There is a tempting way to learn with AI.
You find something you don’t understand. You ask ChatGPT. It explains it. You read the explanation and think:
Yeah. That makes sense.
Then you close the chat.
Two days later, someone asks you to explain the same idea and suddenly it has vanished.
That is one of the strange problems created by AI.
We now have access to something remarkably close to an endlessly patient private tutor. It can explain the same subject five different ways, generate questions, inspect your work, role-play an examiner and create a study plan in seconds.
But it can also become an extremely efficient machine for helping you avoid learning.
The difference often comes down to one question:
Are you using AI to produce the answer, or are you using AI to make your brain produce the answer?
Those are not the same activity.
Research into AI-supported education is beginning to show exactly why that distinction matters. General-purpose AI can improve performance while students have access to it, yet those gains do not necessarily survive when the assistance disappears. Structured AI tutors that force learners to participate in the reasoning process appear to work considerably better.
So instead of treating AI as an answer machine, treat it as something closer to a learning environment.
Here is how.
1. Start by Finding Out What You Actually Need to Learn
People often begin self-study with something hopelessly broad.
Teach me marketing.
Or:
I want to learn Python.
Or:
Teach me accounting.
The problem isn’t AI. The problem is the size of the request.
Before studying the subject, get AI to map it.
For example:
I want to learn financial modelling well enough to build a three-statement model independently. Break the skill into the major concepts and dependencies. Show me what I need to understand first, what can wait, and what practical abilities would demonstrate that I understand each part.
Now you have something closer to a curriculum.
Do not immediately ask AI to teach every item.
Look at the map.
Remove things you don’t need. Add things you know are important. Ask why one subject comes before another.
You should still own the curriculum.
AI simply makes the first draft easier.
2. Make AI Diagnose You Before It Teaches You
This is one of the easiest improvements you can make.
Instead of:
Explain SQL joins.
Try:
I am trying to learn SQL joins. Before explaining them, ask me five questions to determine what I already understand. Ask one question at a time and adapt the next question based on my answer.
Now something useful happens.
AI stops assuming where you are.
Perhaps you understand tables perfectly but don’t understand primary and foreign keys. Perhaps you already understand joins conceptually and only struggle with syntax.
A good human tutor does this instinctively.
Your AI tutor needs to be told.
3. Stop Asking for the Answer First
Suppose you’re learning statistics and get stuck on a problem.
The fastest prompt is:
Solve this.
It is also frequently the least useful.
Try:
Don’t solve this for me. Ask me what I think the first step should be. If I’m wrong, give me a hint. Only reveal the solution after I have attempted it.
That tiny change turns AI from a solution generator into a tutor.
There is good reason to care about this. Research reviews increasingly distinguish between learning-oriented AI that guides learners and general-purpose AI that simply provides answers. Guided interaction appears much better suited to developing metacognition and self-regulation.
Some struggle is useful.
If AI removes every difficult moment, it may also remove the part where your brain has to reorganize what it knows.
4. Explain the Subject Back to AI
You can reverse the relationship entirely.
Instead of having AI teach you, teach AI.
After studying something, close your notes and write:
I’m going to explain compound interest to you from memory. Do not interrupt me. When I’m finished, identify:
- anything I misunderstood,
- anything important I omitted,
- anything I explained unclearly.
Then explain it.
Badly, if necessary.
This exposes something reading does not.
You discover the difference between:
“I recognise this.”
and:
“I can reconstruct this myself.”
Research on AI-assisted learning increasingly emphasizes retrieval, knowledge reconstruction and error analysis because passive answer-taking can bypass these processes.
You can make the technique harder as you improve.
Ask AI to behave like a skeptical beginner.
Every time I use jargon, ask what it means.
Or:
Challenge any claim where I have skipped a reasoning step.
That can become uncomfortable rather quickly.
Which is probably a good sign.
5. Turn AI Into an Examiner
Reading is comfortable.
Testing yourself isn’t.
So test yourself.
Suppose you have just spent an hour learning basic economics.
Ask:
Test whether I genuinely understand supply and demand. Give me 10 questions ranging from basic recall to practical scenarios. Don’t show the answers. Ask them one at a time. Keep track of the concepts I struggle with.
Afterward:
Based on my answers, what should I revise?
Now your study plan changes according to your weaknesses rather than according to Chapter 1, Chapter 2, Chapter 3.
You can also ask for deliberately tricky questions:
Give me five situations where someone who only memorised the definition would probably answer incorrectly.
Those are often much more revealing.
6. Ask AI to Change the Explanation, Not Just Repeat It
Sometimes you don’t understand something because the explanation is wrong for you.
AI is unusually good at changing representation.
If a technical explanation fails, try:
Explain this visually using a physical analogy.
Then:
Now explain it without the analogy.
Then:
Give me a concrete example.
Then:
Give me a counterexample.
Then:
Explain what beginners usually misunderstand about this.
A concept viewed from several directions tends to become easier to hold.
But there is a catch.
Analogies generated by AI can sometimes sound convincing while being conceptually sloppy. Your own anti-AI-writing research repeatedly identifies plausible-sounding but incoherent metaphors as one of the problems with machine-generated prose.
So use analogies as scaffolding.
Do not mistake them for proof.
7. Give AI Your Actual Learning Material
This may be one of the biggest improvements you can make.
Instead of asking:
Teach me Chapter 4 of corporate finance.
Upload the chapter, notes, slides or transcript you are actually studying.
Then say:
Use this document as the primary source. Teach me the concepts in it. When something isn’t supported by the source, tell me rather than filling the gap from memory.
AI systems can confidently provide incorrect information.
Restricting an AI tutor to curated course material can reduce that problem. A 2025 higher-education study using retrieval-augmented AI tutors found good engagement, while also finding that some answers were still incorrect or wandered beyond the supplied knowledge base.
So even source-grounded AI deserves checking.
It is better.
It isn’t infallible.
8. Use AI to Generate Practice, Not Merely Notes
This is where AI becomes genuinely useful for independent learners.
Imagine you are learning Excel.
Instead of producing another twenty pages of notes, ask:
Create a fictional sales dataset with 100 rows and give me five exercises requiring SUMIFS, XLOOKUP, PivotTables and conditional logic. Don’t give me the solutions yet.
Learning copywriting?
Give me a badly written landing-page headline. I’ll rewrite it. Critique my version but don’t rewrite it for me unless I ask.
Learning guitar theory?
Give me a chord progression and ask me to identify the key and harmonic function.
Learning a language?
Have a conversation with me using vocabulary around A2 level. Correct only errors that interfere with meaning until the conversation ends.
AI can manufacture almost unlimited practice material.
That is far more interesting than manufacturing unlimited summaries.
9. Ask AI to Attack Your Understanding
One of the most useful prompts has nothing to do with getting information.
Try:
What is the weakest part of my understanding?
Or:
What question could you ask that would reveal whether I only memorised this?
Or:
What assumption am I making here?
Your writing reference makes a similar point: asking AI to critique your work can preserve your own thinking better than asking it to rewrite everything for you.
Apply the same principle to learning.
Don’t always ask AI to improve your answer.
Ask it to find the hole in it.
10. Occasionally Remove AI Completely
This may be the most important step.
Once a week, study without it.
Solve a problem.
Write the explanation.
Create the spreadsheet.
Speak the language.
Write the code.
Whatever you’re learning, perform the skill without asking AI what comes next.
The goal is not to prove that you can survive without technology.
The goal is to discover what has actually become yours.
The danger of AI-assisted study appears when temporary performance is confused with durable learning. In the previously mentioned GPT tutoring experiment, students using unrestricted AI performed better during assisted practice but worse after the tool was taken away.
An uncomfortable independent session can therefore tell you more than ten smooth AI-assisted ones.
A Better AI Self-Learning System
You don’t really need dozens of AI learning hacks.
A repeatable loop works better.
Before learning
Ask AI to map the subject and test what you already know.
While learning
Ask for explanations, examples and hints, but attempt the reasoning yourself first.
After learning
Close the explanation and reconstruct the idea from memory.
Practice
Have AI create problems, scenarios and exercises.
Test
Make AI examine you without giving hints.
Diagnose
Ask where your reasoning failed.
Repeat
Study the weak area and test it again later.
And occasionally:
turn AI off.
That last step keeps the relationship clear.
You are learning.
The machine is assisting.
What AI Should Do for You
There are jobs I would happily give an AI tutor.
Generating ten more practice problems because I exhausted the textbook? Fine.
Explaining something from another angle because the first explanation made no sense? Useful.
Quizzing me at midnight when no human tutor is available? Excellent.
Finding holes in my notes? Useful again.
Your source material on human-sounding AI use makes essentially the same case in another context: AI works well for finding omissions, identifying material that can be cut, and assisting with revision, provided the human remains responsible for judging the output.
Learning works the same way.
The interesting question isn’t whether AI belongs in education anymore.
It already does.
The more important question is what part of the job we give it.
Give it the repetitive work. Give it the extra questions. Give it the explanations, simulations, quizzes and criticism.
Keep the remembering.
Keep the struggle.
Keep the judgment.
Keep the part where you have to sit there for a moment and think:
Wait. Do I actually understand this?
That part is still yours.
Frequently Asked Questions
Can ChatGPT really be used as a personal tutor?
Yes, but the quality of the experience depends heavily on how you use it. Asking for direct answers creates a very different learning process from asking for hints, questions, feedback and testing.
What is the best way to use AI for studying?
Use AI around the act of learning rather than in place of it. Let it explain, quiz, generate exercises, challenge your reasoning and diagnose weaknesses. You should still retrieve information from memory and solve problems independently.
Can you learn an entire subject using AI?
AI can help organize and accelerate self-study, but important subjects should still be grounded in reliable books, courses, documentation, research or expert material. Treat AI as the interface through which you work with knowledge, not automatically as the authority behind that knowledge.
How do I prevent myself from becoming dependent on AI?
Regularly work without it. Attempt problems before asking for help, explain concepts from memory and occasionally complete an entire study session without AI assistance.
Can AI create a learning plan for me?
Yes. Give it your goal, current knowledge, available study time and intended outcome. Then review the proposed curriculum yourself rather than accepting it unchanged.
Tools and Resources
You don’t need a complicated collection of apps to begin.
A general-purpose AI assistant such as ChatGPT can handle questioning, explanations, practice generation and feedback. Add reliable source material such as textbooks, official documentation, research papers, course notes or transcripts whenever accuracy matters.
For more serious self-study, the useful combination is surprisingly simple:
good source material + AI tutor + deliberate practice + independent testing.
The AI is the easiest part of that system.
The other three are what make it work.
