Creating Math Practice Problems with AI

Practice problems should serve a mathematical purpose. A set might build fluency with a procedure, help students interpret a representation, strengthen understanding of a concept, or develop problem-solving strategies. Before asking AI to create items, decide which goal matters and what students have already learned. A request that says only “make math problems” can produce mismatched grade levels, ambiguous wording, incorrect answers or repetitive exercises that do not address the intended reasoning. The What Works Clearinghouse practice guide on mathematical problem solving in grades 4–8 offers evidence-informed recommendations for educators. IES materials also describe the value of generating and discussing more than one solution strategy, which can help students reason flexibly about when approaches fit. Those principles suggest that practice need not consist solely of answer drills. Students can represent a situation, explain why an operation fits, compare methods or identify a mistake. The exact balance depends on the unit and learner readiness. AI can rapidly draft parallel items, change surface contexts, suggest hints or create a worked example for educator review. It can also silently alter the mathematics. A supposedly equivalent problem might change a denominator, omit a condition, mix units or have no valid solution. Solve each item independently, check representations and confirm that distractors reflect plausible reasoning errors rather than arbitrary wrong answers. If an item is intended to have multiple strategies or solutions, establish those before presenting it. Use the model as a drafting assistant and keep students’ mathematical thinking at the center. A useful prompt states the grade band, topic, prerequisite skills, target reasoning and constraints, and asks for an answer key and verification steps. Educators should not use AI-generated difficulty labels or scores without judgment. Watch how students respond, offer supports that make the problem accessible, and revise items that assess reading complexity or hidden assumptions instead of mathematics. Protect student data and follow local rules for external tools.