This deck explores how a new AI Filter quick filter could work inside the existing hotel search filter bar of a B2B travel platform. Each concept is a separate page focused on one distinct interaction model. All concepts are intentionally low-fidelity, grayscale, and structural — the goal is to compare interaction logic, not visual polish.
Phase 1 rule: AI may only map natural language to filters that already exist in the system today. Anything else is either ignored or shown as "not applied".
AI takes over the bar; original pills hidden behind 'Back to standard filters'.
AI chips appear in a second row labeled 'From your request' below the untouched native bar.
AI-mapped pills sit inline with native pills with a tiny 'from AI' indicator.
Preview panel lists 'We found 2 supported filters' with Apply / Edit / Cancel.
One expandable summary pill: 'AI: Breakfast + Wifi + Parking'.
Input bar stays open and editable; chips render below for iterative refinement.
Diff view: Current vs. AI will add; 'added by AI' highlights fade after apply.
Click AI Filter → suggestion tray with 4 example prompts before a blank input.
Segmented control: Strict / Balanced / Flexible — same query maps differently.
Persistent slim banner above the bar: 'AI request active: …' with edit / clear actions.