As large language models (LLMs) are increasingly integrated into digital mental health services, understanding how users experience AI-guided support is essential for designing trustworthy and effective interventions. In this study, university of applied sciences students (N = 50) participated in a classroom-based intervention where they interacted with two prototype conversational agents (CAs), designed to be embedded in an internet-based Acceptance and Commitment Therapy (iACT) program: one supporting post-exercise reflection and another guiding personal values clarification. Post-interaction surveys were completed, measuring trust (HCTS), satisfaction (CSAT), barriers to help-seeking (BHSS), and preferred support formats. Results showed that student perceived the risk of using the values-oriented CA as lower than using the exercise reflection CA, though trust overall was comparable. Students with lower emotional control barriers were more likely to prefer AI-guided support, particularly among students in the fields of health and welfare. A logistic regression analysis revealed user satisfaction and emotional control as statistically significant predictors of students’ preference for AI-guided support, indicating that students with higher overall satisfaction with the CA interactions and lower emotional control (i.e., more open to talk about their emotions) were more likely to prefer AI guidance over other guidance modalities (personal guidance, group guidance and self study). As an exploratory study, these findings offer early insights into how student’s help-seeking barriers and experiences of interacting with CAs shape openness to AI-based support. We discuss implications for future CA design and call for longitudinal, behaviorally grounded research to further explore trust-barrier dynamics in AI-supported psychological interventions.