Redundant Entry Points Fix Low Discoverability

In a study of 6 Chinese users (ages 18–45), 5 of 6 defaulted to traditional delivery apps like Meituan for milk tea orders, expecting GUI flows over chat prompts—even frequent genAI users saw chatbots as Q&A tools, not transactional. Qwen countered this with dual activation: chat requests or a prominent 'Qwen's Treat' button opening categorized task menus that prefill prompts. This redundancy helped novices, but broad categories like 'Milk Tea & Catering' auto-prompted wrong items (e.g., noodles instead of milk tea), causing confusion. Fix: For narrow scopes like movie tickets, prefilling works; for broad ones, prompt clarifying questions first (e.g., 'What drink?') with options to avoid mismatches and extra steps.

Mirror Familiar Patterns to Cut Learning Friction

Users regained confidence when Qwen transitioned chat to delivery-app-like interfaces: category navigation, item lists, product pages matching Meituan's layout. One user said, 'This looks familiar,' easing mechanics focus. But carousels for shop recommendations hid total options (e.g., showing one at a time without counts), leading users to assume limited choices and switch apps. Use grids, lists, or counts (e.g., '12 shops nearby') for larger sets; proven patterns like carousels suit small subsets only, preventing perceived scarcity.

Transparent Data Access Prevents Privacy Alarms

Users distrusted aggressive promotions (e.g., free milk tea) and freaked when Qwen auto-surfaced full Taobao-saved addresses post-authorization—before item selection—due to buried gray-text explanations. Perceptions of broad data leaks (e.g., shopping history) drove 2 participants to abandon for traditional apps. Mitigate with: (1) active acknowledgment on auth screens stating exact data accessed; (2) contextual explanations before display (e.g., 'Using ZIP for nearby shops'); (3) minimal data reveal (ZIP during browse, full address at checkout only). This matches minimal-needed principle, rebuilding trust.

Full Transparency Preserves Autonomy and Decisions

Qwen let users edit mid-flow, review orders, and pay via AliPay authentication, retaining control. But hidden fees jumped milk tea from ¥1.6 to ¥10.2 (adding delivery/packaging, unmet ¥16 minimum), misleading users to compare elsewhere. Flight options lacked baggage info, forcing app switches despite narrowed choices. Always surface decision-critical details (fees, thresholds, allowances) upfront—efficiency without info erodes trust; informed control fosters repeat use.