Overview
- An Oxford University study in Nature Medicine with nearly 1,300 participants found chatbot guidance was no more effective than standard internet searches at identifying health problems.
- The trial documented large interaction gaps, with small changes in how questions were phrased producing different answers and users often omitting crucial details that skewed the advice.
- Separate evaluations reported that ChatGPT undertriaged more than half of gold‑standard emergencies, steering people to delayed 24–48 hour evaluations instead of the emergency department.
- Research from the London School of Economics and MIT Jameel Clinic found models systematically downplayed women’s symptoms and more often recommended lower levels of care, with informal or typo‑filled prompts more likely to discourage seeking treatment.
- Consumer interest remains high and companies are expanding offerings, with surveys showing strong willingness to use AI for health and Microsoft unveiling Copilot Health as OpenAI’s ChatGPT Health debuted in January.