When AI feels too human: The effects of avatar realism on user perceptions and decisions
Rayeed Khan · Aaltodoc (Aalto University) · 2026
This thesis examines the influence of different levels of AI avatar realism on the psychological perception and behavioural outcomes of users in a recommendation-based customer service setting. A quantitative between-subjects experiment has been conducted by assigning participants to randomly interact with one of three avatars that had varying levels of visual realism: a 2D cartoon avatar, a stylized 3D avatar, and a photorealistic human avatar. The study focused on key constructs such as trust, comfort, eeriness, perceived helpfulness, recommendation acceptance, and time to decision, and investigated their correlation with different avatar types. The moderating effects of digital skill on perceived eeriness were also tested. The data was collected from 186 valid respondents and analysed using statistical techniques such as a one-way ANOVA, chi-square tests, and a univariate general linear model. Overall, the results indicate that a higher level of AI avatar realism doesn’t necessarily improve the user experience in a customer service setting but rather may create discomfort and a feeling of eeriness that supports the Uncanny Valley perspective. The study aims to contribute to the growing research on human-AI interaction and offer practical insights for businesses that design AI avatars for customer service interfaces.