Students in higher education face a range of academic challenges, which leads to high levels of stress among many students. While coaching has been shown to help students address these challenges, it is resource-intensive and therefore not broadly accessible. AI-based chatbots may help broaden access by providing interactive and personalized guidance at scale, yet empirical evidence on their effectiveness remains limited, particularly from longitudinal experimental studies. Addressing this gap, we conducted a longitudinal between-subjects experiment with four measurement points across an academic semester, randomly assigning 101 students to one of three conditions: AI-based chatbot coaching, literature-based self-study (active control), or a waitlist control condition. Students in the chatbot coaching and literature study conditions engaged in repeated interactions with the intervention, averaging approximately 100 minutes of cumulative interaction time accumulated across four sessions of 23–28 minutes each. We examined how students’ goal attainment, university-related self-efficacy, and stress developed over the course of the study. Results showed that students in the AI-based chatbot condition achieved significantly stronger increases in goal attainment over time compared to students in both control conditions. Self-efficacy also increased in the AI-based chatbot condition over time; however, between-condition differences were not consistent across time points, suggesting that the intervention had a more limited impact on self-efficacy relative to goal attainment. Furthermore, no significant effects were found for university-related stress. The findings suggest that AI-based chatbot coaching can effectively support students in achieving university-related goals and strengthening their self-efficacy, while its impact on stress reduction may be limited.