AI explains it well. Do you actually understand it?

In short

  • In a 2025 PNAS study, students using standard GPT-4 did about 48% better on practice problems but about 17% worse on a test without AI.
  • Clear explanations create an illusion of understanding (processing fluency). Retrieval practice beats rereading: 61% vs 40% recall after a week.
  • Ask AI for questions instead of answers, explain concepts in your own words, and review over time.

Ask ChatGPT about a difficult concept and you get a clear explanation in seconds. It can summarize research papers, break down math problems and walk you through solutions step by step.

But everything makes sense while you are reading it. Then the next day, when you try to explain the same concept on your own, you struggle to remember it. You are spending more time studying, but you are not sure how much you are learning.

Are we really learning, or just getting better at feeling like we have learned?

1. Used the wrong way, AI can hurt your learning

A 2025 study in the Proceedings of the National Academy of Sciences (PNAS) looked at how AI affects math learning among nearly 1,000 high school students. They were split into three groups: standard GPT-4, an AI tutor designed to support learning, and no AI.

Practice vs. real test
Standard GPT-4About 48% better on practice problems, about 17% worse on a test without AI.Answers on demand
AI tutor with hintsGuiding questions instead of answers. No statistically significant decline.Made them think
The tutor gave hints and guiding questions rather than immediate answers.

This does not mean AI is bad for learning. It means how you use it matters as much as whether you use it. A general-purpose AI helps you solve problems faster, but it can also remove the need to think them through yourself. There is a big difference between AI being good at solving problems and AI helping you become good at solving them.

2. Feeling like you understand is not understanding

Cognitive psychology calls it processing fluency: when information feels easy to process, we believe we understand it. AI is incredibly good at making complicated ideas feel simple, with familiar examples, clear language and intuitive analogies. While you read, everything clicks.

But recognizing an explanation as correct is not the same as reconstructing it yourself. One of the best ways to test understanding is retrieval practice: instead of rereading, you actively try to recall from memory. In one comparison of study methods, recall after one week looked like this.

Recall after 1 week
Repeated reading
40%
Retrieval practice
61%
A specific experimental setting, so the exact gap will not apply everywhere. The direction is the point.

How easily you understand something while studying is not a reliable sign of how well you will remember it. What matters is whether you can retrieve it without help.

3. How to study with AI instead

Use AI in ways that force you to think.

Ask for questions, not answers

When a problem is hard, resist asking for the full solution. Try this prompt:

"Don't give me the answer. Ask me one question at a time to help me figure it out myself."

Explain it in your own words

After reading an explanation, close the response and explain the concept yourself. Then ask:

"Here's my understanding of this concept. Can you identify any misconceptions, missing details, or gaps in my reasoning?"

This exposes the difference between what you think you understand and what you actually do.

Review over time

Understanding something today does not guarantee you will remember it next week. Ask AI to generate review questions from what you learned, focusing on concepts you struggled with before. The goal is not to understand once, but to make it stick.

Understanding matters more in the age of AI

AI has made information easier to reach than ever. But the speed of access is not the speed of turning it into knowledge. What matters is not how many explanations you have read, but whether you can explain concepts independently, apply them to unfamiliar problems and remember them over time.

Use AI to think more deeply, not to do the thinking for you.

That is the problem behind obbler, a study app built around genuine understanding rather than passive reading. Upload your own PDFs, web pages and notes, then learn through conversation with an AI tutor that checks how well you understand, finds gaps in your knowledge and helps you work toward mastery.

Check your understanding with your own material →

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