Dataset for: Acceptance of Health-Apps with and without AI – An experimental test of an extended TAM

Background: Health apps with and without AI offer potential to improve patient care but face limited adoption. This study examined factors influencing health apps acceptance by extending the Technology Acceptance Model (TAM) with diagnostic source, rating valence, and trust. Method: A preregistered 3 × 2 between-subjects experiment (N=824) manipulated diagnostic source (dermatologist, AI, personalized AI) and rating valence (positive vs. negative) using a realistic app interface. Participants evaluated behavioral intention to use (BI), perceived usefulness (PU), perceived ease of use (PEOU), trust and willingness to pay (WTP). Results: MANCOVA results indicated a non-significant multivariate effect of diagnostic source (p = .079, Ω² = .004), whereas rating valence had robust effects on PU, PEOU, trust and BI (p < .001, Ω² = .13). No significant interactions between diagnostic source and rating valence emerged (p = .222, Ω² = .002). Mediation analyses showed that the effect of rating valence on BI was fully mediated via PU and trust, and partially via PEOU (all p< .001). The effect of diagnostic source on BI was fully mediated through trust (p= .05). Lasso regression identified trust as the strongest predictor of BI, followed by PU. None of the variables reliably predicted WTP. Conclusion: The study identifies positive ratings as a key driver of trust and PU in health apps. Interestingly, no direct effect of diagnostic source emerged, contrasting with prior research on algorithm aversion. This may be due to the implicit presentation of diagnostic source within a realistic app context without explicit comparison or choice.