This study aims to empirically test the theoretical mechanisms of relational perceptions in the context of building and testing a relational artificial intelligence (AI) chatbot for improving physical activity (PA) behaviors among a sedentary adult population in the U.S.

The aim of the study is to build and experimentally test relational capacities of AI chatbot in inducing positive human-AI relationship and leading to higher PA behavior change intention. During the 7-day intervention, the relational chatbot will educate participants on physical activity using 5 types of relational messages during a PA intervention including 1) social dialogue, 2) empathy, 3) self-disclosure, 4) meta-relational communication, and 5) humor. On the other hand, the non-relational chatbot will only deliver PA intervention messages, without relational cues. Relational chatbot condition will be compared to the non-relational chatbot condition to assess its effectiveness.

The objective of this study is to test the efficacy of the mobile app intervention leveraging chatbots in increasing participants' relationship perception and physical activity behavior change.

Despite the recognition of the importance of communication for relationship and trust building in healthcare programs (Ward, 2018), theoretical work and empirical testing of relational capacities in AI chatbot designed for changing health behaviors is lacking in previous research. In our systematic review of AI chatbot-based interventions on PA, diet, and weight loss outcomes, it was found that although majority of studies programmed AI chatbots to engage in relational communication behaviors (e.g., personalized greetings, showing empathy and compassion), none of these studies tested whether AI chatbot's relational capacities contributed to human-AI trust and relationship building (Oh et al., 2021). Beyond knowing some chatbots were perceived to be useful and friendly, we do not know theoretical mechanisms for what relational conversational strategies contribute to higher quality relationship perception and behavior outcomes.

By developing an AI chatbot that provides access to informational, motivational, and socio-emotional aspects of care will open new opportunities for delivering accessible interventions to improve sedentary population's PA behaviors. More broadly, the test of the AI Chatbot Behavior Change Model (Zhang et al., 2020) will provide the first empirical evidence on the model's utility and working mechanisms accounting for how relational AI chatbot can change health behaviors. If successful, the theoretical model and design of the relational chatbot will be able to generalize to related behavior change fields.

This field experiment will randomly assign participants to relational chatbot condition or non-relational (control) chatbot condition. Both conditions will involve an information component addressing physical activity education sessions.

The objective of this study is to test the efficacy of the mobile app intervention leveraging chatbots in increasing participants' relationship perception and physical activity behavior change.

Inclusion Criteria:

• Living in the United States
• Adults (18 years or older)
• Able to speak and read English
• Physically inactive at work and/or during leisure time
• Having internet access to use the chatbot
• Smartphone users

Exclusion Criteria:

• Adults unable to consent
• Individuals who are not yet adults (infants, children, teenagers)
• Prisoners
• Pregnant women
• Physically active individuals who are meeting physical activity guidelines
• Individuals who have participated in a physical activity intervention in the past 12 months

Individual participant data will not be shared with other researchers.