Purpose: This article, rooted in the Uncanny Valley Theory, investigates how human-like similarity in AI-based avatars influences individuals' emotional responses, highlighting the mediating role of perceived anthropomorphism and creepiness, and the moderating effect of users' digital literacy in shaping their reactions during the service delivery process.

Design/methodology/approach: Through a quantitative research design, in Study 1, we prepared and pre-tested the fictitious avatars to be used as stimuli in the main study. In Study 2, that is the Main study, a two-cell experimental design was conducted on a sample of 206 individuals interacting with fictitious avatars exhibiting low versus high levels of human-like similarity investigating, by using validated scales, users' perceived anthropomorphism, creepiness and their emotional responses, together with their self-reported digital literacy, in a simulated healthcare service delivery context.

Findings: Results explain some dark and bright sides of the AI, by revealing that AI-based avatars with high human-like similarity (vs. lower) trigger stronger negative emotional responses, mediated by perceived anthropomorphism and creepiness. Notably, individuals with higher digital literacy experience significantly attenuated negative emotions, suggesting a buffering effect. This confirms the relevance of digital literacy as a cultural and psychological asset in human-avatar interaction.

Research limitations/implications: The study was conducted in a simulated healthcare context and used self-reported digital literacy. Future research could explore different service settings, physiological or behavioral data and expand the model to other personality traits or cultural contexts.

Practical implications: The findings shed light on insights for marketers and managers by illustrating the emotional risks associated with excessive anthropomorphism in AI-driven avatars. Managerial strategies leading to service design should focus not only on the calibrated design of avatar features but also on enhancing users' digital literacy, which plays a crucial role in moderating negative emotional responses. This dual approach can optimize user experience, foster trust and improve service acceptance, particularly in digital health contexts.

Originality/value: This study advances the literature on the Uncanny Valley by identifying the emotional components tied to anthropomorphic avatar design, by using as a setting the healthcare services. Importantly, it introduces digital literacy as a novel moderating variable, establishing a theoretical link between AI design, user affective response and technological competence. By integrating perspectives from emotional psychology and service innovation management, the research may contribute to a better understanding of user-technology interaction.

© Emerald Publishing Limited