A Structured Prompt Framework for AI-Generated Homework in Junior High Science: A Case Study of Molecular Thermal Motion

Abstract

Generative artificial intelligence (AI) offers potential support for designing homework on abstract scientific concepts, but the quality of generated tasks depends substantially on prompt design. This study examines a structured Role–Task–Audience–Goal prompt framework for junior high science homework, using molecular thermal motion as a single-case example. A three-round iterative design compared a zero-instruction baseline, a partially structured prompt containing Role and Task, and a refined prompt containing all four elements. Homework generated in the three rounds was independently evaluated by three junior high science experts using four dimensions: conceptual accuracy, contextual appropriateness, cognitive-level matching, and curriculum alignment. Descriptively, expert mean ratings increased across successive prompt rounds in all four dimensions, with the largest relative change observed in cognitive-level matching. The case analysis indicates that Role and Task help define the instructional identity, item format, and content scope, whereas Audience and Goal provide additional constraints related to learner characteristics and curriculum-linked outcomes. The study also considers transfer to other abstract science concepts with similar macro–micro mapping demands and identifies more complex mathematical topics as a boundary condition requiring additional prompting support. Overall, the case provides a practical, subject-specific procedure for teachers seeking to structure prompts for AI-assisted homework design while retaining teacher control over core pedagogical decisions.