When AI draws forces: evaluation of free-body diagrams generated by ChatGPT, Gemini, and Copilot

Abstract Generative AI tools are increasingly capable of producing visual representations, but their accuracy and pedagogical reliability in physics education remain underexplored. This study investigated the capacity of three AI systems—ChatGPT, Gemini, and Copilot—to generate free-body diagrams (FBDs) for a mechanics problem and evaluated the quality of the outputs through expert review. The analysis focused on the correctness of force identification, placement, and vector orientation, as well as adherence to conventional FBD practices. Results revealed persistent limitations across all tools, including inaccurate force application points, misrepresented directions, and the inclusion of non-force quantities or unnecessary contextual elements. While Copilot produced slightly more coherent diagram, none fully met the standards for accurate conceptual representation. The findings suggest that, although AI can serve as a rapid tool for generating illustrative diagrams, students must critically evaluate these outputs, and educators should screen and guide their use to prevent the propagation of misconceptions.