Digital mental health tools, such as chatbots, address the growing demand for psychological support but often lack dynamic, empathetic, and personalized interactions. This study evaluates the integration of Large Language Models into structured chatbots to enhance the delivery of the World Health Organization’s Self-Help+ intervention. A within-subject experiment with 44 participants compared a standard state-machine chatbot to a Large Language Model augmented version. Participants rated the chatbots using a specialized evaluation scale and provided qualitative feedback. The Large Language Model enriched the chatbot significantly; its integration improved emotional support, memory retention, and user satisfaction. Challenges included hallucinations, verbosity, and limited conversational flexibility due to the retainment of the finite-state structure. These findings highlight Large Language Models’ potential to improve digital mental health tools and that structured conversation approaches are perceived as a major drawback for users, even when interacting with Large Language Models. Future efforts should, therefore, address hallucination mitigation and explore multi-agent architectures to enhance adaptability and user experience.