JMIR Publications

ABSTRACT

Background:

Generative artificial intelligence (GenAI) chatbots are increasingly used for mental health support, but trust in these emotionally vulnerable and relationally sensitive interactions remains poorly understood.

Objective:

This study aims to synthesize empirical evidence on how trust is conceptualized, shaped, and associated with outcomes in GenAI-based mental health support, with attention to differences across AI roles.

Methods:

This systematic review was conducted in accordance with the PRISMA 2020 guidelines. Peer-reviewed empirical studies were identified by searching five electronic databases. Two reviewers independently screened records, selected eligible studies, extracted data, and assessed methodological quality using the Mixed Methods Appraisal Tool. Data were synthesized descriptively and thematically.

Results:

Of 1180 citations retrieved, 28 studies were included. Trust was rarely explicitly defined and was most often operationalized through affective and relational indicators, including emotional comfort, perceived empathy, and psychological safety, rather than technical competence alone. Trust-related antecedents involved user vulnerability and attitudes, system reliability and emotional responsiveness, interactional continuity, and contextual constraints. Trust was associated with engagement and emotional relief, but also with relational and safety concerns, including excessive reliance on AI support, blurred role boundaries, and inadequate responses to crisis-related disclosures. Role-based synthesis suggested that lower-engagement roles (eg, functional assistants and structured facilitators) mainly involved cognitive trust, whereas more emotionally engaging or autonomous roles (eg, empathetic co-therapists and therapeutic companions) involved broader affective and alliance-like trust, together with greater relational and safety risks.

Conclusions:

Trust in GenAI-based mental health support should not be treated as a uniform or inherently desirable outcome. Role-sensitive evaluation and governance are needed to align user trust with system capability, safety boundaries, and responsibility.