Structure Matters: Evaluating Multi-Agents Orchestration in Generative Therapeutic Chatbots

2026-02-28

AI Summary

This study addresses the challenge of balancing clinical protocol adherence with conversational naturalness in generative psychotherapy chatbots. Building upon Self-Attachment Therapy (SAT), the authors propose a multi-agent architecture that integrates a finite state machine with shared long-term memory to explicitly embed therapeutic phase structure into the dialogue system. Structured guidance is achieved through coordinated multi-agent interactions and prompt engineering. In an eight-day randomized controlled trial involving 66 Persian-speaking participants, the proposed approach significantly outperformed both single-agent and unguided large language model baselines, demonstrating superior performance in dialogue naturalness, human-likeness, and most clinical efficacy metrics. These results validate the effectiveness of structured multi-agent design in enhancing the quality of therapeutic conversations.

Technology Category

• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

While large language models (LLMs) excel at open-ended dialogue, effective psychotherapy requires structured progression and adherence to clinical protocols, making the design of psychotherapist chatbots challenging. We investigate how different LLM-based designs shape perceived therapeutic dialogue in a chatbot grounded in the Self-Attachment Technique (SAT), a novel self-administered psychotherapy rooted in attachment theory. We compare three architectural variants: (1) a multi-agent system utilizing finite state machine aligned with therapeutic stages and a shared long-term memory, (2) a single-agent using identical knowledge-base and the same prompts, and (3) an unguided LLM. In an eight-day randomized controlled trial (RCT) with N=66 Farsi-speaking participants, balanced across the three chatbots, the multi-agent system is perceived as significantly more natural and human-like than the other variants and achieves higher ratings across most other metrics. These findings demonstrate that for therapeutic AI, architectural orchestration is as critical as prompt engineering in fostering natural, engaging dialogue.

Problem

• Generative Therapeutic Chatbots
• Multi-Agent Orchestration
• Structured Dialogue
• Psychotherapy
• LLM-based Design

Innovation

• multi-agent system
• finite state machine
• therapeutic chatbot
• Self-Attachment Technique
• architectural orchestration