Generative AI in healthcare: Effect of explanations and the role of AI on trust, perceived privacy, and intent to use.

Abstract

Background The use of artificial intelligence (AI), such as ChatGPT, in healthcare has grown rapidly, yet public perceptions of trust, privacy, and intent to use AI for medical purposes remain underexplored. This study aimed to investigate how ChatGPT’s role (diagnostic tool vs. explanation assistant) and the level of explanation provided (why, confidence, or none) influence these key factors. Methods A within-subjects experimental design was employed using a Wizard-of-Oz methodology to systematically manipulate ChatGPT’s role (diagnostic tool vs. explanation assistant) and level of explanation (none, confidence, why). Ninety-eight UK-based participants recruited from the general public evaluated hypothetical healthcare scenarios via an online survey. Dependent variables (trust, perceived privacy, and intent to use) were measured using validated scales with satisfactory internal consistency in the present study. Data were analysed using repeated-measures twoway ANOVA to test the main and interaction effects of AI role and explanation level. Results Trust (F(1, 97)=12.40, p < 0.001) and intent to use (F(1, 97)=14.61, p < 0.001) were significantly higher when ChatGPT acted as an explanation assistant compared to a diagnostic tool. Perceived privacy was highest for “confidence” explanations overall (F(1.84, 178.85)=4.22, p = 0.019) among all types of explanations. The interaction between AI role and explanation level was significant for all outcomes (p < 0.001). Among all conditions, a “why” explanation yielded the highest scores when ChatGPT was an explanation assistant for trust, perceived privacy, and intent to use. Interestingly, when ChatGPT was a diagnostic tool, trust, perceived privacy, and intent to use were higher when ChatGPT provided no explanation than when it provided “why” explanation. Conclusions This study highlights the importance of context in AI design for healthcare. ChatGPT was trusted and preferred more as an explanation assistant, especially when providing “why” or “confidence” explanations. The findings suggest that AI integration into healthcare should emphasise transparency and context-sensitive roles to build public trust and optimise usability.