This mixed-methods systematic review synthesises current evidence on the application of generative artificial intelligence (GAI) in sport and exercise science. Following PRISMA guidelines, 64 eligible studies were identified from Web of Science, PubMed, SPORTDiscus, Scopus and other sources, and methodological quality was appraised using JBI critical appraisal tools. Evidence was integrated using bibliometric mapping, Latent Dirichlet Allocation topic modelling, and narrative synthesis. Findings indicate that GAI is increasingly applied across physical education, personalised training, exercise rehabilitation, and health promotion. Two dominant research trajectories were identified: (i) methodological evaluation of GAI-generated content and (ii) application oriented deployment in real-world practice. Overall, GAI demonstrates moderate to good performance in content generation, guideline adherence, and decision-support efficiency, particularly under structured, high-information prompting. However, most studies remain evaluative or exploratory, with limited high-quality randomised trials, short intervention durations, and insufficient real-world validation. Persistent limitations include inadequate individualisation, instability across outputs, safety concerns in high-risk populations, and emerging ethical challenges related to data governance, bias, and accountability. In conclusion, GAI represents a promising adjunct tool in sport and exercise science, but current evidence is insufficient to support widespread implementation or paradigm-level change. Future research should prioritise rigorous experimental designs, clinically and performance-relevant outcome measures, domain-specific system development, and robust ethical governance frameworks.