AI Journaling and Companion Apps for Mental Health: Mechanisms, Outcomes, and Safe Design Principles

Abstract

AI-powered journaling and companion apps reach millions daily, yet most lack regulatory oversight and long-term safety data. This paper synthesizes efficacy and harm evidence from randomized controlled trials (RCTs) and observational studies, maps mechanistic pathways, and proposes a five-principle safe design framework, covering reality anchoring, use moderation, vulnerability detection, privacy by design, and clinician integration. Short-term benefits are real but modest; harm concentrates in heavy users and vulnerable subgroups. Short-term benefits are documented across both categories, though empirical evidence for AI journaling applications remains limited relative to companion app research, and long-term outcomes for both remain insufficiently characterized. Regulatory implications for the FDA digital health framework and the EU AI Act are discussed.