ESR-Coach: Leveraging Large Language Models for Training People to Provide Emotionally Supportive Responses

Overview

ESR-Coach is an LLM-based coaching system designed to train individuals in emotionally supportive communication. The system employs multiple AI agents to generate practice scenarios, provide reference responses, and assess user practice replies. The work addresses a gap in accessible and practical training opportunities for developing emotional support competencies, which constitute critical interpersonal skills.

Methods and approach

The system architecture leverages multiple AI agents configured for distinct functions: scenario generation, exemplar response production, and response assessment. The evaluation comprises two components: proficiency assessment of the system's performance across the three core tasks, and a user study with N=20 participants. System performance was measured on fidelity of case generation, quality of exemplary responses, and validity of response assessments. User outcomes were quantified through pre- and post-training metrics of response helpfulness and analysis of strategy diversification and effectiveness.

Key Findings

ESR-Coach demonstrated high-fidelity performance across all three functional components: case generation, exemplary response provision, and response assessment. In the user study, participants achieved an average improvement of 17% in response helpfulness following training. Post-training analysis indicated that participants employed more diverse and effective strategies in emotionally supportive communication compared to baseline performance.

Implications

The findings demonstrate that LLM-based coaching systems can effectively facilitate development of emotionally supportive communication skills through structured practice scenarios and iterative feedback. The results suggest that computational approaches leveraging language models possess sufficient social intelligence to provide meaningful training in interpersonal domains. This work establishes a foundation for exploring broader applications of LLM-based coaching in real-world scenarios where accessible skill development infrastructure is limited.