XInsight: Integrative Stage-Consistent Psychological Counseling Support Agents for Digital Well-Being

TL;DR

Experiments show improved paradigm alignment, multi-therapy integration, interaction depth, and interpretability over existing multi-agent counseling systems, indicating that XInsight provides a practical blueprint for integrating counseling-inspired support agents into web applications for digital well-being.

Abstract

Web-based platforms are becoming a primary channel for psychological support, yet most LLM-driven chatbots remain opaque, single-stage, and weakly grounded in established therapeutic practice, limiting their usefulness for web applications that promote digital well-being. To address this gap, we present \textbf{XInsight}, a counseling-inspired multi-agent framework that models psychological support as a stage-consistent workflow aligned with the classical \textit{Exploration-Insight-Action} paradigm. Building on structured client representations, XInsight orchestrates specialized agents under a unified \textit{Reason-Intervene-Reflect} cycle: an Exploration agent organizes background and concerns into a structured Case Conceptualization Form, a Routing agent performs Adaptive Therapeutic Routing (ATR) across SFBT, CBT, and MBCT, a unified Therapeutic agent executes school-consistent submodules, and a Consolidation agent guides review, skill integration, and relapse-prevention planning. A Recording agent continuously transforms open-ended web dialogues into standardized psychological artifacts, including case formulations, therapeutic records, and relapse-prevention plans, enhancing interpretability, continuity, and accountability. To support rigorous and transparent assessment, we introduce \textbf{XInsight-Bench} with a Scale-Guided LLM Evaluation (SGLE) protocol that combines therapy-specific clinical scales with general counseling criteria. Experiments show improved paradigm alignment, multi-therapy integration, interaction depth, and interpretability over existing multi-agent counseling systems, indicating that XInsight provides a practical blueprint for integrating counseling-inspired support agents into web applications for digital well-being.

Figures (6)

• Figure 1: Overview of the three-stage psychological counseling paradigm, from client input through Exploration for trust-building and case conceptualization, Insight for goal refinement and multi-school interventions, and Action for integration, assessment, and relapse prevention with follow-up review.
• Figure 2: Overview of the XInsight framework. (a) Client representation encodes basic information and the chief complaint into a structured case prompt. (b) Multi-agent counseling workflow reinterprets the three counseling stages-Exploration, Insight, and Action-supported by Adaptive Therapeutic Routing (ATR) and specialized agents. (c) Psychological tools formalize outputs such as case conceptualization, therapeutic overgeneralization, and relapse prevention.
• Figure 4: Overview of XInsight-Bench, a benchmark designed for evaluating multi-agent psychological counseling systems. (a) Client question distribution across therapeutic schools (SFBT, CBT, MBCT) and major categories (e.g., stress/adaptation, emotion, family, somatic). (b) Client age distribution with kernel density estimation and normal fit. (c) Client occupation distribution spans service, students, IT, medical, and other groups. (d) Scale-Guided LLM Evaluation (SGLE) mode, where GPT-4o acts as a psychometric rater to score counseling quality using both school-specific standardized scales (FIT, CTS-R, MBCT-AS) and a general HPEC, enabling consistent and comprehensive evaluation across therapeutic schools.
• Figure 5: CTS-R scores of CBT-LLM na2024cbt, CACTUS lee2024cactus, and XInsight on XInsight-Bench@CBT. XInsight consistently outperforms these multi-agent counseling systems.
• Figure 6: Confusion matrix of therapy choice.