Math: Step-by-Step Problem Solving and Exercises

The Domain Where It All Comes Together

Math is arguably the subject where ChatGPT delivers the most immediate value—and where the risk of misuse is most obvious. A student who pastes in an exercise gets a neatly written solution in seconds. The entire pedagogical challenge lies in putting the tool at the service of the student’s process, not in place of it.

Make the Model a Tutor, Not a Calculator

By default, a language model answers the question you ask. Two levers change this behavior. The first is built into ChatGPT: Study Mode, which asks questions, proceeds step by step, and checks for understanding instead of handing over the answer. You activate it by typing @study, using the + menu, or going to chatgpt.com/studymode. The second is a tutor prompt that you write once and paste at the start of a conversation, or place in a Project’s custom instructions:

You are a math tutor. When a student asks you a problem:
1. NEVER give the final answer directly
2. First ask what the student has already tried
3. Identify the step where they are stuck
4. Give a hint for that step only
5. Let the student continue
6. Check their answer at the end and correct if necessary

Let’s see what this looks like. A high school student studying quadratics asks to solve 2x² + 3x − 5 = 0. With this prompt, ChatGPT doesn’t display the solutions. Instead, it asks which method the student knows for this type of equation. The student mentions the discriminant; the tutor asks them to recall the formula, then asks what result they get for b² − 4ac with a = 2, b = 3, and c = −5. The student calculates, ChatGPT verifies, and guides them to the two solutions. At every step, the student does the work; the tool provides only the scaffolding, and that scaffolding is removed.

Keep the context in mind: this student-led tutoring has a place in the classroom only starting in 8th grade (4ème), and in high school within the boundaries you set. Before that, you lead the exchange at the board, guiding the whole class.

Build Custom Exercise Sets

The second benefit addresses a situation every teacher knows: in a single middle school class, some students still struggle with the basics while others are bored. For fractions, ask for a progression in four levels—identical denominators like 3/7 + 2/7, then denominators that are multiples of each other like 1/3 + 1/6, next arbitrary denominators like 2/5 + 3/7, and finally word problems. Each student works at the level that moves them forward, and you can generate ten additional exercises for a given level in seconds. This lets you circulate among the rows instead of running copies.

Vary the skills you target, because not all exercises are equal. Calculating is calculation with immediate correction. Reasoning requires an open-ended problem where the process matters more than the result. Modeling demands translating a real-world situation into an equation: the price of three items with a discount, the average speed of a trip. Communicating means writing out a proof or explaining a method to a peer. A balanced sequence draws from these families rather than piling up twenty calculations. And always check every answer key: a generated problem statement can contain inconsistent data or a wrong answer.

How the Reasoning Level Changes Things

In ChatGPT’s model selector, the Thinking slider sets GPT-6’s reasoning effort: Instant for quick responses, Medium for standard reasoning, High for deep reasoning, and Extra High for the most advanced level. On Free and Go plans, the Think button plays this role, and available options depend on your plan. For math exercises, set it to at least Medium: the model works through its steps before answering, which reduces errors without eliminating them. Also ask for formulas in LaTeX when you want to reuse them in your own documents.

From Mental Math to Proof by Induction

In elementary school, teachers use it for preparation: multiplication tables turned into games, word problems drawn from daily life like grocery shopping or sharing a cake, unit conversions backed by concrete examples. In middle school, students enter algebra, equations, functions, statistics, and probability—all chapters where generating progressive exercises shines. In high school, the tutor supports analysis: function studies, sequences and proof by induction, conditional probability, and solid geometry.

To get started, write your tutor prompt and place it in a Project dedicated to math, set the reasoning level, then generate a series of ten progressive exercises on the current chapter. Finally, test the guidance by playing the impatient student who insists on getting the answer: if the model eventually gives in, tighten the prompt. And teach your students to verify a sensitive result: that is a mathematical skill in its own right.

Your Turn

Put the tutor to the test.

1. Write your tutor prompt and place it in a Project’s custom instructions
2. Set the reasoning level to Medium or High
3. Generate ten leveled exercises on the current chapter, then verify every answer key
4. Play the impatient student: demand the answer three times, in three different ways
5. Tighten the prompt until the model holds firm, then test Study Mode on the same problem

Official sources: Using study mode in ChatGPT · GPT-6 and other models in ChatGPT · Projects in ChatGPT · Getting started with ChatGPT as an educator (OpenAI Help Center) · Les intelligences artificielles et leurs usages en éducation (éduscol)

Key Takeaways

• Study Mode and a tutor prompt enforce guidance rather than providing answers
• Student-led tutoring begins in the classroom starting in 8th grade (4ème)
• Leveled exercises let each student advance at their own pace, provided you verify the answer keys
• For math, increase the reasoning level: errors decrease but do not disappear
• Verifying a result is a skill to teach