AI Chatbot Feasibility Studies Reveal High Accuracy and Satisfaction

Two recent feasibility studies examined large‑language‑model (LLM) chatbots in medical settings. One trial in a pediatric intensive care unit used a HIPAA‑compliant GPT‑4‑based assistant to answer parents' questions, achieving a 96% real‑time satisfaction rate and 99.3% accuracy across 1,225 generated sentences. Healthcare providers rated the responses highly, and the study concluded the tool is ready for a larger randomized trial.

A separate study evaluated an LLM‑powered chatbot named Lyra for an eight‑week lifestyle‑modification program aimed at adults with overweight or obesity. All 2,288 chatbot messages were judged safe, and participants rated Lyra as helpful (69%), empathetic (88%) and easy to use (94%). The program was delivered at a cost of $1.78 per participant per week, demonstrating acceptable user engagement and safety, though larger efficacy trials are recommended.