Evaluating the efficacy of Amanda: A voice-based large language model chatbot for relationship challenges

Abstract

Digital health interventions are increasingly necessary to bridge gaps in mental health care, providing scalable and accessible solutions to address unmet needs. Relationship challenges, a significant driver of individual well-being and distress, are often under-supported due to barriers such as stigma, cost, and limited access to trained therapists. This study evaluates Amanda, a GPT-4-powered voice-based chatbot, designed to deliver single-session relationship support and enhance therapeutic engagement through natural and collaborative interactions. Participants (N = 54) completed a range of clinical outcome measures and their attitudes toward chatbots and digital health interventions pre- and post-intervention as well as two weeks later. In the interactions with the chatbot, the participants explored a range of relational issues and reported significant improvements in problem-specific outcomes, including reduced distress, enhanced communication, and greater confidence in managing conflicts directly after the interaction as well as two weeks later. While generic relationship outcomes showed only delayed improvements, individual well-being did not significantly change. Participants rated Amanda highly on usability, therapeutic skills, and working alliance, with reduced repetitiveness compared to the text-based version. These findings underscore the potential of voice-based chatbots to deliver accessible and effective relationship support. Future research should explore multi-session formats, clinical populations, and comparisons with other large language models to refine and expand AI-powered interventions.