A Reddit account says ChatGPT talked a family out of letting their father sleep off what turned out to be a stroke.
According to the poster, their father dropped his chopsticks repeatedly at dinner, his right hand went numb, his mouth looked slightly uneven, and his speech turned strange before he briefly lost his balance. The symptoms eased soon after, which nearly persuaded the family he was just tired. TechRadar, which surfaced the anecdote from Reddit, reported that when the family described the episode to ChatGPT, the bot said it sounded like a stroke and pushed them to seek emergency care immediately, despite the apparent improvement. A hospital later confirmed an ischemic stroke and treated him.
This isn't an isolated data point: the same TechRadar report cites a second Reddit user who says ChatGPT Voice kept rousing them and prompting a glucose check during a 90-minute low blood-sugar episode, suggesting a pattern of the tool nudging people toward care rather than diagnosing them outright. That distinction matters because OpenAI has been explicit that ChatGPT Health is meant to support decisions, not replace a doctor, and the useful skill here is recognizing an emergency threshold, like the stroke warning signs the CDC and the B.E.F.A.S.T. framework already describe, not identifying a blocked artery.
Two viral posts are not a clinical study, and nobody tracks how often a chatbot reassures someone who should have gone to the ER instead - so treat this as a promising anecdote, not proof the tool reliably gets the call right.