How does the visual complexity of architectural signage affect wayfinding search efficiency? Nonlinear evidence from a semantic-computational dual-channel experiment

Abstract Signage is the primary visual interface guiding people through complex public buildings; however, designers lack quantitative evidence on how its visual complexity shapes wayfinding performance, and clutter research largely assumes that simpler methods are always faster. This study examines whether this relationship is linear or nonmonotonic, developing the Architectural Visual Search Toolkit (AVST), a semantic-computational dual-channel framework that integrates multimodal large language model scoring with computational visual analysis to yield decomposable complexity metrics. Using AVST, 242 AI-generated signage stimuli were evaluated, and an online click-search experiment with 150 participants yielded 10,233 valid reaction-time trials analyzed via linear mixed models, analysis of variance, and bootstrap mediation. The results reveal a U-shaped relationship: medium-complexity signage produced the fastest search, whereas both low-complexity and high-complexity signage slowed it, challenging the prevailing linear assumption. Specifically, (1) color organization entropy predicted faster search; (2) mounting type moderated the complexity‒efficiency relationship; and (3) the figure‒ground index partially mediated the complexity‒reaction time pathway. These findings establish moderate complexity as an evidence-based target for wayfinding design, signage standards, and architectural planning policy.