Generative AI-assisted design problem exploration in interaction design: response disclosure, prompt initiative, and cost-sensitive prompting behavior

Generative AI is increasingly used in early-stage design exploration, yet experimental evidence is limited on how system-side response disclosure and user-side prompt initiative jointly shape the exploration of design problems that are not yet well defined. In a 2 × 2 within-subject experiment, 48 design students completed four interaction-design tasks involving unfamiliar user groups under one-time versus progressive response disclosure and low versus high prompt initiative, with the composition time of every submission logged. Progressive disclosure was associated with greater perceived understanding change, higher self-reported engagement and germane cognitive load, lower extraneous cognitive load, and more prompts. High prompt initiative showed the same pattern on these four self-reported measures but significantly fewer prompts, reversing the predicted direction; composition time accounted for 18% of the exploration window under high initiative and under 2% under low initiative. Both sets of main effects held after multiplicity correction and adjustment for task and period. An exploratory interaction for germane cognitive load did not survive correction. These findings extend a mixed-initiative view of AI-assisted exploration by identifying response structuring and question formulation as separable design levers. They also indicate that prompt frequency should be interpreted together with the cost of producing a submission.