Human factors in generative AI companions: a systematic literature review of trust, relational dependence, psychosocial outcomes, and responsible adoption

Abstract

Context Generative AI companions and socially oriented chatbots are systems designed or used to sustain social, affective, or relational interaction. They increasingly function as persistent interaction partners across systems ranging from earlier rule-based, retrieval-based, and pre-LLM architectures to contemporary generative AI companions. Prior reviews and conceptual models have largely examined user-facing relational mechanisms or specific ethical and legal issues. Yet, how these mechanisms connect with platform, organizational, and regulatory conditions remains fragmented. Objectives This study addresses this gap by synthesizing human-factor and responsible-adoption concerns across technology generations and developing an evidence-calibrated multi-stakeholder framework spanning human, organizational, technological, and regulatory layers. Methods We conducted a PRISMA-guided systematic literature review of records retrieved from Scopus, ACM Guide to Computing Literature via the ACM Digital Library, Web of Science Core Collection + MEDLINE, and IEEE Xplore. We examined English-language, peer-reviewed publications made available between 2005 and 2024. From 1699 identified records, 42 publications were included and analyzed using thematic synthesis. Results Four analytical themes were identified: usability and relational affordances in interaction; trust, attachment, over-reliance, and psychosocial outcomes; responsible adoption, platform design, and organizational mediation; and governance, privacy, and accountability. Technology emerged as a cross-cutting mediator linking design features, user experience, organizational practice, and governance concerns. The first two themes drew on 29 and 33 materially contributing publications, whereas the latter themes each drew on 12 and represent emerging synthesis areas. Conclusion The review shows that responsible adoption of AI companion systems requires attention not only to user experience, but also to platform incentives, design affordances, data practices, service continuity, and governance mechanisms. The framework's distinctive contribution is to connect these concerns across stakeholder layers while preserving differences in evidentiary maturity. It is a provisional analytical heuristic, not a validated causal model.