Ash AI boosts mental health functioning in 1,284-user study

Researchers from Columbia University and other institutions evaluated Ash, a conversational AI designed for mental health support, in a single-arm observational cohort study with 1,284 new users. Participants completed in-app measures of psychological functioning (life satisfaction, relationship satisfaction, sleep quality, behavioral activation), working alliance, and grandiosity (inflated self-perception) at baseline and after four weeks. At baseline, users reported below-average life satisfaction and fair sleep quality. After four weeks, within-person improvements were significant across all functioning indicators and working alliance (all p<.001, Cohen's d ranging from 0.14 to 0.26), while grandiosity showed no significant change, alleviating concerns about inflated self-perception often raised for AI-based therapy tools.

Engagement metrics were strongly predictive of outcomes. Active days, total sessions, and total minutes consistently predicted Week 4 psychological functioning and working alliance (p≤.006, partial R² range: 0.58-2.15%, controlling for baseline). However, user message volume did not correlate with improvements, suggesting that consistent engagement over time matters more than sheer number of messages. These findings extend previous mental health AI research beyond symptom reduction to broader day-to-day functioning, offering preliminary evidence that purpose-built conversational AI can support psychological well-being at scale without adverse effects on self-perception.