Generative AI Assisted Workflows in Architectural Conceptual Design: Performance, Creative Self-Efficacy, and Cognitive Load

Details

As design schools rush to integrate GenAI tools, this study pumps the brakes with actual data. Thirty-six architecture students completed a two-phase design task: first independently, then revising with either GenAI image generation or ArchDaily-based precedent search. Eight judges rated outcomes. The headline result is a null finding across the main outcomes: no significant difference in design performance, cognitive workload, or task-specific creative self-efficacy between the two workflows. This matters because advocates and critics of GenAI in education often project large effects that empirical study does not support. Three secondary findings deserve attention. General creative self-efficacy — students' broader confidence in their creative ability — showed a significant relative decline under GenAI. A subgroup analysis suggested novice students may perform better with GenAI than precedent search, but the paper warns this finding is exploratory and should not be over-interpreted. Iterative, task-specific prompting strategies were associated with lower cognitive load, though this did not survive multiple-comparison correction. The authors conclude that GenAI's educational value depends on pedagogical framing, learner characteristics, and prompt literacy — not the tool itself.