In recent years, there has been growing recognition of the threats to health posed by loneliness. One of the main strategies that has been recommended to address this is social prescribing (SP). This typically involves general practitioners (GPs) and other health practitioners directing clients who are experiencing loneliness and related conditions to take part in social activities—typically in recreational and community contexts. However, evidence for the effectiveness of SP is mixed—leading some to suggest that enthusiasm for it might be misplaced. In this review, we argue that a core problem with most existing approaches to SP is that they lack a strong theoretical base. This has been a barrier to (a) understanding when SP will work and why, (b) designing optimally effective SP programmes, and (c) developing practitioner skills and appropriate infrastructure to support them. As a corrective to this state of affairs, this review outlines a three-tier social identity framework for SP and five associated hypotheses. These hypotheses predict that SP will be more effective when (a) clients join groups and (b) these groups are ones with which they identify, and when SP is supported by (c) social-identity-enhancing social infrastructure, (d) a social-identity-based therapeutic alliance, and (e) identity leadership that builds and shapes this alliance as well as clients’ identification with prescribed groups. This framework is supported by a range of evidence and provides an agenda for much-needed future research and practice.