The biggest mistake when studying with ChatGPT: is the answer the AI gave you really your knowledge?
In short
- An AI explaining something well does not mean you understood it.
- In one study, students who practiced with a general chatbot scored higher while using it, but lower on a test without AI.
- Get the explanation from AI, then have it write questions, answer them yourself, and retry without AI a few days later.
When a concept stumps you, you ask ChatGPT. A few seconds later, a neatly organized explanation appears. It unpacks difficult jargon and adds a fitting example for the hard parts. You read it and nod.
"Ah, now I get it."
Material that used to take several readings of a textbook or a search for a good lecture now seems to click after a short chat. But there is a problem that is easy to miss. The fact that AI can explain a concept perfectly does not mean that you, having read the explanation, understand it perfectly.
Recent studies suggest that producing better work with AI and building the ability to solve problems without AI may be two different things.
1. The illusion created by an easy explanation
Cognitive psychology has a concept called processing fluency. Processing fluency is how easily and quickly you can process a piece of information. Take two explanations of the same idea: one full of dense jargon and one using familiar words and concrete examples. The second feels much easier to understand. The catch is that we use this ease of processing as a clue to how well we understand something.
A feeling that it "reads easily" can turn into a judgment that you know it well. ChatGPT is a tool that makes this situation easy to create. It organizes complex material step by step, offers easy analogies, and restructures its explanation to fit your question. That is of course a real strength. When a hard concept is explained simply, it reduces the mental load of learning and gives you a foothold for new material.
But here you need to separate three things.
Suppose ChatGPT explains opportunity cost in economics. While you read, the concept may feel simple. But the next day, in a new decision where you have to calculate or explain opportunity cost yourself, it is a different story. What you need then is not the ability to recognize sentences the AI wrote, but the ability to use the concept from your own memory and understanding.
So however good the AI's explanation is, you should not conclude that learning has happened just because you read it.
2. Using AI and getting worse grades?
In June 2025, an interesting study on AI and learning was published in the Proceedings of the National Academy of Sciences (PNAS). Researchers tested how generative AI affects math learning with about 1,000 high school students. Students solved problems under three conditions.
First group: Used regular study materials with no AI help.
Second group: Could use an AI similar to a standard ChatGPT.
Third group: Used an AI tutor that gave hints and made students work through the problem rather than giving the answer right away.
When AI was available, students' problem-solving performance rose substantially. The group using the standard AI scored about 48% higher on practice problems than the control group, and the group using the AI tutor designed to support learning scored about 127% higher. From this alone, you might conclude that AI is highly effective for learning.
But the researchers then had the students take an exam without AI. The results changed.
The exam scores of the group that used the standard AI were about 17% lower than those of the group that did not use AI. In the group that used the AI tutor designed to support learning, most of this negative effect was reduced. The important point of this study is not the simple conclusion that AI is bad for learning.
It is that the ability to solve problems while using AI and the ability to solve them without AI can be different. When the researchers analyzed how students used AI, those with the standard AI tended to ask for the answer directly and use it. Students with the learning-focused AI tended to use it to get hints or to try the solution themselves.
Of course, this was a study in a specific school's math learning environment, so it cannot be applied equally to every subject and age. Still, the study leaves an important question.
Failing to tell these two apart is the mistake to guard against when learning with AI.
3. What happens when you hand the thinking to AI
This can also be explained with the concept of cognitive offloading. Cognitive offloading means handing some of the work of remembering or thinking to an external tool to reduce your own mental load. We use cognitive offloading in daily life too. We save a phone number in contacts instead of memorizing it, use a calculator for complex arithmetic, and put important dates in a calendar. None of this is inherently bad.
Used well, external tools cut unnecessary load and let you focus on more important problems. But studying is a little different. If the purpose of learning is itself to build a particular way of thinking or a body of knowledge, handing the whole process to an external tool does not necessarily help.
Say you are studying math problems. If you ask ChatGPT for the solution the moment you read the problem, you get the answer and working in seconds. That cuts the time it takes to resolve the problem considerably.
But the learner may not get enough practice choosing the formula, planning a strategy, or spotting their own mistakes.
Choosing the formula, planning, and finding mistakes are done by the AI.Little thinking practiced by you
You choose, plan, and find your own errors.You go through the thinking
Here the AI stops being a tool that helps learning and becomes a tool that does the mental work the learner should be doing. What matters is not putting in as much mental effort as possible, but experiencing the thinking process that the learning requires. Leaving a hard new topic for you to figure out entirely alone can also be inefficient. There, the explanations and examples an AI provides help.
Conversely, when you are practicing something you already partly understand, producing answers yourself and correcting errors is what matters. In the end, the AI's role should change with the learner's current level.
4. So should you not study with AI?
Not at all. There is also research showing the benefits of learning with AI. In 2025, Scientific Reports published a randomized controlled trial with university students comparing an AI tutor with in-class active learning. In that study, students who used an AI tutor designed around sound teaching principles learned more in less time than those in the comparison class. Student engagement and motivation were also positive.
Does that contradict the study above? Not necessarily. The two studies differed in how AI was used, the learning environment, and the comparison condition. A systematically designed AI tutor that supports learning and a general chatbot that lets users freely request answers can play different roles in the learning process.
AI is neither a tool that weakens learning just by being used, nor one that improves it just by being used. A systematic review published in 2026 on generative AI and university students' higher-order thinking reached a similar conclusion. Synthesizing 89 studies published between 2024 and 2026, the researchers found that the effect of generative AI depends on how it is used and how it is designed educationally.
The key is whether AI is used as a tool that extends the learner's thinking or as one that replaces it. So the important question is not "Can I use ChatGPT?" It should be "How should I use ChatGPT so that I actually learn better?"
5. Three common mistakes when studying with AI
Mistake 1. Checking the answer before understanding
When you hit a hard problem, everyone wants to check the answer quickly. But if it is a problem the learner can solve, it is better to try it yourself before looking at the answer. Checking the answer is not the problem in itself.
The problem is that you lose the chance to attempt the solution and check your understanding. Try changing it like this.
Asking this way lets you use the AI as a tutor that supports learning, not an answer generator.
Mistake 2. Finishing studying after reading the AI's explanation
Just because ChatGPT explained a hard concept simply does not mean learning is complete. Even if it feels clear while you read, you may not be able to recall it later. What you need here is retrieval practice. Retrieval practice means recalling what you learned directly from memory instead of rereading it.
For example, if you learned a concept with AI, pause the conversation and answer these questions.
- What is the core of the concept you just learned?
- Why does this principle hold?
- Can you come up with a concrete example yourself?
- How does it differ from similar concepts?
Questions like these show whether you merely followed the explanation or can actually remember and reconstruct the content.
Mistake 3. Assuming the AI's answer is always accurate
Generative AI is excellent at producing sentences that fit the context naturally. But the fact that an answer is logically smooth and written with confidence does not guarantee that it is accurate. AI can invent facts that do not exist, oversimplify complex concepts inaccurately, or present a wrong solution convincingly. This is commonly called hallucination.
Learners meeting new material in particular lack background knowledge in the field, so they may find it hard to notice the errors an AI presents. So when using AI, build the habit of checking important concepts, figures and sources against a textbook or other reliable material whenever possible. The ability to critically examine an AI's explanation is as important a learning skill as the ability to understand it.
6. A four-step way to use ChatGPT well
So how do you keep the strengths of AI while raising the real learning effect? The following applies methods from existing educational psychology research, such as retrieval practice, feedback and self-checking, to learning with AI.
STEP 1. Have the AI explain the concept
For a concept you are learning for the first time, it is good to understand it through sufficient explanation and examples. At this stage you can make full use of AI's strength in putting hard material into easy language. Try asking like this.
If the explanation is hard, ask for an easier example or ask specifically about the parts you do not understand. But it matters not to decide that you are done just because you read the explanation.
STEP 2. Ask the AI to write questions
Once you have learned the concept well, switch from the role of receiving explanations to the role of answering.
Asking this way lets you use the AI as a tool for assessing learning. When you answer, try not to look at the original explanation if you can. The purpose of this step is not only to get many answers right but to find out where your memory and understanding stand.
STEP 3. Analyze why you got things wrong
If you got something wrong, do not just check the answer and move on. The same wrong answer can have different causes. You may have misunderstood the concept, failed to remember a needed fact, or understood the concept but struggled to apply it in a new situation.
Here you can ask the AI like this.
This process helps you recognize your weak points and study the missing parts again. Feedback from an AI can also contain errors, so check important points against the original material.
STEP 4. A few days later, try again without AI
To remember what you learned for a long time, you need to practice recalling it after time has passed. This is called spaced practice. Test what you studied today the next day, a few days later, and a week later. You do not need to study everything from scratch at each review. It is better to separate the concepts you remember accurately from the ones that are still hard, and spend more time on the weak parts.
The important point is not to ask the AI for an explanation as soon as you start reviewing. First build your answer from your own memory, then use the AI to check the result. Checking whether you can explain it without AI is an important clue to your actual learning.
7. In the age of AI, the standard for studying has to change too
Until now we have used several measures to judge how well studying went. How long you studied, how many pages you read, how many lectures you took. With AI, new measures have been added. How many questions you asked, how quickly you got the answer, how much content you summarized.
But these numbers do not accurately describe real learning. Say two people studied the same material.
Which one learned more? You cannot tell from the fact that the study time was the same. What matters in the end is what the two people remember, understand and can actually use after studying.
So in the age of AI, pay attention not only to the time spent or the output the AI generated, but to how the learner's own knowledge and ability changed. The ability to judge what you know, what you do not know, and what you need to study again is called metacognition. As AI advances, this ability may matter even more.
AI can answer almost any question, but it does not automatically guarantee how much the person reading the answer actually understood.
8. From today, ask the AI for questions instead of answers
The mistake to guard against most when studying with ChatGPT is not the act of asking the AI questions. It is mistaking the answer the AI produced for your own ability. In an age when AI can explain everything, the process of checking what you yourself can explain matters even more.
Look again at the PDFs, lecture materials and notes you studied today. Then ask yourself.
"Could I explain this without looking at anything?"
"Can I tell the difference between having memorized a concept and having understood it?"
"Where have I not yet properly understood?"
If these questions are hard to answer, it can help to check your own understanding from what you have studied so far before you ask the AI for a new explanation. Recently, there are also study tools that create quizzes and flashcards from the PDFs, notes and web material you studied, and check your understanding through your answers.
What matters is not how many questions you asked the AI. It is what you actually learned from those questions. From studying that gets answers from AI, to studying that makes you able to answer on your own through AI. The small difference that changes the direction of your studying starts here.
Study your own material with obbler
Add a PDF, photos, a link or a video. Chat about it and turn it into notes, flashcards and quizzes.
Download on the App Store