Study: AI designers oscillate between vague and specific language
Expert UI/UX designers move from vague intentions to specifics and back again when using generative AI.
A new preprint from Daisaku Sato, submitted to the International Journal of Human-Computer Interaction, examines how expert designers form intent when working with generative AI tools. Five UI/UX designers with 11-20 years of experience (mean 15.4) designed hero sections for landing pages using a generative AI design tool. Through think-aloud protocols and retrospective interviews, the researchers used reflexive thematic analysis with a sensitizing lens of lexical granularity: L1 (vibe), L2 (design-domain language), and L3 (operational language). The study challenges the assumption that designers move unidirectionally from vague prompts to precise code.
Instead, designers exhibited what Sato calls "lexical oscillation" — they repeatedly returned from operational specificity back to ambiguity. Mismatches with AI outputs were not seen as failures but as opportunities for designers to reconsider what they meant, shifting engagement from instruction to consultation. One non-oscillating trajectory suggested that conceptual misalignment may limit design exploration. The paper positions AI as a non-neutral generative interlocutor that makes ambiguity a creative resource, implying that design tools should support iterative back-and-forth rather than fixed one-shot prompting.
- Five expert UI/UX designers with 11-20 years experience showed lexical oscillation between three levels: L1 vibe, L2 design-domain, L3 operational language.
- Designers did not move linearly from ambiguity to specificity; they cycled back to ambiguous language after giving precise instructions.
- Mismatched AI outputs prompted designers to reconsider their original intent, turning instruction into a consultation-like process.
Why It Matters
Suggests AI design tools should support ambiguity and iterative refinement rather than one-shot prompts.