Maternal health chatbot fails to improve anxiety and depression

A generative AI chatbot did not improve anxiety or depression at 12 months compared with an active control group in a University of Maryland-led trial involving 400 maternal patients. Participants were pregnant or postpartum and from racial minority backgrounds, populations that often face inequities in healthcare access and outcomes. The findings suggest chatbot support alone may be insufficient to improve perinatal mental health.

The experimental group received access to Rosie, while the control group received a free children’s book once a month for a year and access to study staff as needed. Rosie was trained using community-engagement methods, feedback from diverse pregnant and postpartum populations, and evidence-based mental health knowledge. It could flag possible self-harm, severe symptoms and other distress for real-time review and follow-up by a clinical social worker.

More frequent Rosie use was associated with higher reported anxiety but better sleep. Whether anxiety prompted greater use or chatbot interactions amplified anxiety remained unclear. Benefits also appeared to vary with social support and parenting stage, though subgroup findings require cautious interpretation and further study. Chatbots may be better suited to supplementing care than replacing interventions that address caregiving burdens, financial strain, social support and access to clinical treatment.