Why Graded Tests Should Ban AI: A Teacher’s Perspective
It happened during a unit on the American Revolution. The test was straightforward: five short-answer questions about causes, key figures, and the role of the French alliance. One student, let’s call him Marcus, turned in a response that was almost too good. Every sentence was perfectly structured, with precise dates, even a quote from a letter I hadn’t assigned. I’d taught that unit for ten years, and I knew what a tenth-grader could produce in 45 minutes. This wasn’t it.
I didn’t say anything right away. Instead, I waited until the next morning and asked Marcus to stay after class. “Can you tell me more about the Stamp Act Congress?” I said. He glanced at the ceiling. “It was, like, a meeting of people who were upset about taxes,” he said. I asked about the quote in his answer. He shrugged. “I remembered it.” Then he looked at his shoes. We both knew.
It wasn’t the first time I’d seen that pattern, and it won’t be the last. Marcus is a good kid, not lazy or dishonest by nature. He just figured out that a free AI tool could do the heavy lifting for him, and no one had ever told him why that was a problem. Or rather, no one had told him in a way that made sense.
For a while, I tried to work around it. I changed my test questions to be more personal: “Describe a time you had to persuade someone to see things your way, and connect it to Jefferson’s arguments in the Declaration of Independence.” That worked for about a week. The next essay was polished to a sheen, with no personal details that matched anything the student had actually written before. It was like reading a marketing brochure for a person’s life.
That’s when I started to think seriously about a blanket ban—no AI, not for any small part of a graded test, not even for brainstorming. No exceptions.
I know that sounds extreme. A lot of my colleagues feel differently. They argue that AI is a tool, that students need to learn how to use it ethically, that the real world will be full of AI assistants. I hear that. I really do. But a graded test has a specific purpose. It’s the one place where we can see what a student can do alone, without a search engine, without a parent’s editing, without a chatbot’s confident paraphrase of Wikipedia. If that’s gone, then we’re not measuring learning. We’re measuring access to technology and willingness to use it.
Last semester, I had a student named Elena who struggled with writing. She would spend an hour on a single paragraph, erasing and rewriting until the paper was worn thin. On a midterm, she wrote a full essay in thirty minutes about the causes of World War I. It wasn’t perfect—there were run-ons, a few clunky transitions, a misremembered date. But it was clearly hers. I could hear her voice in it. I could see the effort.
I told her that, actually. “This is really good,” I said. “I can tell you know this stuff.” She looked relieved. “I didn’t use any AI,” she said. “I wanted to see if I could do it by myself.” That’s the moment I want for every student. Not a perfectly polished answer, but the quiet pride of knowing you did something on your own.
A ban on AI during tests doesn’t solve everything, but it creates a space for that moment. It tells students: this part of your education is about your brain, not your device. It’s about retrieving knowledge, organizing your thoughts, and putting them on paper. That’s a skill you’ll use for the rest of your life, whether you’re writing a report, explaining a problem to a coworker, or just trying to argue your way out of a speeding ticket.
Practically, a ban means a few things. For me, it means tests need to be designed so that AI is not even tempting. That might sound old-fashioned, but it works. I’ve gone back to paper tests for most graded work. Not scanned PDFs and online portals—actual paper. Students sit in rows, their phones are in their backpacks, and there’s a clear boundary between a test and a regular class activity. I also make my questions more specific and more tied to the material we covered in class. I avoid generic prompts like “Discuss the causes of the Civil War,” which AI can answer instantly. Instead, I ask things like “Explain how the Fugitive Slave Act of 1850 changed the way Northerners thought about slavery,” or “Use at least three primary sources from our reading to argue whether Lincoln’s suspension of habeas corpus was justified.” These aren’t impossible for AI to answer, but they’re harder to fake. And the act of answering them requires actually having read the material.
For parents, a ban might feel like a relief, but it also requires some adjusting. It means you can’t pretend not to notice when your kid is typing a question into a chat window five minutes before a test. It means you have to set a rule at home that goes beyond “don’t cheat.” It means saying something like, “For any graded assignment, you’re on your own with your own brain. We can talk about what you don’t understand, but I’m not going to help you generate answers.” I’ve seen parents who are exhausted by this. They want to help, and AI feels like a way to level the playing field. But I tell them: your kid doesn’t need a perfect test score. Your kid needs to learn how to think under pressure, how to be wrong, how to sit with a blank page and figure out a way forward. That’s hard. That’s what school is for.
There’s also a deeper reason for the ban that I didn’t fully understand until I had my own daughter. She’s in third grade now, and she’s learning to read. Sometimes she gets stuck on a word, and my wife or I will prompt her: “Sound it out.” It would be faster to just tell her the word. It would be more efficient. But then she wouldn’t learn the decoding strategy. She wouldn’t build the neural pathway. The struggle is the learning.
The same is true for a high schooler writing an essay on a test. The struggle of recalling a fact, organizing an argument, choosing a verb—that struggle is where learning sticks. When AI does that work, it’s like telling your child every word instead of letting them sound it out. They get the answer, but they don’t get the ability.
I don’t think we can ban AI everywhere. Students will use it on homework, on projects, on take-home essays. And honestly, some of that is fine. I want them to learn how to work with a tool that will be in their professional lives. But a graded test is the one time we can see what they know without any scaffolding. If we’re serious about assessment, we have to protect that.
I’ve started telling my students this at the beginning of the year. I don’t lecture them about cheating. I just say: “When you take a test in this class, it’s just you and your brain. That’s a small thing, but it’s important. It’s a chance to find out what you actually know. You might be surprised. You might know more than you think.”
Marcus, the kid from the Revolution unit, took a different path. After our conversation, he didn’t come to class for a couple of days. Then he sent me an email, apologizing, but also asking if he could retake the test. I said yes. He came in after school, sat at a desk in the empty classroom, and wrote for forty minutes. When he finished, his answer was messier, less polished, and full of real understanding. He knew the cause and effect. He could name the players. He didn’t quote a letter, but he explained why the French alliance mattered, and he used his own words.
I haven’t completely solved the AI question. I still wonder about the student who is a slow writer, or the one who freezes under pressure. A ban doesn’t help them automatically. But I’ve learned that if I make the test about the thinking, not the production, they often do better than they expect. And that’s enough for now.
Please indicate: Thinking In Educating » Why Graded Tests Should Ban AI: A Teacher’s Perspective