KFF poll: frequent AI health-chat users more likely to believe vaccine myths

A KFF poll released Tuesday found a blunt pattern: US adults who frequently seek health advice from artificial intelligence chatbots are more likely to believe myths about vaccines. The survey, conducted in May, polled a representative sample of 2,480 US adults and examined whether chatbot use tracks with belief in specific falsehoods.

In that May survey, frequent use of AI tools and chatbots correlated with belief in vaccine falsehoods, including the claim that vaccines cause autism. It also included another widely circulated myth, that the measles vaccine poses more danger than the corresponding virus. The key point here is not that AI chatbots are the sole cause of misinformation. It is that the correlation persisted even after KFF controlled for factors such as age, race, education, and political partisanship.

Why executives should care is that this is exactly how product risk becomes policy risk. When a technology category touches health decisions, regulators tend to look past intention and focus on outcomes: what people believe, what they do, and what harms follow. KFF's finding that the connection remains after controlling for demographics means the issue is not neatly explained by who uses chatbots. It also implies the risk is tied to the interaction itself, the framing people encounter, or how people rely on AI responses in moments when they are uncertain.

The mechanics matter. In the AI chatbot era, health questions often start as “what’s true?” and end as “this feels credible.” Even if a system is not designed to promote myths, it can still produce persuasive-sounding content, summarize misleading narratives, or fail to correct a false premise quickly. A correlation like KFF’s is a warning light for product teams: if frequent users are more likely to hold false beliefs, then the user journey is doing more than answering questions. It is shaping confidence.

From a governance standpoint, boards should treat this as a liability-and-trust issue, not just a communications issue. The survey controls for age, race, education, and political partisanship, which means executives cannot simply say “it’s demographics” or “it’s politics.” For public companies, investors are increasingly sensitive to “adjacent harm” where a product is not medically regulated in the way a drug is, but its use influences health behavior and belief. For boards, that means risk committees should ask whether the company has a documented approach to misinformation, health context handling, and escalation when a user asks about vaccination safety.

There is also a market reality here: “frequent” use is a different category than casual curiosity. People who repeatedly seek health advice from chatbots are likely using AI as a trusted intermediary. That changes the duty-of-care question, because the more a tool is relied on, the more any error can cascade. KFF’s May survey involved 2,480 representative US adults, giving the finding statistical weight, but the strategic implication is bigger than the numbers: the AI health user is not a one-off experiment. They are building habits.

Second-order implications follow quickly. First, expectations for safety and accuracy rise, even without a single new law hitting tomorrow. When a reputable poll links chatbot use to belief in vaccine myths like “shots cause autism” and “measles vaccine poses more danger than the virus,” it becomes harder for any operator to argue that the problem is purely educational. Second, partnerships and distribution channels may face scrutiny. If a health-focused AI product is embedded in a platform, the partner may be pulled into the same reputational gravity even if the platform did not author the content.

Finally, for peers across AI, the strategic stakes are straightforward: companies that are serious about long-term adoption need to earn trust in high-stakes domains. A poll finding that the correlation remains after controlling for age, race, education, and political partisanship suggests the challenge is structural, not demographic. Executives should see this as an early signal of where user trust can break, where regulators will pay attention, and where boards will eventually be asked uncomfortable questions about how AI systems handle health misinformation under real-world use patterns.