Effects of Visual Interface Elements within Generative AI Multimodal Systems on User Experience and Satisfaction

This study investigates how interface modality in multimodal generative AI systems shapes user experience and satisfaction.Although systems such as Open AI's GPT-4o increasingly support multimodal interaction, research has emphasized technical performance more than user-centered interface design.Addressing this gap, we examine how modality affects social presence, perceived enjoyment, concentration, preference, and satisfaction.A betweensubjects experimental survey used imagined-interaction stimuli for three conditions: (1) GPT-4o's default text interface, (2) GPT-4o's native audio interface, and (3) a visually enhanced interface in which participants selected one of the five MetaHuman characters, providing simplified, non-embodied visual cues to manipulate visual presence.A total 922 participants were randomly assigned to conditions.Established scales mesasured all constructs.Data were analyzed using ANOVA and structural equation modeling (SEM).ANOVA revealed significant modality-based differences in social presence and perceived enjoyment, but not in concentration.SEM clarified this pattern: higher social presence significantly increased both concentration and enjoyment.Concentration strongly predicted enjoyment, which increased preference and, in turn, satisfaction.Notably, the text-based interface produced higher enjoyment than the MetaHuman-based visual condition, suggesting that simplified or nonrealisitc visual cues do not necessarily improve user experience.This is among the first empirical studies to test how visual interface elements in multimodal generative AI shape user experience via presence.