AI-Based Social Connection Interventions for Older Adults: A Scoping Review Identifying Core Categories of Relational, Emotional, Functional, and Risk-Detection Support

Background and Purpose: Social connection is a critical determinant of well-being in later life, yet older adults face relational deficits due to reduced social interaction. In response, artificial intelligence (AI)-based interventions have emerged as scalable approaches to support social connection, and a growing body of research has examined their potential to address the mental health challenges faced by older adults. However, despite this expanding evidence base, little attention has been paid to identifying the cross-cutting categories of support that shape social connection across diverse AI technologies and care contexts. This study addresses this gap by developing a typology of core support categories and mapping how these categories are configured across AI-based interventions to shape social connection among older adults, drawing on a scoping review of empirical studies.

Methods: We searched five major databases (PubMed, MEDLINE, CINAHL, PsycINFO, and Web of Science) for peer-reviewed studies published between 2016 and 2025. Studies were included if they examined AI-driven interventions designed to support social connection among older adults, conceptualized as a multidimensional construct encompassing emotional, relational, and behavioral dimensions (e.g., loneliness reduction, social interaction,and behavioral engagement). A total of 50 studies met the inclusion criteria. Data were analyzed using an inductive, theory-informed approach that combined descriptive mapping and thematic analysis to identify recurring patterns in how AI systems engage with users. Through iterative cross-study comparison, core support categories and their configurations were identified and synthesized.

Results: Four core support categories—relational interaction (forming ties), emotional regulation (enhancing interaction quality), functional support (sustaining engagement), and risk detection (reducing constraints)—were identified. Together, these categories shape social connection through distinct but interrelated processes. Relational interaction and emotional regulation were present across all studies, jointly supporting the initiation and quality of social interaction and serving as the foundation for sustained social connection. Functional support and risk detection were observed in 72% and 16% of studies, respectively. Across the included studies, interventions did not rely on a single category in isolation; instead, categories operated in combination, highlighting the importance of their configuration in shaping social connection outcomes. The most prevalent configuration combined relational interaction, emotional regulation, and functional support, forming a companionship-plus-routine pathway that sustained engagement in everyday contexts. Functional support enabled sustained participation in daily routines and activities, thereby reinforcing opportunities for ongoing social interaction. In contrast, risk detection, while less prevalent, functioned as a constraint-reducing condition by lowering barriers to participation and supporting continuity of engagement.

Conclusion and Implications: This review advances a conceptual framework of core support categories underlying AI-based interventions, shifting attention from specific technologies to the interactional configurations that enable AI to support social connection across contexts. The findings suggest that social connection outcomes vary depending on how core support categories are configured within interventions. This typology provides a transferable framework for designing AI-based interventions that are responsive to the multidimensional nature of social connection.