Conversational AI for perinatal mental health: promise, limits, and a human-AI stepped-care framework

Perinatal mental health conditions are common yet persistently underdetected and undertreated. Recent advances in generative artificial intelligence (AI) and multimodal interaction offer new opportunities for scalable, continuous support, making the concept of the “digital doula” particularly compelling. Because perinatal care relies heavily on emotional support, repeated contact, and timely escalation, conversational AI is attractive as a potential adjunct to existing care. We argue that digital doulas should not be framed as autonomous substitutes for clinicians or human doulas, but rather conceptualized as AI-enabled relational interfaces embedded within a stepped-care model. We propose four core functions for digital doulas: companionship, symptom interpretation, navigation, and sentinel monitoring. These roles can reduce disclosure barriers, extend support between clinical visits, and strengthen linkage to appropriate care. However, conversational AI also introduces significant risks, including the illusion of empathy, crisis recognition failures, algorithmic bias against vulnerable populations, data privacy vulnerabilities, and unsafe deployment. To address these risks, we outline a safety-by-design agenda emphasizing human oversight, clear escalation protocols with medico-legal accountability, workflow integration, and equity-centered evaluation. Ultimately, the clinical value of digital doulas depends on safely strengthening the perinatal mental health care cascade without replacing its core human relationships.