Why do AI chatbots make mistakes with maths?

It can be surprising: an AI that explains calculus beautifully sometimes gets a simple multiplication wrong. This is a known limitation, and it has a clear explanation.

Why does AI struggle with arithmetic?

1. It predicts text, it does not calculate. A language model generates the most likely next pieces of text. For "2 + 2 =" the answer is so common that it is reliable. For "4,837 × 291 =" the model must effectively predict digits, which can go wrong.
2. Numbers are split into tokens. Large numbers may be broken into chunks in ways that make digit-by-digit calculation harder.
3. Long calculations compound errors. One small slip early in a multi-step problem carries through to the final answer.
4. Word problems need interpretation. The AI might misunderstand which quantity to use or what the question asks.

Modern AI has improved a lot. Reasoning models that think step by step make far fewer mistakes, and many assistants can run code to calculate exactly. But errors still happen, especially in long calculations, percentages, compound interest, unit conversions and anything involving many numbers.

How to get reliable maths from AI:

1. Ask it to use code or a calculator tool. "Use Python to calculate this" makes many assistants run actual code, giving precise results.
2. Use reasoning or thinking modes for multi-step problems.
3. Ask for step-by-step working, so you can check each step.
4. Verify with a calculator. For anything important, such as invoices, tax calculations, loan EMIs or exam answers, check key numbers yourself.
5. Use spreadsheets for repeated calculations. Ask AI to write the Excel formula, then let Excel do the arithmetic.
6. Restate the problem clearly. Give all numbers and specify units and rounding rules.

For students, AI is excellent for understanding how to solve maths problems: explaining concepts, showing methods and suggesting approaches. But do not rely on it for final numerical answers without checking. In exams, you will need to calculate yourself anyway.

For businesses, never let AI-generated figures go into financial documents, quotations or tax filings without verification. Use AI for the logic and structure, and use proper tools for the numbers.