AI-generated medical advice is rated favourably but GP advice is more trusted in India

Artificial intelligence (AI) is increasingly used to provide health information and advice, yet its acceptance depends not only on how credible the information appears but also on whether individuals are willing to rely on it. Existing research has documented resistance to medical AI and source-labelling effects, but evidence from non-Western settings and factorial designs examining source, transparency, and clinical risk together remains comparatively limited. This study examined how individuals evaluate medical advice presented as AI-generated versus general practitioner (GP)-generated, whether source-related differences vary across transparency and clinical-risk conditions, and how these evaluations relate to stated willingness to follow the advice.

We conducted a mixed-methods, partially within-subject 2 × 2 × 2 factorial experiment with 797 participants in Mumbai, India. Each participant evaluated four vignettes randomly drawn from eight conditions varying by source (AI or GP), transparency (high or low), and clinical risk (high or low), yielding 3,188 vignette observations. Following examination of the item structure, trustworthiness and a two-item uniqueness-recognition score (α = 0.944) were analysed as separate principal epistemically relevant outcomes.

Perceived accuracy and binary behavioural intention were examined separately. The originally constructed three-item epistemic-value composite (α = 0.15) was retained only as an exploratory supplementary analysis. Data were analysed using linear and logistic mixed-effects models, participant-level paired comparisons, and equivalence testing using operational ± 0.20-SD bounds.

Qualitative responses were analysed using inductive thematic analysis. Perceived recognition of individual uniqueness did not differ significantly between AI-generated and GP-generated advice, and equivalence was supported within the operational bounds. In contrast, GP-generated advice was rated as more trustworthy and more accurate than AI-generated advice and elicited greater stated willingness to follow.

Nevertheless, willingness to follow remained high for both sources, with 86.5% of AI-generated advice responses and 97.3% of GP-generated advice responses indicating an intention to follow the advice. Transparency and clinical risk did not significantly influence the continuous evaluative outcomes. Qualitative findings indicated that AI-generated advice was valued particularly for accessibility and informational clarity, whereas GP-generated advice was more strongly associated with professional expertise, accountability, safety, and relational care.

Favourable responses to AI-generated medical advice can coexist with a relative preference for advice attributed to a medical professional. Although AI- and GP-generated advice were comparable in perceived recognition of individual uniqueness within the operational equivalence bounds, GP-generated advice retained advantages in trustworthiness, perceived accuracy, and intended reliance. These findings suggest that responses to medical AI are multidimensional and support positioning AI as a complementary source of health information rather than as an interchangeable substitute for human medical expertise.