Beyond User Illusions: An Analysis of the Relationship Between Subjective Satisfaction and Objective Performance in Voice Agent Interaction

Abstract

This study investigates how voice agent (VA) response styles and task cognitive levels affect users’ information processing performance and satisfaction in voice user interface environments. Prior research focused primarily on response length or naturalness, leaving a gap in how information structuring influences objective performance across task difficulty levels. We conducted a 3×3 Latin Square experiment with 30 participants, manipulating response style (Narrative, Enumerative, Categorical) and cognitive level (Recall, Understand, Analyze). Using ART-ANOVA and LMM as a robustness check, we found that the Style × Cognitive Level interaction was significant for performance in both analyses (ART: p=.020; LMM: p=.026) but non-significant for satisfaction. This asymmetry suggests a performance–satisfaction discrepancy that we refer to as “User Illusion,” further supported by a significant Style × Cog interaction on the normalized Gap (ART: p=.035; LMM: borderline). These findings indicate that VA response design must dynamically adapt to task cognitive level, integrating both objective and subjective evaluation metrics.