Using AI-Generated Videos to Simulate Different People Communicating the Same Healthcare Needs: An Exploratory Study of Care Prioritisation and Trust Across Health Professional and Public Groups

Abstract

Digital technologies increasingly shape how health information is presented and interpreted in communication-intensive contexts. In healthcare, such communication is cognitively and emotionally demanding, particularly in contexts involving stigma and uncertainty. While evaluations of artificial intelligence (AI) systems typically focus on performance and usability, less is known about how different viewer groups interpret AI-generated health communication in these sensitive care contexts. This exploratory study examines whether AI-generated video case studies can support controlled comparison of how health professionals and members of the public interpret healthcare needs and care priorities. Four AI-generated video case studies were developed, each simulating a client profile presenting concerns related to obesity and mental health within a healthcare consultation, delivered as a one-way, non-interactive communication. Each case followed a shared core narrative with structured variation in sociocultural context, creating distinct client profiles and enabling comparison of interpretation under consistent informational conditions. Participants were randomly assigned to view one video within an online survey and completed ranking tasks on care priorities alongside evaluations of AI's perceived impact on communication and trust. Across participant cohorts, psychological and lifestyle-related care were prioritised over biomedical interventions, indicating a shared biopsychosocial framing of care. Responses to AI-generated communication were characterised by ambivalence, with concerns centred on loss of human touch, oversight, and relational risk rather than technical performance. Variability in viewer perception was observed across AI-generated client profiles among public participants, suggesting sensitivity to sociocultural cues despite standardised content. These findings suggest that ambivalence and variability may reflect adaptive responses to cumulative cognitive, emotional, and organisational demands rather than resistance or error. Trust is interpreted as a conditional judgement shaped by relational expectations and organisational context. Implications are discussed for the design of AI-generated communication systems in emotionally complex care domains.