Towards user-centered AI counseling: a quantitative approach exploring user trust and preference dynamics

As generative AI becomes increasingly embedded in mental health services, concerns have emerged regarding its trustworthiness and appropriateness as a source of emotional support. Although recent studies have highlighted its therapeutic potential, fewer have examined how users form trust in AI counseling and whether such trust translates into acceptance relative to human counselors. Adopting a user-centered perspective, this study explores how individuals perceive, trust, and choose AI counseling in simulated scenarios. Drawing on the Human–AI Trust Framework and UTAUT, we examine how counselor presentation, contextual variation, and user characteristics jointly shape trust and preference, with particular attention to privacy concerns, attitudes toward AI, and prior AI experience. This study employed a scenario-based mixed-design experiment with one between-participant manipulation of AI framing (helper vs. neutral) and repeated evaluations across two counseling scenarios drawn from four topic × severity conditions. Within each scenario, participants evaluated five counseling methods: AI text, AI voice, AI voice + visual representation, human text, and human calling. Data from 375 Chinese university students were analyzed using hierarchical linear modeling and hierarchical generalized linear modeling to examine predictors of trust and counselor preference. The results showed a divergence between trust judgments and counselor preference. Richer AI presentation formats were associated with higher trust in AI counselors. In the preference model, privacy concern consistently reduced the likelihood of choosing AI, whereas positive attitudes toward AI consistently increased AI choice. The main model suggested lower AI preference in the social-related severe condition, but this topic-related effect was attenuated after scenario position was included in a supplementary sensitivity analysis. These findings suggest that trust in AI counseling can be strengthened in simulated settings, but comparative choice between AI and human counselors remains constrained by privacy concern, AI attitudes, and contextual conditions.