Background Artificial intelligence (AI) and machine learning (ML) are transforming sports sciences by enabling precise performance analysis, injury prevention, and rehabilitation. However, inconsistencies in protocols, lack of standardization, and practical deployment challenges limit their real-world impact. Methodology This PRISMA-guided systematic review was registered with PROSPERO (CRD42021235527). Literature searches were conducted across Scopus, PubMed, and Web of Science for studies published between 2018 and 2024. The PICO framework guided selection, and methodological quality was assessed using AMSTAR-2. Results A total of 40 studies were included. CNNs (n = 18; accuracy = 96 ± 1.5%) and LSTMs (n = 10; 92 ± 2%) performed best for motion tracking and gait analysis. SVMs (n = 17; 81 ± 3%) and RFs (n = 16; 80 ± 2.8%) were widely used in EMG, GPS, and workload analysis. Real-time feedback (n = 18) showed latency below 100 ms but was mainly validated in controlled settings; only 25% tested systems outdoors. Wearable sensors (e.g., IMUs, gyroscopes) were used in 34 studies, most frequently on the lower limbs. Participants averaged 42 ± 28 in number, with a bias toward young, healthy males. Key limitations included algorithm opacity, small homogeneous samples, inconsistent sensor setups, and limited interpretability or clinical integration. Conclusion AI/ML tools show high accuracy and responsiveness for biomechanical monitoring, injury risk detection, and training adaptation. However, scalability remains hindered by technical, demographic, and practical barriers. Future studies must prioritize diverse populations, standardize sensor and data protocols, emphasize model explainability, and ensure deployment in real-world sports environments to close the gap between algorithmic performance and applied utility. Trial registration CRD42021235527
© Copyright 2025 Sport Sciences for Health. Springer. Alle Rechte vorbehalten.