2026 ASEE Annual Conference & Exposition

Artificial Intelligence (AI) is increasingly embedded in higher education, and students now routinely use open-access AI tools to support their learning. In engineering education, research has shown that spatial visualization ability is a key indicator of entry and success within STEM fields, so students may be inclined to use AI for spatial skill training and assessment. This study investigates: (1) how effective widely available AI language models are at completing spatial visualization tasks and (2) how effectively they explain the reasoning processes behind their answers. Four established instruments are used: the Purdue Spatial Visualization Test (PSVT), the Differential Aptitude Test: Space Relations (DAT:SR), the Mental Rotation Test (MRT), and the Mental Cutting Test (MCT). These tests were administered to several popular AI systems that students may have access to (e.g., ChatGPT, Claude, Gemini, Perplexity), and quantitative data was collected on the accuracy of answers for each test across repeated runs to capture non-deterministic behavior. A common prompt sequence was administered to all AI tools to model realistic student-like interactions and change in both answers and explanations were examined across runs. Overall performance on tests across all AI systems was well below typical post-instruction of human performance reported in the literature, suggesting that current systems struggle with complex visual-spatial tasks relative to textual problems. Rather than directly evaluating human learning or training outcomes, this work benchmarks current AI performance and explanation quality on spatial visualization tasks and uses these findings to inform when and how such tools might be used as supplementary supports for spatial skills development in engineering education. Future work will involve more in-depth tests for each instrument, modified item formats that may be more compatible with current AI tools, and studies that connect model behavior to actual student learning outcomes.