MyListener is an Android app that generates AI diary prompts from contextual data — step count and sleep from a Fitbit Luxe smartband, plus screen time and app usage from the phone — then replies to what users write with a GPT-4-generated message of empathy and self-disclosure, designed to build intimacy with the AI agent and encourage further reflection.

In a two-week study with 11 university students who frequently experience depression and loneliness, mean PHQ-9 scores dropped from 8.27 to 6.09 and UCLA-3 loneliness scores from 50.18 to 47.27, both statistically significant.

As second author, conducted user interviews and derived design implications on human-AI intimacy for a context-aware journaling app. Interviews surfaced AI self-disclosure as a "double-edged sword" — a consistent tone built trust and a sense of being understood, but an inconsistent tone or a misread diary entry could instead deepen the very depression and loneliness the app aimed to relieve.

Published at ACM UbiComp'24.