AI-Personalized Digital Exercise for University Students (AIPDEUS)

This was an 8-week, two-group randomized controlled trial conducted among undergraduate students at Transilvania University of Brașov, Romania. The study compared an AI-personalized digital exercise programme with a standardized digital exercise programme delivered within the same digital study environment.

Participants assigned to the AI-personalized group received exercise recommendations through MoveAI-Student version 2.1.0 using the version-controlled v1.3-adaptive system. The system combined prespecified safety rules with a supervised recommendation model. Personalization used participant characteristics, sport participation, weekly training volume, countermovement jump performance, simple and choice reaction time, readiness, pain, previous-week adherence, and safety information. Exercise recommendations could specify session frequency, duration, intensity, and additional emphasis on lower-limb power or reactive drills. Recommendations were updated once per intervention week. Progression was limited by predefined safety rules, and exercise load could be reduced when pain or an adverse-event flag was present. Human override of recommendations was permitted when required.

Participants assigned to the standardized digital comparator used the same digital study environment but did not receive the adaptive weekly prescription procedure. The comparator programme prescribed three exercise sessions per week, with each session lasting 35 minutes.

The primary outcome was mean weekly exercise adherence, calculated from completed sessions relative to prescribed sessions across the 8-week intervention. Secondary outcomes included total physical activity assessed with the International Physical Activity Questionnaire Short Form, WHO-5 wellbeing, countermovement jump performance, simple reaction time, choice reaction time, moderate-to-vigorous physical activity, application engagement, satisfaction, perceived usefulness, usability, sitting time, body weight, body mass index, and adverse events.

The study was designed to estimate the effect of the adaptive programme as a complete intervention compared with standardized digital exercise. It was not designed to isolate the independent effect of the algorithm from differences in exercise prescription or completed exercise exposure.