Regular physical activity plays a critical role in health promotion and athletic performance, necessitating personalized exercise and training prescriptions. While traditional methods rely on expert assessments, artificial intelligence (AI), particularly generative AI models such as ChatGPT and Google Gemini, has emerged as a potential tool for enhancing personalization and scalability in training recommendations. However, the applicability, reliability, and adaptability of AI-generated exercise prescriptions remain underexplored. A comprehensive search was performed using the UnoPerTutto metadatabase, identifying 2891 records. After duplicate removal (1619 records) and screening, 61 full-text reports were assessed for eligibility, resulting in the inclusion of 10 studies. The studies varied in methodology, including qualitative assessments, mixed-methods approaches, quasi-experimental designs, and a randomized controlled trial (RCT). AI models such as ChatGPT-4, ChatGPT-3.5, and Google Gemini were evaluated across different contexts, including strength training, rehabilitation, cardiovascular exercise, and general fitness programs. Findings indicate that generative AI-generated training programs generally adhere to established exercise guidelines but often lack specificity, progression, and adaptability to real-time physiological feedback. AI-generated recommendations were found to emphasize safety and broad applicability, making them useful for general fitness guidance but less effective for high-performance training. GPT-4 demonstrated superior performance in generating structured resistance training programs compared to older AI models, yet limitations in individualization and contextual adaptation persisted. A critical appraisal using the METRICS checklist revealed inconsistencies in study quality, particularly regarding prompt specificity, model transparency, and evaluation frameworks. While generative AI holds promise for democratizing access to structured exercise prescriptions, its role remains complementary rather than substitutive to expert guidance. Future research should prioritize real-time adaptability, integration with physiological monitoring, and improved AI-human collaboration to enhance the precision and effectiveness of AI-driven exercise recommendations.
Harnessing Generativ e Artificial Intelligence for Exercise and Training Prescription: Applications and Implications in Sports and Physical Activity—A Systematic Literature Review
Puce L.;Bragazzi N. L.;Trompetto C.
2025-01-01
Abstract
Regular physical activity plays a critical role in health promotion and athletic performance, necessitating personalized exercise and training prescriptions. While traditional methods rely on expert assessments, artificial intelligence (AI), particularly generative AI models such as ChatGPT and Google Gemini, has emerged as a potential tool for enhancing personalization and scalability in training recommendations. However, the applicability, reliability, and adaptability of AI-generated exercise prescriptions remain underexplored. A comprehensive search was performed using the UnoPerTutto metadatabase, identifying 2891 records. After duplicate removal (1619 records) and screening, 61 full-text reports were assessed for eligibility, resulting in the inclusion of 10 studies. The studies varied in methodology, including qualitative assessments, mixed-methods approaches, quasi-experimental designs, and a randomized controlled trial (RCT). AI models such as ChatGPT-4, ChatGPT-3.5, and Google Gemini were evaluated across different contexts, including strength training, rehabilitation, cardiovascular exercise, and general fitness programs. Findings indicate that generative AI-generated training programs generally adhere to established exercise guidelines but often lack specificity, progression, and adaptability to real-time physiological feedback. AI-generated recommendations were found to emphasize safety and broad applicability, making them useful for general fitness guidance but less effective for high-performance training. GPT-4 demonstrated superior performance in generating structured resistance training programs compared to older AI models, yet limitations in individualization and contextual adaptation persisted. A critical appraisal using the METRICS checklist revealed inconsistencies in study quality, particularly regarding prompt specificity, model transparency, and evaluation frameworks. While generative AI holds promise for democratizing access to structured exercise prescriptions, its role remains complementary rather than substitutive to expert guidance. Future research should prioritize real-time adaptability, integration with physiological monitoring, and improved AI-human collaboration to enhance the precision and effectiveness of AI-driven exercise recommendations.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
