Physical education teachers increasingly use generative AI tools to support lesson planning, yet the effectiveness of AI-generated plans relative to human-designed and collaboratively developed plans remains empirically untested. This three-arm randomized parallel-group study (N = 90 secondary school students; Kef, Tunisia) compares AI-only, human-only, and human–AI hybrid session design conditions across four outcomes: basketball shooting accuracy, lesson plan quality (AQTR), student engagement (PACES), and planning efficiency. Data collection is complete. This is a post-enrollment, pre-analysis registration—no inferential analyses have been conducted at the time of registration.