ASSESSING THE IMPACT OF AI-MEDIATED VERSUS HUMAN COACHING ON PHYSICAL PERFORMANCE

Coaching influences physical performance through both physical instruction and by shaping how effort is perceived and regulated during demanding tasks. The use of artificial intelligence (AI) in digital health and exercise settings is increasing, but it is unclear whether AI-mediated coaching influences performance in a manner comparable to human coaching. This study aims to examine how different coaching modalities affect performance control, physiological load and perceptual responses during high-effort exercise.
Using a within-subject repeated-measures design, healthy male and female participants aged 18 -30 years completed three experimental conditions: no coaching, AI-mediated digital coaching delivered via adaptive text-to-speech prompts and live human coaching. Data has been recorded from 40 subjects, with recruitment and data collection ongoing. Habitual physical activity levels were assessed using the International Physical Activity Questionnaire. Under each coaching condition, participants performed three physical tasks: a wall sit to volitional failure with a four-minute cap, a 60-second maximal handgrip task, and a 30-second Wingate anaerobic cycling sprint. Performance measures were collected alongside physiological indicators of exercise load including heart rate and muscle activity. Ratings of perceived exertion and session-specific motivation were recorded following each task to capture responses to the different coaching conditions.
Preliminary results indicate that both live and AI coaching result in better performance when compared with no coaching, and their effect on performance differs between tasks. This work informs research on performance control and may support the development of AI-based coaching systems that better reflect responses to feedback during exercise.