AI can answer an A-level maths question in seconds. That is useful, but it is not the same as learning how to answer it yourself.
The danger is obvious. You paste in a question, ChatGPT produces six neat lines of algebra, and each line appears to make sense. You close the conversation feeling as though you understand it. Then you meet a slightly different question without the explanation beside you and cannot begin.
I know that feeling from worked solutions, textbooks and tutoring. An explanation can look completely reasonable while it is in front of you. The test is what remains when it is taken away.
Used badly, AI removes the exact thinking that A-level maths requires. Used well, it can behave more like a patient tutor: asking what you notice, finding the first gap in your understanding and giving you just enough help to take the next step.
The difference is not really the AI. It is what you ask it to do.
Ask for teaching, not an answer
If your prompt is “solve this”, the most likely response is a solution. If your prompt is “teach me how to think through this without solving it for me”, you have changed the job.
Start a conversation with clear rules:
I am studying A-level maths. Act as a Socratic tutor. Do not solve the question for me or reveal the next step immediately. Ask me one short question at a time. Use my answers to find the first thing I do not understand. If I am stuck, give the smallest useful hint, then ask me to continue. Make me explain why each step is valid. Only show a complete solution if I explicitly ask for one at the end.
Then give it the topic, your exam board if it matters, and the exact question. If you have already attempted it, include your working. Your attempt is much more useful than saying “I don’t get it” because it shows where your reasoning changed direction.
For example:
I am learning differentiation and I am stuck on this question: [paste the question]. Here is my attempt: [paste your working]. Do not restart from the beginning. First identify the earliest line where my reasoning becomes uncertain or incorrect, but ask me a question that helps me find it myself.
That last sentence matters. You do not need another polished route through work you already understand. You need help at the first broken decision.
What the Socratic method actually means
The Socratic method teaches through questions. Instead of telling you a fact or method, the tutor asks questions that expose what you believe, test whether it is consistent and lead you towards a stronger conclusion.
In maths, that might sound like:
- What is the question asking you to find?
- Which information have you not used yet?
- What does the gradient represent here?
- Why are you allowed to divide by that expression?
- Does your answer satisfy the original equation?
- What would change if this value were negative?
These are not decorative questions placed around an explanation. Answering them is the work.
Suppose you are solving (x^2 - 5x + 6 = 0). An answer machine factorises it and reports (x=2) or (x=3). A Socratic tutor might begin by asking what two numbers must add to and multiply to. If you know that, it asks you to use them. If you do not, it steps back and checks whether you understand how the middle term appears when two brackets are expanded.
The conversation moves backwards only as far as the missing idea, then lets you move forwards again. That is what makes it teaching rather than answer delivery.
Make ChatGPT ask one question at a time
AI has a habit of being too helpful. Ask it for guidance and it may give you three hints, a mini-lesson and most of the solution in one response. By the time you reach the question at the bottom, there is very little left to think about.
Tell it to ask one short question and wait. This slows the exchange down and makes your answer determine what happens next.
A useful conversation might look like this:
You: I need to find the stationary points of (y=x^3-3x^2-9x+5). Ask one question at a time and do not solve it for me.
ChatGPT: What is true about the gradient at a stationary point?
You: It is zero.
ChatGPT: How can you write the gradient of this curve as an expression in (x)?
You: Differentiate it: (3x^2-6x-9).
ChatGPT: Good. What equation must you now solve?
Each answer is small, but each one belongs to you. If you say that the gradient is the value of (y), the tutor now knows the problem is not factorising the quadratic. The missing idea comes earlier: what a stationary point means.
This is also why you should answer in full sentences occasionally. “Differentiate” may be enough to continue, but “I am differentiating because a stationary point has gradient zero” proves that you understand the connection between the method and the question.
Ask it to diagnose your understanding
“I’m stuck” can mean several different things:
- you do not understand the concept;
- you know the concept but cannot recognise when to use it;
- you chose the correct method but your algebra failed;
- you can complete the process but cannot explain why it works;
- you misread what the question asked for.
Those problems should not receive the same lesson.
Ask ChatGPT to diagnose before it explains:
Before teaching me, ask up to five short diagnostic questions, one at a time. Work out whether my problem is the underlying concept, choosing a method, algebraic fluency or interpreting the question. Then tell me what you think the gap is and teach only that part Socratically.
You can also ask it to test the foundations beneath a topic:
I keep getting integration by substitution questions wrong. Test the prerequisite skills I need, one question at a time. Do not assume the problem is integration itself.
A-level maths is built on earlier maths. A problem with logarithms may really be a problem with indices. A calculus mistake may begin with rearranging a fraction. Repairing the foundation is slower than copying the answer to the current question, but it is much faster than making the same mistake for the rest of the course.
Control the size of the hint
When you are stuck, do not jump from no help to the complete solution. Ask for the smallest intervention that lets you continue.
Try this ladder:
- A question: “Ask me something that draws my attention to the useful information.”
- A nudge: “Name the relevant idea, but do not tell me how to apply it.”
- A first step: “Show only the first step, then stop.”
- A partial example: “Give me a simpler example using the same idea, then return to my question.”
- A solution: “Show the full solution and explain why each decision was made.”
Begin at the top. Move down only when the previous level genuinely gives you nothing to work with.
This preserves productive struggle. That does not mean staring at a blank page for forty minutes. It means spending enough time retrieving, choosing and testing ideas that the eventual help has somewhere to attach.
Ask “why?” more often than “what next?”
“What do I do next?” keeps a question moving. “Why is that valid?” builds a method you may recognise again.
Useful follow-up questions include:
Why did you choose that method instead of another one?
What feature of the question should have made me think of this method?
Which assumption are we using at this step?
What common wrong step would a student make here, and why is it wrong?
Can you ask me to explain this step back to you, then challenge anything vague in my explanation?
That final prompt is particularly valuable. Explaining an idea reveals gaps that are easy to hide while reading. Ask ChatGPT to be unconvinced by phrases such as “you just do this”, “it cancels” or “that is the rule”. If you cannot say what cancels, under which operation and why, your understanding is probably less secure than it feels.
Make it test the idea from another direction
Completing one guided question is not the finish. You need to find out whether the idea survived the conversation.
Once you have solved the original problem, ask:
Give me a new A-level question that tests the same idea but looks different on the surface. Do not tell me which method it uses. Let me attempt it without help, mark my working, and only then return to Socratic questions if I am wrong.
You can make the test more demanding:
Change one condition so that the method I just used no longer works. Ask me to identify what changed.
Give me three possible first steps, including two plausible mistakes. Ask me to choose one and justify it.
Ask me to create my own example where this method applies, then check it.
Return to this topic later in our conversation without warning and give me a retrieval question.
The aim is not to preserve the wording of the explanation. It is to recognise the structure when the surface changes.
A prompt you can reuse
Copy this at the beginning of a study conversation and change the parts in brackets:
I am an A-level maths student studying [topic] for [exam board, if relevant]. I want to learn, not receive answers.
Act as a patient Socratic tutor. Ask one short question at a time and wait for my response. Begin by checking what I already understand. Do not reveal a full method or solution unless I explicitly request it. If I make a mistake, do not immediately correct it: ask a question that helps me inspect the earliest incorrect step. If I remain stuck, give the smallest useful hint. Ask me to explain why important steps are valid and challenge vague explanations. When I reach an answer, ask me to check it, then give me one different-looking question that tests the same idea without naming the method.
Here is the question: [question]
Here is my attempt: [working, even if incomplete]
You do not have to use the whole prompt every time. Once ChatGPT begins giving too much away, remind it:
You have started explaining instead of questioning. Return to one Socratic question at a time. Do not reveal the next step.
Do not outsource checking
ChatGPT can make mathematical mistakes. It can misunderstand a diagram, silently change a condition or confidently use a method your specification does not require. Treat it as a tutor you can question, not an answer book you must trust.
For anything important:
- copy the original question accurately;
- keep exact values until the question asks for rounding;
- substitute answers back where possible;
- compare terminology and permitted methods with your specification;
- use an official mark scheme or a trusted worked solution to verify the final route.
If two sources disagree, do not ask AI to sound more confident. Ask it to state its assumptions, show the disputed step and test the result. Bring the question to a teacher when you still cannot resolve it.
Close the chat and do the question again
The final step happens without AI.
Close the explanation. Take a clean sheet of paper. Rebuild the solution from the question alone, saying why you are taking each important step. Then attempt a related question later, after enough time has passed that you must retrieve the method rather than imitate it.
When I improved at maths, it was not because somebody was always beside me supplying the next line. I improved by doing questions, discovering the gaps, getting help at the point where I was stuck and returning to the work until I could do it independently.
AI can make that help available almost whenever you need it. But it should not remove the struggle on your behalf. Ask it to question you, test you and make you explain yourself.
The best AI tutoring conversation should end with the AI doing less and you doing more.