ChatGPT Can Make You Worse at Exams. Here's How to Learn Faster with AI Instead
Published on
In a study published in PNAS in 2025, researchers followed nearly 1,000 high-school math students in Turkey. Each class was assigned one way to practice:
- a plain ChatGPT-style interface
- a GPT Tutor: same model, but nudging students toward the solution without handing it over
- no AI at all: textbook and class notes only
During practice, the AI groups crushed it. Compared to the students without AI:
- +48% for the plain ChatGPT group
- +127% for the GPT Tutor group
Then everyone took a closed-book exam, no AI allowed:
- the plain ChatGPT group scored 17% lower than the students without AI
- the GPT Tutor group scored about the same as the students without AI
The students who leaned on ChatGPT looked like the best learners in the room, right until the help was gone. Their performance went up while their learning went down.
So should you stop studying with AI?
AI can make you learn twice as fast
In 2023, Harvard ran a randomized trial in its largest introductory physics class. Each of the 194 students learned one topic in class, with experienced instructors running active learning sessions, and another topic at home with an AI tutor.
At the end:
- students learned more than twice as much with AI (+1.75 points on the post-test vs +0.75 in class)
- they did it faster: 49 minutes on average vs 60 in class
And the in-class sessions weren’t boring lectures: active learning is already one of the best ways to teach.
Same story in Nigeria. The World Bank ran a 6-week after-school program where secondary students practiced English with GPT-4. The gains were equivalent to 1.5 to 2 years of regular schooling.
So what did Harvard’s tutor do that ChatGPT didn’t? It never gave the answer. It asked questions. When students got stuck, it broke the problem into smaller steps. When they guessed right without understanding, it pushed them to explain why.
The AI wasn’t the variable. How students used it was.
How to learn with AI without falling into the trap
The rule behind all of the studies above: the AI should make you think harder, not less. Here’s how to apply it.
1. Try first, ask second
Struggle with the problem before opening the chat. Even a wrong attempt prepares your brain to remember the right answer. If you ask first, you skip the part where learning happens.
2. Ask for hints, not answers
Tell the AI up front how to behave:
I’m learning [topic]. Don’t give me the solution. Ask me one question at a time to guide me, and only give me a hint when I’m stuck.
That one prompt turns a plain ChatGPT into something much closer to the GPT Tutor.
3. Make it quiz you
Reading an explanation feels like learning. Retrieving it from memory is learning. After each session, ask:
Quiz me on what we just covered. One question at a time, and don’t tell me the answer until I’ve tried.
4. Explain it back
Once you think you understand something, explain it to the AI in your own words and ask it to find the gaps. If you can’t explain it simply, you don’t understand it yet. That’s the Feynman technique, and AI makes a very patient listener.
5. Come back later
One session isn’t enough, no matter how good it was. Quiz yourself again after a day, then a few days, then a week. Each time you remember something just before forgetting it, it sticks longer. That’s spaced repetition.
6. Test yourself without the AI
Don’t wait for your exam to face the same reality as the Turkish students: regularly solve a problem with the chat closed. It’s the only honest measure of what you know.
Conclusion
AI doesn’t make you a better or worse learner by itself. Used as an answer machine, it makes you feel productive while your exam score drops. Used as a tutor that questions you, quizzes you and makes you come back, it can make you learn twice as much in less time.
This is what I built Cleverose for: every course is structured around active recall, the Feynman technique and spaced repetition, so you can just start studying.
Next time you open a chat to study, ask yourself one question: who’s doing the thinking here?