How to Ask an AI Tutor Better Questions (Real Examples)

Quick answer: To ask an AI tutor better questions, put four things in every message: your exam board and level, the exact problem copied accurately, the specific step you are stuck on, and what you already tried. Then ask for the method rather than the answer, and end the session by asking it to quiz you.

I used an AI tutor badly for about three months. I would paste a whole past paper question, get back a tidy wall of text that solved it perfectly, feel like I understood, then score four out of ten on the same topic a fortnight later. The tool was not the problem. I was asking it to finish my homework instead of asking it to teach me. Once I changed how I typed the question, the same chat box became the most useful thing on my laptop. Here is exactly what I changed, using real examples from my Cambridge IGCSE maths and my A-Level history essays.

The four parts of a question that actually gets a good answer

A strong question to an AI tutor contains level, problem, sticking point, and attempt. Miss any one of them and you get a generic textbook reply.

1. Level. "Cambridge IGCSE Extended" and "CBSE Class 10" produce different explanations of the same topic, because the syllabuses cut off in different places. Naming your board stops the AI wandering into content you will never be examined on. If you finish onboarding, this gets attached to every reply automatically.
2. The exact problem. Type the numbers correctly. I once wrote 0.05 instead of 0.5 and spent twenty minutes arguing with an explanation that was perfectly correct for the question I had actually asked.
3. Your sticking point. "I don't understand" is useless. "I don't understand why the sign flips on the third line" is a question a tutor can answer in two sentences.
4. What you already tried. This is the part almost nobody includes, and it is the one that changes the reply most. If you say what you tried, the AI corrects your reasoning instead of replacing it.

Turning a weak question into a strong one, with real numbers

Here is the same maths question asked two ways, and what came back.

Weak version: "solve 3x^2 - 5x - 2 = 0"

That gets you x = 2 and x = -1/3. Correct, useless, forgotten by dinner.

Strong version: "Cambridge IGCSE Extended maths. Solve 3x^2 - 5x - 2 = 0 by factorising, not the quadratic formula. I wrote (3x + ?)(x + ?) but I cannot find the two numbers. Show me how to find them and name the technique. Do not give me the roots yet."

That reply walked me through splitting the middle term:

1. Multiply the first and last coefficients: 3 times -2 = -6.
2. Find two numbers that multiply to -6 and add to -5. Those are -6 and +1.
3. Rewrite the middle term using them: 3x^2 - 6x + x - 2.
4. Group in pairs: 3x(x - 2) + 1(x - 2).
5. Both brackets match, so factor out: (3x + 1)(x - 2).
6. Roots: x = -1/3 and x = 2.

The roots were the least valuable line. Step 1 was the valuable one, because "multiply the outer coefficients, then split the middle term" works on every non-monic quadratic I will ever meet. That is the difference between an answer and a method. If you want the same treatment on your own working, the math solver shows the steps rather than just the result.

Ask for the method first, then for the mark scheme wording

Split every question into two messages. The first gets you understanding, the second gets you marks, and they are not the same thing.

Message one: "Explain the method and name the technique."

Message two: "Now write that as a full-mark answer to a 4-mark question, and tell me where each mark is awarded."

The second prompt is the one that moved my history grade. For a 6-mark explain question I now ask what the examiner is actually counting: usually a mark for the point itself, marks for specific supporting detail such as a date, a name or a figure, and a mark for linking back to the wording of the question. Once I could see the marks laid out separately, I stopped writing beautiful paragraphs that scored three out of six. If you are working on essays, the essay coach does this against your own draft rather than an invented one.

Five follow-up lines that rescue a reply instead of abandoning it

Most students read a disappointing reply and give up. The fix is usually one short sentence.

• "Too long. Same thing in five steps."
• "I don't get the part where you cancelled the 2. Explain only that line."
• "Give me an easier version of this question first, then this one."
• "What is the most common mistake students make at this step?"
• "Where would I lose marks if I wrote it exactly the way you did?"

That last one is the sharpest question in this whole article. It forces the reply to shift from explaining to examining, and it surfaces what quietly costs marks: missing units, unrounded answers, conclusions that never answer the question. You can fire these off in chat as quick follow-ups without re-explaining your situation each time.

What I got wrong for an entire term

Four honest mistakes.

I pasted whole past papers. Six questions at once produces six shallow answers. One question at a time produces one deep answer, and depth is what transfers.

I accepted the first reply. If the explanation did not click, I assumed I was slow. I was not slow. The explanation was pitched wrong, and one line of feedback would have fixed it.

I never asked why twice. Asking "why does that work?" about a step you have already accepted is where the real learning hides.

I copied phrasing into essays. It read fluently, it did not sound like me, and in a silent exam hall I could not reproduce a single sentence of it. Now I read the explanation, close the tab, and write the paragraph from memory. If I cannot, I never understood it.

Finish every session by making it test you

The highest-value message in any AI tutor conversation is the last one. Type: "Ask me five questions on what we just covered, one at a time. Do not tell me the answer until I have replied."

The "one at a time" part is not optional. If you request all five together, your eyes read the later questions while you answer the first, and you end up with recognition instead of recall. Answering blind, one by one, is proper retrieval practice, and five minutes of it beats the fifteen minutes you just spent reading the explanation. When you want the same thing across a whole topic rather than one conversation, the quiz tool builds it from your notes.

Test yourself

1. Name the four parts a good question to an AI tutor should contain.
2. You are given 2x^2 + 7x + 3 = 0. Which two numbers do you look for when splitting the middle term, and why those two?
3. Which single follow-up question turns an explanation into exam-technique feedback?

FAQ

Should I type the question out or paste a screenshot?

Type it, or paste the text. Typing takes fifteen seconds and forces you to read the question properly, which is where half of my careless errors used to come from. It also removes the risk of a misread symbol changing the whole problem.

What if the AI tutor gives me a wrong answer?

It happens, especially on long multi-step arithmetic. Two defences: ask it to check the result by substituting back into the original equation, and compare anything important against your textbook or a mark scheme. Treat replies as a confident classmate rather than an oracle and you will catch the errors yourself.

Is asking for the method still cheating?

No. Asking for the method, working the problem yourself, then being tested on it is ordinary tutoring. Copying a generated answer into work that will be assessed is a different thing entirely. Every prompt in this article is built so the thinking stays with you.

How long should one session be?

Twenty to thirty minutes on a single topic, ending with the quiz. Beyond that you are reading rather than learning, and reading feels productive while doing very little.

In short

The gap between a useless AI reply and a genuinely good one is about fifteen extra words in your question: your level, your sticking point, and what you already tried. Ask for the method before the answer, ask where the marks are, use short follow-ups instead of giving up, and always finish by being quizzed. Same tool, completely different result.