[Paper Review] A Survey of Large Language Models in Psychotherapy: Current Landscape and Future Directions

Hongbin Na, Yining Hua+7 more|ArXiv.org|Feb 16, 2025

Mental Health via Writing3 citations

TL;DR

This survey proposes a tripartite taxonomy (Assessment, Diagnosis, Treatment) for LLM applications in psychotherapy, reviews current studies (69 total) across tasks, and discusses challenges and future directions for end-to-end, context-aware psychotherapy with LLMs.

ABSTRACT

Mental health is increasingly critical in contemporary healthcare, with psychotherapy demanding dynamic, context-sensitive interactions that traditional NLP methods struggle to capture. Large Language Models (LLMs) offer significant potential for addressing this gap due to their ability to handle extensive context and multi-turn reasoning. This review introduces a conceptual taxonomy dividing psychotherapy into interconnected stages--assessment, diagnosis, and treatment--to systematically examine LLM advancements and challenges. Our comprehensive analysis reveals imbalances in current research, such as a focus on common disorders, linguistic biases, fragmented methods, and limited theoretical integration. We identify critical challenges including capturing dynamic symptom fluctuations, overcoming linguistic and cultural biases, and ensuring diagnostic reliability. Highlighting future directions, we advocate for continuous multi-stage modeling, real-time adaptive systems grounded in psychological theory, and diversified research covering broader mental disorders and therapeutic approaches, aiming toward more holistic and clinically integrated psychotherapy LLMs systems.

Motivation & Objective

• Introduce a standardized taxonomy aligning psychotherapy processes with LLM applications (Assessment, Diagnosis, Treatment).
• Survey 69 studies to map the landscape of LLM-based psychotherapy research and identify gaps in coverage, theory integration, and multilingual resources.
• Highlight challenges (bias, limited disorder coverage, model alignment) and propose directions for an end-to-end, integrative framework.

Proposed method

• Propose a conceptual taxonomy (Assessment, Diagnosis, Treatment) with dynamic interrelations (Synthesizing, Framing, Customization).
• Systematically review recent advances in LLM applications across the three components and their methodological approaches.
• Synthesize landscape characteristics: disorder coverage, linguistic resources, psychotherapy theory alignment, techniques used, and model types (commercial vs. open, prompt-based).
• Identify research gaps (linguistic bias, limited disorders, underrepresented therapies) and articulates future directions for holistic integration.

Experimental results

Research questions

• RQ1 How are LLMs currently being applied across assessment, diagnosis, and treatment in psychotherapy?
• RQ2 What are the main methodological approaches, data sources, and model types used in this field?
• RQ3 What gaps and biases exist (language, disorder coverage, therapy models), and how can future work address them?
• RQ4 What would an end-to-end, integrative LLM-based psychotherapy framework look like, and what challenges must be overcome?

Key findings

• 33 studies on assessment, 9 on diagnosis, 32 on treatment (5 overlap across dimensions).
• About 74% of studies used commercial LLMs and ~77% employed prompt-based techniques, indicating reliance on closed models and prompts.
• English-language dominance with multilingual underrepresentation; limited coverage of disorders beyond depression/anxiety and limited psychotherapy theories in use.
• Research shows fragmentation: many studies focus on isolated tasks rather than integrated, end-to-end systems.
• Emerging approaches explore bias mitigation, domain-specific summarization, and multi-agent or hybrid therapy frameworks; however, gaps remain in theory alignment and culturally diverse datasets.
• Future work should pursue integrative, multi-turn systems that span full psychotherapy processes and embrace broader disorders and therapeutic modalities.