The role of prompt, voice, and personality factors in the acceptance and evaluation of AI-generated mindfulness exercises

Abstract

Abstract AI-generated mindfulness exercises have the potential to provide tailored mindfulness interventions. However, the role of quality of AI-generated mindfulness exercises on their acceptance and evaluation is as of yet underexplored. The present work investigates effects of prompting (tailored versus non-tailored versus human), voice (trained versus non-trained versus human), and matching voice personality factors on the uncanniness, human likeness, and acceptance of AI-generated mindfulness exercises. In two experiments, n = 143 participants rated real and different variations of AI-generated mindfulness exercises. It was found that trained AI voices significantly improve the evaluation of AI-generated mindfulness exercises comparable to human controls, while no significant effects of prompting were found. A categorization task further showed that exercises by trained AI voices were indistinguishable from human mindfulness exercises. Furthermore, a mismatch of voice personality and setting (mindfulness) significantly decreased voice evaluation. The results demonstrate the importance of using trained and matching AI voices for the implementation of AI-generated mindfulness interventions.