Artificial Intelligence in Exercise Programming and Coaching: Opportunities and Limitations

Artificial intelligence has the potential to rapidly transform exercise prescription and coaching through machine learning, deep learning, and large language models. However, there is a disconnect between peer-reviewed evidence and real-world deployment in commercial platforms. This narrative review evaluated recent literature (2023-2025) assessing data-driven programming approaches for exercise prescription. Articles were identified across four themes in exercise programming: data integration on wearables and sensors, predictive modeling, chatbots and virtual coaches, and general-use large language models. Current evidence demonstrates the feasibility of artificial intelligence for activity recognition, workload estimation, and short-term performance prediction. However, significant gaps remain in evaluating closed-loop adaptive programming, downstream behavioral outcomes, and long-term effectiveness. Industry systems integrate multimodal data and continuously evolve, operating largely without formal validation. Advancement requires translational research models that bridge academic rigor with industry implementation timelines, prioritize transparency and human-in-the-loop frameworks, and evaluate artificial intelligence as a tool to augment rather than replace professional exercise prescription.