Feasibility of a teacher-supervised ChatGPT-assisted workflow for individualized exercise planning in university physical education: a two-phase study using a fuzzy Delphi process and a cluster pilot trial.

BACKGROUND

University physical education can promote physical activity and fitness, but individualized instruction is difficult to implement in routine classes. This study developed a structured input framework for ChatGPT-assisted individualized fitness teaching and evaluated its feasibility, safety, acceptability, and preliminary effects.

METHODS

This two-phase study involved non-sports-major university students. Phase I used a two-round fuzzy Delphi process to develop the framework. Phase II was a 6-week cluster pilot trial involving two intact classes allocated by coin toss to either AI-assisted instruction (n = 30) or conventional instruction (n = 30). The AI group received teacher-supervised, AI-generated fitness plans. Primary outcomes included feasibility, safety, implementation, and acceptability.

RESULTS

The final framework comprised 49 items, with 96.4% achieving a consensus rate ≥75%. Follow-up completion rates were 96.7% and 93.3% in the AI and control groups, respectively. Attendance, completion rate, and perceived exertion were similar between groups. Time in the target heart-rate zone was higher in the AI group (33.95 ± 4.33 vs. 32.60 ± 3.74 min, P = 0.033), as was perceived personalization (4.0 ± 0.7 vs. 3.2 ± 1.0, P = 0.001). No serious adverse events occurred, and most physical and functional outcomes did not differ significantly between groups.

CONCLUSIONS

ChatGPT-assisted individualized fitness instruction was feasible, safe, and acceptable within a structured teacher-supervised workflow, with its primary short-term benefit being improved perceived personalization.Clinical Trial Registration: Chinese Clinical Trial Registry ChiCTR2500107730.