AI-Driven Personalization vs. Privacy: Design Trade-offs at a Mental Health App

AI · 5 min read

The product team built an AI model that recommended exercises, nudges, and session timing based on in-app behavior and passive signals. Early lab testing showed engagement gains, but interviews flagged anxiety around opaque data use—users wanted to know what the model saw and how recommendations were generated.

Design responded with a two-part solution: transparency-first UI that explains the model in plain language and a granular consent center where users can toggle which signals the model may use. The team introduced 'why this suggestion' cards that surface the top two features influencing a recommendation without exposing sensitive raw data.

A pilot with 1,200 users showed that explicit transparency reduced opt-outs and increased trust scores; users who enabled more signals experienced higher engagement but only when they had clear control. The product team adopted a default privacy-maximizing configuration and designed onboarding to invite progressive disclosure of personalization features.

This case highlights a repeatable pattern for AI features in startups: ship conservatively, make agency and explanations core UX elements, and measure trust outcomes alongside engagement.