Optimizing generative artificial intelligence prompts to engage urges for physical activity in middle-aged and older adults

Background Most middle-aged and older adults do not engage in sufficient physical activity. Text messages have proven effective for promoting physical activity, but little is known about how message content can engage motivational mechanisms. Purpose This study aimed to examine how generative artificial intelligence (AI) prompts could be engineered to create messages that engage physical activity urges. Methods Study 1 involved iterative prompt development and expert evaluations of messages, varying by physical activity behavioral context (preparation vs. execution), benefit horizon (short-term vs. long-term), temporal focus (recalled past vs. imagined future), and behavioral target (physical activity vs. sedentary behavior). Study 2 involved a web-based factorial experiment to assess how these factors affected urges among middle-aged and older adults. Results In Study 1, ratings of confidence that text messages (nMessages = 16) would evoke urges were moderate yet heterogenous across the experts (nExperts = 15). Themes including physical activity cues and affective appeal emerged as key factors. In Study 2, 639 adults (aged 40-85 years, M = 57; 52% female) rated 80 messages. Men reported stronger urges than women, and participants with stronger baseline urges reported stronger urges after reading the messages. Age moderated the effect of context, such that the execution (vs. preparation) prompt advantage increased with age. Two significant 3-way interactions showed that execution-based prompts outperformed preparation prompts, except when prompts targeted (1) long-term future benefits and (2) future physical activity experiences. Conclusions This study identified how specific generative AI prompt features can evoke physical activity urges in middle-aged and older adults, supporting the development of motivationally targeted interventions.