Frontiers in Public Health · Oct 30, 2026
Companionship-oriented conversational AI use and loneliness among community-dwelling older adults: the moderating role of human social connectedness
By Lifeng Chen, Xiucai Yang, Hong Chang
Background: Conversational artificial intelligence (AI) is increasingly used for companionship, creating new questions about loneliness and social relationships in later life. We examined the association between AI companionship use and loneliness, its variation across levels of human social connectedness, and emotional dependence among users. Methods: A cross-sectional questionnaire survey conducted in Guangzhou between March and May 2026 included a nonprobability, complete-case sample of 427 community-dwelling adults aged 60 years or older. Loneliness was measured using the three-item UCLA scale and human connectedness using the six-item Lubben Social Network Scale (LSNS-6). Linear regressions used HC3 covariance estimates and adjusted for demographic, household, health and digital circumstances. An interaction model assessed variation across connectedness levels, and a separate analysis examined study-specific emotional dependence among 168 AI companionship users. Results: Of the 427 respondents, 168 (39.3%) used AI for companionship. At mean connectedness, the adjusted loneliness difference between users and nonusers was −1.175 points (95% CI −1.459 to − 0.891). The AI-by-connectedness coefficient was 0.587 per five LSNS-6 points (95% CI 0.346 to 0.827). Adjusted differences were −1.610, −1.140 and −0.671 points at the lower quartile, median and upper quartile of connectedness, respectively. Among users, each dependence point was associated with 0.752 additional loneliness points (95% CI 0.493 to 1.010), adjusting for frequency, connectedness and contextual covariates. Conclusion: AI companionship use was associated with lower loneliness, particularly among respondents with weaker human connectedness. Greater emotional dependence was associated with higher loneliness among users. These concurrent associations support assessing digital companionship alongside human network resources and emotional reliance, while prospective studies are needed to establish temporal relationships and evaluate intervention benefits or harms.