Generative AI-Assisted Exercise Design in Primary Mathematics

Abstract

The instructional value of generative artificial intelligence in primary mathematics depends on how generated exercises connect mathematical content, representation, and evidence of understanding. This conceptual article develops a framework for that connection through a focused narrative synthesis of research on AI-supported instructional materials, mathematics pedagogy, fraction instruction, and formative assessment. Five design principles specify goal alignment, mathematical completeness, cognitive demand, representational accessibility, and diagnostic potential. A preparation workflow translates these principles into task specification, candidate generation, mathematical verification, pedagogical revision, and sequencing. Three worked examples address equal partitioning, fraction magnitude, and the reference whole. Each links a design problem to a complete exercise, a solution rationale, and an instructional response. The analysis identifies representational precision as a central condition of exercise quality: the whole, partition, scale, and comparison conditions must support the intended inference. The framework connects AI assistance to decisions that teachers can articulate and inspect, providing a practical basis for developing conceptually coherent exercises. The examples are constructed illustrations; the framework’s usability and educational effects require empirical evaluation.