The Evaluation of AI Companions Relative to Family, Friends, and Mental Health Professionals Among Young Adults: The Role of Social Support and AI Beliefs

Generative AI is increasingly used for emotional support, but little empirical work has examined how young adults evaluate AI companions relative to human sources of support. This study examined how college students comparatively evaluate AI companions relative to family, friends, and mental-health professionals, and whether perceived human support and beliefs about AI are associated with these evaluations. Ninety-two undergraduates rated AI companions relative to each human source on a five-point scale and completed measures of perceived social support, beliefs about AI's supportive potential (MIDUS-AI), and perceived AI capabilities. Cumulative link mixed models, Spearman correlations, and sensitivity analyses were conducted. Students rated AI companions as significantly lower than family, friends, and mental-health professionals, with no differences across targets. A one-SD increase in MIDUS-AI more than doubled the odds of a more favorable comparative AI rating (OR = 2.13, 95% CI [1.40, 3.23]); a one-SD increase in perceived AI capabilities raised those odds by approximately 69% (OR = 1.69, [1.08, 2.62]); and a one-SD increase in perceived family support reduced them by 43% (OR = 0.57, [0.38, 0.86]). Latent-scale ICC was .25. College students perceived AI companions as inferior to family, friends, and mental-health professionals in direct comparisons. Findings describe comparative evaluations and do not support inferences about absolute AI appraisals or AI use. • College students rated AI companion support as inferior to human support • Students with stronger human support rated AI companions more negatively (vs. humans) • Specifically, higher family & friend support predicted lower AI ratings (vs. humans) • No differences were reported for friend, family, and professional support • Stronger beliefs in AI support capabilities predicted more positive AI ratings