Personalized Online Counseling via AI-Driven Synchronous-Asynchronous Platform

Thesis Abstract

Online counseling has increasingly shifted toward scalable, accessible interventions, yet existing platforms frequently fail to harmonize real-time and asynchronous support with personalized therapeutic pathways, potentially limiting engagement and outcome equity across diverse client populations. This study addresses the problem by developing and evaluating a AI-driven platform that integrates synchronous video sessions with asynchronous messaging, automated mood and risk monitoring, and adaptive content tailoring to individual therapeutic goals. The aim is to assess whether a personalized, AI-supported online counseling platform improves therapeutic alliance, engagement, and symptom reduction for adults seeking mental health support, relative to standard tele-counseling. Specific objectives include (1) to design and implement an AI-driven decision-support module that personalizes session pacing, resource recommendations, and intervention prompts based on client profiles and real-time interaction data; (2) to evaluate the platform’s impact on therapeutic alliance as measured by the Working Alliance Inventory (WAI) and on user engagement metrics such as completed assignments and session adherence over a 12-week period; (3) to examine changes in clinically significant symptoms using the Depression Anxiety Stress Scales (DASS-21) and the PHQ-9/GAD-7 composite; (4) to explore user satisfaction and perceived usefulness through a structured post-treatment survey and qualitative feedback; and (5) to assess ethical, privacy, and data governance implications of AI-enabled counseling. Methodologically, this mixed-methods study adopts a quasi-experimental design with two groups an intervention group (n = 180) using the AI-driven platform and a control group (n = 180) receiving conventional online counseling, recruited from university-affiliated clinics and community mental health services. Data collection instruments include standardized scales (WAI, DASS-21, PHQ-9, GAD-7), engagement analytics (logins, message latency, assignment completion), session transcripts for thematic analysis, and a post-intervention satisfaction questionnaire. The AI components incorporate natural language processing to detect affective states and risk signals, a recommender system to tailor psychoeducation and exercises, and a decision-support engine guided by established therapeutic frameworks including Cognitive Behavioral Therapy (CBT) principles and Person-Centered Therapy elements. Data analysis comprises quantitative and qualitative strands. Quantitative analysis uses multilevel modeling to account for repeated measures across time, with regression analyses to identify predictors of symptom change and engagement. Mediation analyses test whether therapeutic alliance mediates the relationship between platform personalization and clinical outcomes. ANOVA is employed to compare group differences in primary outcomes at post-treatment and follow-up (3 months). Qualitative data from session transcripts and participant interviews are analyzed via thematic analysis to identify perceived facilitators and barriers to AI-supported counseling, with coding validated through triangulation and clinician peer debriefing. Model specification includes a hierarchical linear model of symptom trajectories and an analytical framework for the AI personalization components to delineate their contribution to outcomes. Ethical considerations focus on informed consent, data privacy, consent management, and clinician oversight, with the study adhering to GDPR-equivalent standards and institutional review board approvals. Potential limitations, such as differential attrition and platform literacy, will be addressed with intention-to-treat analyses and targeted training for participants. Expected findings anticipate that the AI-driven platform will yield higher therapeutic alliance scores, improved engagement metrics, and greater reductions in depressive and anxiety symptoms compared with standard online counseling. The mixed-methods approach is expected to reveal nuanced mechanisms by which personalization enhances therapeutic processes and highlight ethical safeguards necessary for scalable AI-enabled care. Contribution to knowledge includes empirical evidence on how synchronous–asynchronous integration coupled with adaptive personalization influences treatment outcomes in online counseling, a detailed model of the AI decision-support architecture in mental health contexts, and practical guidelines for implementing responsible AI-assisted psychotherapy. The study concludes with recommendations for refining AI personalization strategies, ensuring equitable access, and establishing governance frameworks that balance innovation with client safety, privacy, and professional accountability.