kids know while letting a computer help them along. I’m open to that.
For now, though, I keep coming back to Marcus at the door. He said, “I didn’t use it.” He said it like a confession, but it wasn’t. It was a quiet statement of what he was capable of. We need more of those statements. And a graded test, for all its flaws, is still a place where they can be made.
Please indicate: Thinking In Educating » Why I Tell My Students AI Is Not Allowed on Graded Tests