One Question, Many Forms: Isomorphic Physics Test Design with Generative AI

We present recent progress on leveraging generative AI (ChatGPT) to create isomorphic multiple-choice assessment items in physics education. Our customized GPT employs a structured prompting framework with Python execution to validate answer algorithms. The system analyzes a seed question, identifies the underlying solution path, and then generates corresponding distractors by extracting and recombining algorithmic variants. In addition, a built-in feedback mechanism allows users to rate the quality of generated items, providing data for iterative refinement of the model's question-generation function. This approach demonstrates a scalable pathway toward high-quality, automatically generated test items that preserve conceptual equivalence while diversifying assessment formats.