Routine Generative AI Use and Acceptance of Medical AI: A Cross-Sectional Survey of Healthcare Workers and the Public in Japan.

Background Acceptance of medical artificial intelligence (AI) is critical for successful implementation, yet its acceptance may vary by clinical application and stakeholder group. Whether routine use of generative AI (GAI) is associated with acceptance of medical AI remains unclear. Methods A cross-sectional web survey was conducted in Japan in November 2025 using web-panel convenience sampling from an online research panel. The small analytic sample comprised 200 participants aged 20-69 years, including 100 healthcare workers and 100 non-healthcare workers. GAI use was assessed using two five-point items (GAI use at work and in daily life) and dichotomized as at least monthly use in either daily life or work versus less than monthly use in both settings, based on prespecified criteria. Acceptance of medical AI was measured for five scenario-based applications - AI-assisted imaging interpretation, AI-based health risk prediction, AI-based treatment recommendations, AI-enabled triage guidance, and AI-assisted robotic surgery - using a four-point acceptability scale; responses were dichotomized as acceptable versus not acceptable. Calibration weights were constructed to approximate Japanese internet users by raking on sex-by-age group and household income, normalized to the analytic sample size, and truncated at prespecified bounds. Adjusted prevalence ratios (aPRs) were estimated using modified Poisson regression with robust standard errors, stratified by occupation; covariates were selected a priori using a modified disjunctive cause criterion. This study was exploratory; therefore, adjustments for multiple comparisons were not performed. Results Among healthcare workers, the lower-use group had a higher prevalence of non-acceptance of AI-based health risk prediction (aPR 6.980, 95% CI 1.412-34.506) and AI-based treatment recommendations (aPR 4.364, 95% CI 1.331-14.305), whereas estimates for imaging interpretation, triage guidance, and robotic surgery remained statistically uncertain. Among non-healthcare workers, the lower-use group had a higher prevalence of non-acceptance of AI-assisted imaging interpretation (aPR 13.906, 95% CI 1.624-119.052), AI-based health risk prediction (aPR 4.861, 95% CI 1.133-20.862), AI-based treatment recommendations (aPR 6.288, 95% CI 1.386-28.530), and AI-assisted robotic surgery (aPR 6.173, 95% CI 1.435-26.561); the adjusted estimate for AI-enabled triage guidance was not statistically supported (aPR 1.840, 95% CI 0.521-6.505). Several estimates had wide confidence intervals, indicating substantial imprecision. Conclusions In this small exploratory web-panel convenience sample, lower-frequency GAI use was associated with higher non-acceptance of several medical AI applications. Associations were observed across more applications among non-healthcare workers than among healthcare workers. Because the study was cross-sectional, used convenience sampling, and included imprecise estimates with wide confidence intervals, the findings should not be interpreted causally. Larger studies using longitudinal or interventional designs are needed to clarify temporality and mechanisms.