AI-Driven Counseling Chatbot for Mental Health Support in Schools

Thesis Abstract

The study addresses the escalating gap in timely access to mental health support for students within school environments, where stigma, resource constraints, and limited professional availability impede proactive help-seeking and early intervention. The central aim is to evaluate the efficacy, acceptability, and equity implications of an AI-driven counseling chatbot designed to provide confidential, evidence-based mental health support to secondary school students. Specific objectives include (1) assessing chatbot usability and student engagement over a 12-week deployment, (2) examining changes in help-seeking intentions and self-reported well-being, (3) evaluating the chatbot’s alignment with established therapeutic frameworks and ethical guidelines, (4) identifying differential effects across demographic groups to address equity considerations, and (5) developing a scalable implementation framework for integration with school counselling services. The study is guided by the Technology Acceptance Model (TAM) and the Self-Determination Theory to explore motivation, perceived usefulness, ease of use, autonomy, and relatedness in human–technology interactions, complemented by the Capability–Opportunity–Motivation–Behavior (COM-B) model to elucidate behavior change processes. Methodologically, the research adopts a mixed-methods design combining a quasi-experimental pretest–posttest control group with an embedded qualitative component. The population comprises students aged 13–18 from ten secondary schools within a metropolitan district, with four schools assigned to the intervention arm (n ? 1,600 students) and six to the control arm (n ? 2,400 students) over a 14-week cycle. A stratified random sampling approach yields a sample of 400 students in the intervention group for quantitative analysis, supplemented by 40 focus participants across intervention and control schools for in-depth interviews and 20 teachers and school counselors for stakeholder perspectives. Data collection instruments include (a) the Depression, Anxiety, and Stress Scales (DASS-21) to measure psychological distress, (b) the Strengths and Difficulties Questionnaire (SDQ) for behavioral and social functioning, (c) the General Help-Seeking Questionnaire (GHSQ) adapted for digital modalities, (d) the System Usability Scale (SUS) and a bespoke chatbot satisfaction survey, and (e) semi-structured interview guides for focus groups and key informants. Data collection will occur at baseline (week 0), mid-intervention (week 7), and post-intervention (week 14). Quantitative analysis will employ descriptive statistics, repeated-measures ANOVA to detect changes over time, and hierarchical linear modeling to account for nested data within schools. Mediation analyses will test whether perceived usefulness and ease of use mediate the relationship between chatbot exposure and help-seeking intentions. Subgroup analyses will examine differential effects by gender, socioeconomic status, and prior mental health history. The qualitative component will be analyzed using thematic analysis following Braun and Clarke’s framework, with triangulation to enhance validity and explain quantitative results. The study will ensure ethical rigor through informed consent procedures, data anonymization, adherence to institutional review board guidelines, and a risk management plan for safeguarding participants. Expected findings include (i) higher engagement and satisfaction with the chatbot among students in the intervention group compared with controls, (ii) statistically significant improvements in self-reported well-being and reductions in distress levels as measured by DASS-21, (iii) increased help-seeking intentions and reductions in stigma associated with seeking support, (iv) evidence of moderation by demographic factors indicating differential benefits, and (v) qualitative insights into facilitators and barriers to adoption, including perceived privacy, cultural relevance, and alignment with school counselling workflows. Anticipated challenges include maintaining data privacy, ensuring cultural and linguistic inclusivity, and integrating chatbot interactions with human counsellors in a coordinated care model. The study contributes to knowledge by empirically validating a scalable AI-driven intervention within real-world school settings, extending theories of technology acceptance and behavior change to digital mental health supports for adolescents, and identifying practical implementation levers and equity considerations for school-based mental health services. The expected conclusion underscores the potential of AI-driven counseling chatbots to complement, rather than replace, human professionals, provided there are robust ethical safeguards, transparent governance, and structured referral pathways. Recommendations emphasize scalable integration with school counselling teams, ongoing monitoring of mental health outcomes, routine evaluation of fairness across subgroups, and iterative refinement of the chatbot’s therapeutic content to maintain evidence-based practice and adolescent relevance.