From generative-AI feedback to sport skill learning: the roles of perceived competence and autonomous motivation in university physical education.

INTRODUCTION

Personalized feedback can support motor learning while also increasing reliance on external guidance. This study evaluated a structured personalized generative-artificial-intelligence (GenAI) feedback package in university physical education, focusing on immediate sport-skill acquisition, 14-day delayed retention, perceived competence, and autonomous motivation.

METHODS

In a two-arm randomized trial, 240 undergraduates completed an 8-week basketball unit and received either constrained written GenAI feedback generated from structured performance indicators or conventional oral instructor feedback matched for curriculum, practice dose, technical rubric, and feedback timing. Free-throw accuracy and blinded technique quality were assessed at baseline (T0), immediately after instruction (T1), and 14 days later without feedback (T2). Technique quality and shooting accuracy at T2 were co-primary endpoints with Holm control. Perceived competence and autonomous motivation were measured during the intervention, and mixed-model, binomial, robustness, and bootstrap indirect-effect analyses were conducted.

RESULTS

The GenAI-feedback package produced larger T0-T1 gains in shooting accuracy (differential gain = 6.7 percentage points, 95% CI [4.3, 9.1], d = 0.61) and technique quality (adjusted difference = 1.74 points, 95% CI [1.14, 2.34], d = 0.70). At T2, adjusted between-condition differences remained 6.1 percentage points for accuracy and 1.63 points for technique quality; both co-primary criteria were met after Holm correction. The GenAI condition also showed higher mid-intervention perceived competence and autonomous motivation. Serial indirect effects were statistically consistent with the proposed competence-to-autonomous-motivation sequence, but alternative parallel and reversed-order models were also compatible with the data.

DISCUSSION: The findings support the complete structured GenAI-feedback package relative to an active instructor-feedback package, not an isolated AI-specific effect. The 14-day result supports delayed retention over the tested interval rather than durable long-term learning, and the competence-motivation sequence should be interpreted as theory-consistent statistical evidence rather than a confirmed causal chain.