IntroductionConsistent client engagement with between-therapy session activities (i.e., homework) is a strong predictor of positive psychotherapy outcomes. Because traditional homework activities aren’t always tailored to the individual, they can fail to provide the support patients need to translate session insights into real-life change. The integration of digital tools into weekly psychotherapy presents an opportunity to overcome these limitations through personalized blended care.MethodsThis study describes the development and initial validation of Talkcast, a novel AI-powered tool that enables therapists to generate personalized audio discussions (“episodes”) to reinforce concepts from prior therapy sessions. We examined its feasibility (provider/client usage) and acceptability (provider/client quality ratings) across a five-month period (April–September 2025) on a large-scale virtual therapy platform. We also used Linear Mixed-Effects Models to evaluate associations between Talkcast usage and client engagement metrics (live sessions, provider messages, client messages).ResultsOf 6,032 active providers, 46.27% generated a Talkcast, and of 105,317 active clients, 12.97% received one, with 52.70% of recipients opening it. Over 90% of sent Talkcasts rated by providers were “Helpful,” primarily for “Session Reinforcement/Recap”. Similarly, 76.02% of client ratings were “Helpful,” citing “Personalization & Relevance” and “Specific Activities/Tools”. The analysis found a significant positive association between opening a Talkcast and higher overall platform engagement in all measured metrics (all p < 0.001). Qualitative feedback highlighted the need for improved content specificity and AI authenticity.DiscussionTalkcast demonstrated promising feasibility and high acceptability among both providers and clients. Its positive association with key client engagement metrics suggests that Talkcast is a well-integrated blended care tool used by engaged clients to reinforce the therapeutic process. Future work should focus on establishing causality and refining the AI’s content generation.
Performance of large language models in delivering accurate and comprehensible patient information on heart failure and cardiomyopathy
BackgroundLarge language models (LLMs) are increasingly used by patients seeking cardiovascular health information through digital platforms. However, their accuracy and suitability for providing guidance on