Designing For Generative Assistants: Onboarding UX for a Product-Design AI Tool

AI · 5 min read

Early users of the design tool encountered an onboarding that emphasized model parameters—creativity, temperature, output style—alongside many sample prompts. While power users appreciated the control, beginners were uncertain how to phrase goals or adapt outputs. Initial metrics showed many users rejected generated assets and returned to manual workflows.

The redesign reframed the assistant around designer tasks: “Create a 3-screen onboarding,” “Generate illustration set for fintech,” with task-specific templates and prefilled prompts. Each generated result included a short explainer: why the assistant made layout choices, which layers were editable, and a one-click conversion to native editable components. A sandbox mode let users tweak constraints without affecting production files.

After rollout, first-edit success (users keeping or lightly editing a generated asset) improved by 34% and frequency of rollbacks decreased by 21%. The team also implemented visible provenance and versioning to address ownership and quality questions. The lesson: when AI is a collaborator, onboarding must teach the workflow and afford editing—explainability and low-friction editability matter as much as raw generation quality.