The integration of Large Language Models (LLMs) with Conversational User Interfaces (CUIs) has significantly transformed health information seeking, offering interactive access to health resources. Despite the importance of trust in adopting health advice, the impact of user interfaces on trust perception in LLM-provided information remains unclear. Our mixed-methods study investigated how different CUIs (text-based, speech-based, and embodied) influence trust when using an identical LLM source. Key findings include (a) higher trust levels in information delivered via textbased interface compared to others; (b) a significant correlation between trust in the interface and the information provided; (c) participant's prior experience, processing approach for information with different modalities and presentation styles, and usability level were key determinants of trust in health-related information. Our study sheds light on trust perceptions in health information from LLMs and its dissemination, underscoring the importance of user interface in trustworthy and effective health information seeking with LLM-powered CUIs.

, , ,
doi.org/10.1145/3613905.3650837
2024 CHI Conference on Human Factors in Computing Sytems, CHI EA 2024
Centrum Wiskunde & Informatica, Amsterdam (CWI), The Netherlands

Sun, X., Liu, Y., de Wit, J., Bosch, J., & Li, Z. (2024). Trust by interface: How different user interfaces shape human trust in health information from Large Language Models. In Proceedings of the Conference on Human Factors in Computing Systems (pp. 344:1–344:7). doi:10.1145/3613905.3650837