A new poll from KFF, one of the most respected health research organisations in the United States, found something worth paying attention to if your business touches employee health, benefits, or insurance. Americans who use AI chatbots for health information at least once a week are noticeably more likely to hold false beliefs about vaccines than people who do not use AI for health at all.
The numbers are specific. Among weekly AI health users, 35% believe the MMR vaccine has been proven to cause autism in children. That claim was based on a study published in the 1990s that was fully retracted after its findings were found to be false, and it has been disproved by multiple studies since. Among people who never use AI for health, only 20% hold the same belief. A gap of 15 percentage points, controlling for age, race, education, and political affiliation, is not noise.
Two other false beliefs followed the same pattern. Among frequent AI health users, 29% believe mRNA vaccines can change human DNA, compared to 20% of non-AI users. And 22% of frequent AI users believe the measles vaccine is more dangerous than the measles virus itself, compared to 15% of those who do not use AI for health.
The poll does not prove that AI creates these beliefs. It is equally possible that people already inclined toward health skepticism are the heaviest users of AI for medical questions. Researchers call this a correlation, not a cause. But a separate study published in BMJ Open in April 2026, which tested five of the most popular AI chatbots including ChatGPT, Google's Gemini, Meta AI, DeepSeek, and Grok, found that nearly half of all health-related answers were problematic. About 30% lacked full context, and roughly 20% contained outright inaccurate or misleading information. That is not a fringe finding.
A third piece of research from MIT, published in NEJM AI, found that people consistently overtrust AI-generated medical advice even when that advice is wrong. Participants in that study could not reliably distinguish between a doctor's answer and an AI-generated one, and they rated AI responses as more valid and trustworthy than doctors' responses, even when the AI was factually incorrect.
What makes all of this more consequential is the scale. OpenAI says over 230 million people ask health questions on ChatGPT every week. In January 2026, the company launched ChatGPT Health, a dedicated space where users can connect their medical records, lab results, and wellness apps directly to the chatbot. A separate KFF survey from March 2026 found that about one in three US adults has now turned to an AI chatbot for health advice in the past year.
The regulatory picture makes this harder, not easier. General-purpose AI chatbots like ChatGPT are not regulated as medical devices in the United States. Purpose-built clinical tools are, but the apps people use on their phones for free are not. Recent deregulatory signals from Washington suggest that will not change quickly.
For anyone running a business with employees or customers making health decisions, this matters in a practical way. If your staff benefits program includes health education, if your insurance products depend on customers making informed choices about vaccines or treatments, or if you work in any sector touching healthcare decisions, the channel through which people are now getting health information is unreliable in a measurable way. The answer is not to ban AI use; people will use what is available to them. The more useful response is to assume that the health beliefs of your employees and customers are increasingly shaped by AI sources that are wrong roughly half the time, and to design your communications accordingly: clearer, more direct, and backed by sources people can check.