Under the global trend of population ageing, social isolation and loneliness have emerged as public health issues that can no longer be neglected, and their adverse effects on health are comparable to those of established risk factors such as smoking and obesity, with a particularly profound impact on older populations. As one of the core dimensions of holistic health, social health is interwoven with physical, mental and spiritual health, and together with them determines an individual's overall state of well-being. As community care systems have matured and digital technology has advanced, social-health interventions have expanded from traditional volunteer visits and isolated community activities into a diversified range of models, including social prescribing, health communities, senior co-housing and AI-based virtual companionship. These models offer new solutions for loneliness prevention and social support in an ageing society, while at the same time giving rise to ethical challenges such as algorithmic bias, the digital divide and privacy risk. The present study was designed to examine systematically the theoretical foundations of social health, the development of the service industry and the application of digital technology, and to construct an age-friendly, integrated AI community-service architecture. First, the mechanisms linking social connectedness, belonging and health were reviewed, the behavioural, physiological and psychological pathways through which social support influences health were analysed, and large-scale epidemiological evidence was used to describe the concrete effects of loneliness and social isolation on mortality, cardiovascular disease, cognitive function and health-care utilisation. Second, the operating logic and practical value of industry models including social prescribing, community care, health communities and co-housing were examined, together with the institutional challenges and success factors of cross-sector collaboration and the practical strategies of Online-Merge-Offline (OMO) community operation. Third, the applied value of Social Network Analysis (SNA) in risk detection and community assessment, the benefit boundary of AI virtual companions and social robots, and ethical governance issues concerning algorithmic bias, privacy risk and the protection of vulnerable groups were discussed in depth. Finally, five design principles, a five-stage integrated service process and five diversified business models for age-friendly AI community services are proposed, and a cross-domain ecosystem in which health care, social welfare, industry, government and academia collaborate is constructed.