More AI, Less Care-Seeking? A National Survey Experiment on the Impact of AI Intensity on Patient Care-Seeking Intention in Chinese Family Doctor Services

Background: Artificial intelligence (AI) is increasingly embedded in routine primary care, yet how the levels of integration might affect its acceptability is unknown, especially in relationship-based service models where patients expect visible human stewardship. Prior experimental studies often treat AI adoption as a binary condition, leaving the “intensity gradient” of automation and the role of model specialization under-explored. We examine whether increasing AI integration in the clinical encounter erodes patients’ intention to seek care from family doctors in China, and whether labeling the AI as a medical-specific model buffers such erosion. Methods: We conducted a nationwide online survey experiment in China (N = 2790). Participants were randomly assigned to vignettes that varied by (i) the level of AI integration (low, medium, high) and (ii) the AI type (general-purpose vs. medical-specific large language model), with a human-only care scenario as a reference. Care-seeking intention from family doctors was assessed immediately after exposure. We estimated treatment effects using OLS regression with heteroskedasticity-robust standard errors, and examined the buffering hypothesis through an interaction term between AI integration intensity and AI type. Results: Care-seeking intention declined steadily as AI integration increased (p < 0.001), with the sharpest drop under high-intensity AI integration where clinical decisions were delegated to the AI system. Across all intensity levels, framing the system as a medical-specific AI consistently resulted in higher care-seeking intention than a general-purpose model. However, the interaction between AI intensity and the AI type was not statistically significant (p = 0.508). Conclusions: Patient acceptance of AI in primary care depends not only on whether AI is involved, but on how deeply AI is positioned in the encounter. Medical-specific AI labeling may enhance acceptance across all AI integration levels. The findings underscore the need to preserve human clinical agency in AI-embedded primary care. The results contribute to research on healthcare systems, digital health, and AI–patient interaction.