Building an Emotion‑Aware Onboarding: UX Case Study with an AI Coach

AI · 5 min read

The startup behind a mental fitness coach used multimodal AI to infer user emotional state from typed responses and optional voice samples during onboarding. Designers debated whether to show inferred emotions explicitly, hide them entirely, or use them only to adapt content. Given privacy concerns and regulatory sensitivity, the team ran a privacy-first experiment: keep inferences hidden by default but allow users to opt into seeing and editing labels.

Prototypes compared three flows: (1) adaptive-only (system uses emotion but hides inference), (2) transparent (shows emotion labels and rationale), and (3) editable (shows labels with an edit affordance). Moderated sessions with 30 participants revealed that editable labels created the best mix of perceived control and personalization — users felt heard and could correct misreads, which increased trust in the AI coach.

Post-launch analytics showed a 22% higher 14-day retention for users who engaged with editable labels and a 15% increase in content completion. The design decision to prioritize control and explainability balanced personalization with ethical constraints, demonstrating a pragmatic roadmap for startups combining AI with sensitive domains.