A Meta-Analysis of the Effectiveness of AI-Enhanced Counseling Interventions

As mental health problems increase and the shortage of professionals becomes more severe in modern society, AI-enhanced counseling is gaining attention as an innovative solution. This study conducted a meta-analysis to systematically verify the effectiveness of AI-enhanced counseling. Using the PICOS criteria by Wood and Mayo-Wilson, we collected relevant studies conducted domestically and published in domestic and international academic journals through the Research Information Sharing Service (RISS) and Google Scholar, using search terms such as 'artificial intelligence', 'AI', 'counseling', and 'chatbot'. Finally, 54 effect sizes from 10 studies were analyzed. The analysis showed an overall mean effect size of .313, indicating a small to medium effect size that was statistically significant. By effectiveness factors, career awareness showed the largest effect size (1.638), followed by ADHD-related factors (.266), depression and stress (.263), positive emotions and mental health (.192), social-emotional factors (.154), and anxiety and panic (.118). By AI technology type, AI tuning showed the highest effect size (4.147), followed by Humanoid robots (.399), LLM (.378), and ChatBot (.153). Age group analysis revealed the largest effect size for elementary and secondary students (.718), followed by middle-aged to elderly adults (.341) and university students (.144). By session frequency, 5-10 sessions showed the most effectiveness with an effect size of 1.638. These results suggest that AI-based counseling can be a useful tool in mental health services, particularly showing high effectiveness in career counseling and for elementary and secondary students. Future development of personalized AI systems and the need for age-specific differentiated approach strategies are recommended.