Large Language Model-Based Chatbots and Agentic AI for Mental Health Counseling: Systematic Review of Methodologies,...

Area of Science:

• Digital mental health (MH)
• Conversational artificial intelligence (AI)

Background:

• Large language model (LLM)-based chatbots are emerging as tools for digital mental health (MH) counseling.
• Evidence on their methodological quality, evaluation rigor, and ethical safeguards is fragmented, impacting clinical readiness and deployment safety.

Purpose of the Study:

• To systematically review the methodologies, evaluation practices, and ethical/governance frameworks of LLM-based MH counseling chatbots.
• To identify gaps affecting the validity, reproducibility, and translation of these tools.

Main Methods:

• Searched major academic databases (Google Scholar, PubMed, IEEE Xplore, ACM Digital Library) for studies from January 2020 to May 2025.
• Included studies on LLM-driven MH counseling chatbots, excluding those without LLM agents or evaluable outputs.
• Appraised study quality using a traffic-light framework and synthesized data via narrative synthesis.

Main Results:

• Twenty studies were included, with GPT-based models used in 45% and fine-tuned models in 90%.
• Evaluations frequently combined qualitative assessments with quantitative language metrics.
• Limitations were noted in external validation and reporting on ethics/safety; no studies reported registered randomized controlled trials or independent clinical validation.

Conclusions:

• LLM-based MH chatbots offer potential for scalable, personalized support but require more rigorous evaluation.
• Future research must focus on clinically grounded frameworks, transparent reporting, and robust validation for safe deployment.