Towards a Large Language Model Cognitive Storytelling Intervention: Development and Evaluation in a Laboratory Setting

Dementia progressively impairs memory and identity, leading to social withdrawal. Collaborative storytelling can foster social engagement and cognitive stimulation, but traditional approaches require trained facilitators, limiting scalability. We present a storytelling intervention delivered by a conversational agent blending structured scripting with generative AI to support shared storytelling. Implemented in Botpress, the system combines rule-based dialogue for safe turn-taking and a coherent story arc with a constrained large language model for creative expansions based on user input. A pilot study with 11 healthy adults explored usability and engagement. Participants described the experience as easy to use and enjoyable, reporting strong connections with the agent and feeling transported by shared narratives. They praised the agent’s empathic voice, callbacks to earlier story points, and sense of agency, while identifying latency and occasional factual errors as areas for improvement. These findings suggest storytelling systems support co-creation and offer promising foundations for scalable dementia interventions.