["International Journal of Art &Design Education, EarlyView. ", "\nAbstract\nEarly‐stage design students frequently experience representational friction during conceptual ideation, where limited sketch fluency constrains the externalisation of spatial concepts and weakens creative confidence. This study examines whether a topology‐preserving generative AI framework can function as a pedagogical cognitive scaffold to reduce these barriers without displacing student authorship. We implemented a zero‐shot, structure‐guided AI system (based on ControlNet) that converts hand‐drawn sketches into structurally mediated visual scaffolds while preserving original geometric topology. In a controlled experiment (n=80$$ n=80 $$), this AI‐assisted workflow was compared with static moodboard‐based ideation. To isolate design logic from surface rendering advantages, blind experts evaluated only final pure line‐art outcomes. Results indicated that the AI‐assisted group was associated with higher overall design quality (d = 0.88), with improvements concentrated in structural coherence, functionality, and completeness rather than aesthetic quality. Behavioural traces and retrospective protocols further indicated reduced conceptual abandonment and increased structurally focused iteration. These findings suggest that topology‐preserving generative AI may function as a selective scaffold for spatial reasoning and cognitive offloading in foundational design education, bridging conceptual intent and executable form while preserving learner creative control. Rather than automating creativity, constrained generative systems may serve as potential instructional partners for structure‐centred student–AI co‐creation.\n"]