Designing an AI Assistant’s Prompt UI: From Modal Prompts to Inline Workflows

AI · 5 min read

The initial design exposed users to a single modal prompt box for all queries. While powerful, it required users to translate analytic goals into free-form prompts, which led to inconsistent outcomes and frustration. Designers observed users reverting to support docs or exporting data to manipulate offline. To lower the skill floor, the team created inline prompt builders that surfaced templates, prefilled filters, and suggested clarifying questions based on the dataset.

The redesign included three pattern shifts: (1) context-aware templates that adapt to the current dataset and selection, (2) progressive clarification prompts that ask one targeted question at a time, and (3) an editor that showed both the natural-language transformation and the equivalent SQL preview for transparency. These elements were tested with 500 analytics users in a staged rollout.

Results were notable: task completion for common analytics jobs rose 41%, time-to-insight dropped 27%, and the number of user-generated free-text prompts declined — indicating users relied on structured workflows more. Power users still valued free-form prompts, so the team retained a modal expert mode but surfaced the inline builder as the default.

This case demonstrates that AI UIs benefit from scaffolding: guiding users toward intent specification and making model outputs inspectable reduces errors and builds trust. For analytics products, coupling natural-language interfaces with explicit transformations (like SQL previews) helps bridge expertise gaps while keeping the assistant approachable.