A longitudinal study published in NEJM AI evaluates seven generative AI chatbots for alcohol misuse support, finding that while they excel in empathy and non-judgmental language, they frequently provide low-quality or inaccurate medical information, highlighting critical safety concerns for clinical application.