Abstract
<p>Background: Conversational artificial intelligence (AI) is increasingly used for emotional support. Among sustained users, AI may become an enduring part of their broader support ecology rather than a one-off or short-term coping strategy, yet little is known about how AI and human support are configured within this population.Objective: This study aimed to identify AI–human support profiles among sustained users of conversational AI for emotional support in China and compare their AI use, satisfaction, psychological distress, and positive well-being.Methods: In this cross-sectional study, 484 individuals who had used conversational AI for emotional support at least once per week for more than 1 continuous month were recruited online between February and March 2026, with sustained use verified through uploaded conversation screenshots. Latent profile analysis was conducted using perceived emotional support and emotional closeness from AI and human sources, and differences in AI use, satisfaction, and psychological outcomes were compared across profiles.Results: Participants were on average 23.62 years old (SD 4.24), and 59.3% (287/484) were women. A 3-profile solution was retained (entropy=0.824): human-anchored integrated (221/484, 45.7%), AI–human divergent (167/484, 34.5%), and AI-compensatory (96/484, 19.8%). The profiles were characterized by strong human resources with substantial AI support, greater perceived support from AI than from humans despite greater human closeness, and moderate AI support with near-floor human support, respectively. Profile differences were larger for human support and closeness (partial η²=.878 and .868) than for AI support and closeness (partial η²=.058 and .010). General-purpose AI systems predominated, and no significant profile differences existed in system type or most use behaviors. Significant profile differences existed in total AI use time (P=.019) and satisfaction (P=.013), with the AI–human divergent profile using AI longer but reporting lower satisfaction than the human-anchored integrated profile. No significant differences existed in depression, anxiety, perceived stress, or loneliness (all P≥.050). Significant differences existed in life satisfaction (P=.008) and subjective happiness (P&lt;.001), with the human-anchored integrated profile reporting the highest positive well-being.Conclusions: These findings highlight heterogeneity in how AI and human support are configured within sustained users’ broader support ecologies and suggest that human-anchored integration is associated with more favorable positive well-being, despite broadly similar AI use behaviors and psychological symptom levels across profiles.</p>