U.S. — A patient in her 50s presented to an eye clinic with flashes of light in one eye. The patient stated she used ChatGPT to determine her condition was migraine with visual aura. A clinical exam identified a retinal tear, and the patient received laser treatment for the tear on the day of the exam. Before using ChatGPT, the patient described her original symptoms as briefer, peripheral, recurrent flashes, especially noticeable in the dark.

Generative AI chatbots are being used by a segment of the population for health information. A 2026 KFF poll found that 32% of U.S. adults said they had used AI chatbots for health information in the past year. The poll also found that many users who asked AI chatbots about physical or mental health did not follow up with a clinician afterward.

Research has explored the effects of generative chatbots on memory and diagnostic accuracy. A 2024 study from MIT and the University of California-Irvine found that people who recalled an event through a back-and-forth with a generative chatbot formed more than three times as many false memories as a control group. The study, co-authored by psychologist Elizabeth Loftus, PhD, also found that participants' confidence in their false memories remained elevated a week later.

A randomized clinical vignette study published in JAMA found that clinicians became substantially less accurate when shown biased AI diagnostic predictions. This decrease in accuracy occurred even when the AI model's explanations were displayed. Additionally, a 2026 Nature Medicine study found that members of the public using large language models were no better than controls at identifying relevant conditions or choosing the right course of action. That study also found that the large language models performed much better when tested alone than when used by members of the public.