Can AI Replace an IGCSE Maths Tutor?

Your child gets stuck on a question, opens ChatGPT, and has a full worked solution back within seconds. It looks right, and it's natural to start wondering whether a tutor is still needed at all.

Most parents we speak to aren't looking for a reason to ban AI at home, and that's probably not a realistic goal in any case. The more useful question is a narrower one: what AI is actually good at for IGCSE maths, where it tends to fall short, and what a tutor should still be doing in 2026 that AI can't.

Where AI helps

AI is genuinely useful as an explainer. It can go over completing the square several different ways late at night without losing patience, and a child can ask the same question twice without feeling embarrassed about it. It's also an effectively unlimited source of practice questions. For a subject like maths, where progress depends on repetition and quick feedback, that's a real advantage.

This matches what parents say when they're asked directly. In a 2026 US survey on AI in schools, the argument parents found most persuasive in favour of AI in education was its ability to personalise learning to a child's own pace, ahead of arguments about future careers or job readiness. Parents seem to value that patience and personalisation. What they're less certain about is what happens after.

Where AI falls short for IGCSE maths

It makes mistakes, and it tends to make them on the questions that decide the grade. AI language models don't calculate the way a calculator does. They predict what a correct-looking answer should read like, based on patterns in the text they were trained on. That works reasonably well for routine steps and becomes less reliable as a problem gets longer. One 2026 analysis found current AI chatbots still make arithmetic errors in roughly three in ten complex, multi-step calculations. A study testing ChatGPT against real maths exam questions found a more specific pattern: close to full accuracy on low-mark, single-step questions, dropping to around fifty percent on the higher-mark questions that require several linked steps of reasoning.

This matters for IGCSE in particular, because it mirrors how the paper is structured. The one and two-mark questions near the start rarely decide the final grade. The four, five, and six-mark multi-step questions in the second half of the paper, the ones that separate a 6 from an 8, are where AI tends to be least reliable.

It doesn't know how Cambridge or Edexcel actually mark the paper. Both exam boards award most of the available marks for method, not for the final answer. A four-mark question is often structured so a student can earn most of the marks with a correct approach and a wrong final number, provided the working is visible and follows the expected steps. Some questions specifically require a stated reason, not just a calculated value, and lose marks if that line is missing even when the number is right. An AI chatbot has no concept of any of this. It produces a route to a correct-looking final answer, not a script written the way an examiner is trained to credit. A student who copies AI-style working onto their own exam paper can do the maths correctly and still lose marks, because the working isn't presented the way the mark scheme rewards.

It has no memory of a child's specific pattern of mistakes. A tutor working through six weeks of past papers with a student starts to notice things that repeat: always dropping a negative sign when expanding brackets, always forgetting units in context questions, always mishandling upper and lower bounds. That pattern only becomes visible across many attempts, over time, from someone paying attention. Each conversation with an AI chatbot is answered on its own, with no memory that this is the twentieth time the same error has come up.

It can slow learning down if it's used the wrong way. Research comparing the two approaches found that students who attempt a problem first, then use AI to check their reasoning, learn more than students who use AI to complete the problem before attempting it themselves. The second pattern, getting the answer first, is easy to fall into under deadline pressure, and it can undermine the problem-solving skill an IGCSE paper is actually testing. This is close to the dependency that parents already say they're worried about when their children use AI for schoolwork.

Where this leaves things

AI is a useful explainer and a good source of practice questions. It isn't an examiner, it has no accountability for whether a child improves over ten months, and it doesn't distinguish between an answer that's mathematically correct and one that's correct in the way Cambridge or Edexcel actually mark it.

That changes what a tutor's role looks like in 2026. Less time goes into reteaching a concept from the beginning, since a chatbot can often do a reasonable first pass at that on its own. More time goes into the parts AI isn't set up to do: working through real past papers against the real mark scheme so a student learns to write answers examiners are trained to credit, catching the specific recurring error that's costing marks before it becomes a habit, and holding the structure and accountability across a full syllabus that a chat window isn't built to provide.

If your child is already using AI to help with their maths, that isn't something to worry about on its own. The more useful question is whether anyone is checking that what it's teaching them lines up with what the exam board actually rewards.