Human expertise and artificial intelligence in coaching: A comparative analysis of feedback for triathlon front crawl technique

Abstract

Recent advances in AI create new opportunities for performance analysis in endurance sports. This study compared expert human coaching with GPT-4o in biomechanical analysis and technique feedback for triathlon front crawl. Fifteen certified coaches and an AI agent analysed a 25-frame video of an intermediate triathlete completing one stroke cycle. Both identified errors, prioritised corrections, and suggested drills to address them. Responses were thematically analysed and mapped to four stroke phases: glide, pull, push, and recovery, within a structured feedback framework based on a deterministic biomechanical model. Human coaches emphasised holistic aspects (such as breathing, alignment, and rhythm), while AI provided detailed biomechanical cues (e.g., early vertical forearm, stroke length) and drill-specific recommendations. Agreement was strongest on high-impact errors in the catch phase. Findings suggest that AI can complement coaching by enhancing error detection, providing structured feedback, and reducing cognitive load, while preserving the coach's role in contextual adaptation and individualised strategies.