A generative AI framework for preference-driven indoor visualization: an exploratory questionnaire-based approach

Abstract

Abstract Although residential spatial qualities are consistently recognized as a determinant of occupant spatial preference and perceptual experience, early-stage practice lacks systematic, user-centered mechanisms for capturing and translating individual preferences into design concepts. This study aims to examine whether a structured, questionnaire-driven generative AI framework can produce preference-aligned residential interior visualizations, and which spatial preference dimensions are most reliably translated. To address this aim, a 15-item multiple-choice questionnaire, grounded in eight environmental psychology and architectural theory frameworks, maps user responses through a predefined parameter layer into structured prompts, guiding Google Gemini's integrated generative pipeline to produce individualized interior visualizations across three residential typologies: living rooms, bedrooms, and home offices. Evaluation by 122 participants across 25 countries employed four perceptual constructs — Atmospheric Coherence, Emotional Comfort, Overall Satisfaction, and Expectations Alignment — operationalized strictly as indicators of perceived design suitability and preference resonance, not as measures of psychological or mental health outcomes. All four constructs exceeded the neutral midpoint at a Bonferroni-corrected threshold with large effect sizes (M = 4.03–4.61; d = 1.20–2.52), and internal consistency of the multi-item constructs was acceptable to good (α = 0.72–0.84). The theoretically predicted construct hierarchy — with atmospheric parameters translating most reliably and identity-expressive parameters least — was preserved across all typologies and evaluative profiles, establishing preliminary feasibility of the framework as a preference-elicitation and visualization instrument for early-stage design.