Turn AI into your personal tutor: 5 practical ways to learn

AI explains a difficult topic, and everything makes sense as you read. Then you try a similar problem on your own and realise you do not know where to start.

That is why it helps to include practice without AI. Can you explain the topic in your own words or solve a new problem without looking at the answer?

You can ask AI to set practice questions, offer hints and give feedback on your answers. Try answering for yourself before asking AI for a complete solution.

This guide gives you five practical tips and prompts to use in your own study. It also examines three studies of AI and learning, including what their findings can and cannot tell us.

Decide what you need: a finished task or a new skill

If you need to send a well-written email, an AI draft may be exactly the help you need. If you want to improve your writing, you also need practice and feedback. Your goal determines what to ask AI to do.

Worked examples can help too. Study the steps, put the solution aside and try a similar problem yourself.

In my own study, I have used questions that require answers from memory. In 2018, I needed to learn the answers to the US citizenship test. Cramming did not appeal, so I built a practice app: it read the questions aloud, and I had to recall the answers without multiple-choice options. An engineer solves a study problem by writing a program.

Now I can ask AI to quiz me on a topic and give feedback on my answers. I no longer need to build an app for that.

5 ways to use AI as a study partner

These tips are practical suggestions informed by the research below. They have not been tested as a set, and a general-purpose chatbot is not equivalent to a tutor built for a study.

1. Try the task yourself, then ask for feedback

Start the task on your own. Notice how far you get and where you need help. Show AI your work even if it is unfinished: a few sentences, the start of a calculation or a code snippet is enough.

Here is the task and my attempt: [task and attempt]. Tell me what I have done correctly. Then choose the most important point to correct and give me a hint. Let me try again before showing me the solution.

If you cannot get started, ask for a worked example first. Then try a similar task yourself.

2. Ask AI to quiz you

Answering a question requires you to put the explanation into your own words. Ask AI to wait for your answer before giving feedback or moving to the next question.

I am studying [topic], and I want to be able to [goal]. Ask me one question at a time, starting with the basics. Wait for my answer, give me feedback and adapt the next question to my understanding. Do not reveal the correct answer before I have tried.

3. Explain the topic in your own words

Put the explanation you have read aside and write your own. Ask AI to check for errors and missing information.

Here is my explanation of [topic]: [explanation]. Check for factual errors or anything important I have left out. Tell me what I should check and why. Do not rewrite the explanation for me.

Check the feedback too. AI can be wrong while sounding confident (see hallucination). If you do not yet know enough about the topic to judge an answer, verify important points in a textbook or an authoritative source.

4. Ask for one hint at a time

If you do not know the next step, ask for a hint about that step. Try to continue before requesting more help.

I am stuck here: [point of difficulty and what I have tried]. Give me the smallest hint that will help me continue. Do not show me the full solution. I will ask for another hint if I need one.

5. Ask for a different explanation

If an explanation is unclear, identify the part you do not understand. Ask for an example related to your work or another familiar situation.

I do not understand [specific point] yet. Explain it using a concrete example related to [my work or a familiar subject]. Then give me a short new task to check whether I understand.

Example: a company makes a profit, but its cash falls

Suppose you are learning to read financial statements. A company has made a profit, but it has less cash than before. How is that possible?

This fictional conversation illustrates how to practise with hints.

You: Profit is the money the company has left over. If it has less cash, the profit cannot be real.

AI: Profit and cash measure different things. Consider the purchase of a machine: when does the company pay for it, and when is the purchase recognised as an expense?

You: The company pays for the machine upfront, but recognises its cost as depreciation over several years. It can also make a sale on credit and recognise a profit before the customer pays.

AI: Correct. Try a new problem: a company sells and delivers goods for €100,000. The customer pays 60 days later. The goods cost €70,000, which the company pays in the month of the sale. Assume there is no other revenue, expense or cash flow, and ignore taxes. How much profit does the sale generate, and how does cash change during that month?

Answer the last question before continuing the conversation with AI. This tests whether you can distinguish profit from cash flow in a new problem.

What does research tell us about learning with AI?

Studies have found both improved learning outcomes with AI and worse performance on a subsequent test without it. These studies used different tools and exercises, and measured learning in different ways.

Harvard: test scores improved more with an AI tutor

On an introductory physics course at Harvard, 194 students studied two topics: one at home with an AI tutor and the other in a classroom using active learning methods. The order was randomised, so each student served as their own comparison.

Students took tests before and immediately after studying. The median increase in test scores was more than twice as large with the AI tutor as with classroom instruction. Median study time with the tutor was 49 minutes, compared with an hour allocated to the classroom lesson.

The tutor guided students through tasks step by step and had been given the correct answers in advance. The trial covered two topics on one course. Because learning was measured immediately after the lesson, the results do not establish long-term retention.

Mathematics: users of a general-purpose assistant scored lower without AI

Nearly a thousand high-school students in Turkey practised mathematics over four 90-minute sessions. One group used a general-purpose ChatGPT-style assistant, another used a tutor designed with teachers' input, and a third practised without AI.

During practice, scores were 48% higher with the general-purpose assistant and 127% higher with the tutor than in the control group. On a test taken without AI, the general-purpose assistant group scored 17% lower than the control group. The tutor group performed about as well as the control group.

The researchers interpreted this as evidence that some students relied on the general-purpose assistant without learning to solve problems independently. Students also overestimated how much they had learned.

Programming: AI users scored lower on a skills test

In Anthropic's study, 52 mostly early-career programmers learned to use an unfamiliar programming library. Those who used AI averaged 50% on the skills test, compared with 67% among those who worked without it.

When researchers examined how participants used AI, the lowest scores came from those who let it write the code, and the highest from those who asked for explanations or asked about concepts. These groups contained only two to seven people each. This finding does not establish that the way they used AI caused the difference.

What can we conclude?

The results cannot be compared directly: the subjects, tools and study designs differed. They also do not establish that these five prompts will make a general-purpose chatbot as effective as the tutors used in the studies.

My practical takeaway is to avoid judging learning solely by whether you can complete a task with AI. Try a new task without help as well. If you cannot do it, return to the step where you still need guidance.

Try it today: start one task without AI

Choose a practice task on a topic you are studying. Spend five minutes trying it without help. Then show AI your answer and ask for feedback on one point that needs correcting. Read the feedback, put it aside and make the correction yourself.

Finally, try a new task without help. You can also return to the topic a week later to see what you remember. That later check is my suggestion, not a method tested in these studies.

Start with one task: try it yourself, ask for feedback and try again.

Disclosure and sources

The research section draws on three published studies. The practical tips are my suggestions based on their findings, and my own experience is an individual observation. One study was conducted by the AI company Anthropic, which readers should consider when assessing its findings. I have no financial interest in any of the companies mentioned here.