SketchifAI study: Sketching boosts AI ideation fluency, but users prefer text prompts

A new study from the University of Melbourne tackles a critical tension in AI-assisted design: as text-based AI tools (like ChatGPT) streamline ideation, they risk eroding the reflective sketching practices central to designerly thinking. The researchers built SketchifAI, a prototype AI creativity support tool (AI-CST) that accepts multimodal inputs: text only, sketch only, and sketch plus tags. In a within-participants study with design students, they measured how each modality affected perceived expression ability, creativity support, and divergent thinking (fluency, variety, originality, quality).

Preliminary results show that sketching—especially sketch-only—tended to enhance fluency, meaning participants generated more ideas when they could draw. However, there was inconclusive evidence that sketch modalities improved variety, originality, or quality over pure text prompts. More strikingly, despite the fluency boost from sketching, participants overwhelmingly preferred text prompting. This paradox suggests that designers find text faster or more comfortable, even when sketching yields more ideas.

The study, presented at ACM Creativity and Cognition 2026, argues that current AI tools optimize for quick solution generation at the expense of reflection-in-action—the iterative, back-and-forth process designers use to refine ideas. The authors call for AI-CSTs that deliberately introduce “productive friction” to encourage sketching, such as requiring a rough sketch before accepting a text prompt. Without such interventions, essential design skills may be lost as students gravitate toward the path of least resistance (text). The implication: AI tools for creative work must be designed to support, not replace, the messy, visual thinking that drives innovation.