When It Comes to Vaccines and Autism, It Is User Error Not AI Disinformation

A KFF poll released this week found that adults who use AI chatbots weekly for health information are more likely to lean toward the false claim that MMR vaccines cause autism. 35 percent called it probably or definitely true, compared to 20 percent among those who never use AI for health. The Guardian and others read this as evidence that chatbots are spreading disinformation. What it actually reveals is the stubbornness of false beliefs. But don’t take my word for it. Ask ChatGPT, Gemini, Claude, or Copilot whether vaccines cause autism. Each replies no and provides the evidence to back the conclusion. Published evaluations of chatbot vaccine answers reach the same result, rating default responses as accurate and aligned with WHO and CDC guidance. The fault lies not in the AI systems but in ourselves. KFF’s data identifies what accounts for this particularly noxious form of user error. Adults without a trusted health-care provider were far more likely to endorse vaccine falsehoods—46 percent believed the claim that COVID-19 vaccines killed more people than the virus, against 24 percent among those with a trusted provider. It is unsurprising that people who distrust physicians are also the people most likely to route their health questions to a chatbot. KFF anticipated the selection bias problem and controlled for age, race and ethnicity, education, and partisanship, but those are weak stand-ins for disposition and bias. Prior distrust of health institutions or the absence of a regular doctor is not a demographic variable, and each arguably drives both the decision to ask AI about health and the willingness to believe the myth. Chatbots are agreeable by design, and a user who arrives convinced that vaccines cause autism (or kill unwary individuals) can with some effort coax a model toward agreement. In fact, there’s a cottage industry of developers building models for exactly this purpose. Fewer than one in ten adults are firm believers in popular health myths, and 55 percent reject all of them. This makes one wonder if the real disinformation campaign is not about health falsehoods but against AI systems themselves.