Frequent AI chatbot users likelier to believe vaccine myths

TL;DR:

• A KFF poll of 2,480 US adults found that frequently seeking health advice from AI chatbots correlates with believing vaccine falsehoods.
• Among weekly AI health users, 35% thought it “definitely or probably true” that MMR vaccines cause autism, against 20% of non-users — a link that held after controlling for age, education and politics.
• The poll did not record which chatbots people used, and different models produce different levels of misinformation.

A new survey has put a number on a long-standing worry: that turning to AI for medical advice may leave people more misinformed. Health research firm KFF found that US adults who consult chatbots for health information at least weekly were consistently more likely to endorse debunked claims about vaccines than those who never do.

Correlation, with caveats

The pattern spans several myths. Frequent AI health users were more likely to believe that mRNA vaccines alter DNA (29% versus 20% of non-users) and that the measles vaccine is more dangerous than measles itself (22% versus 15%). The autism falsehood — a central pillar of the anti-vaccine movement, traced to a retracted 1990s study — showed the widest gap. Crucially, the association persisted after controlling for age, race, education and political leaning.

The findings come with real limits. Correlation is not causation: people already inclined to distrust mainstream medicine may gravitate to chatbots that confirm their views, rather than the tools creating the belief. KFF also did not ask which models respondents used, and social media use showed an even stronger link to misinformation. Notably, AI health-seekers skewed towards higher-income, college-educated users, while social media health advice skewed lower-income.

For UK readers, the direct data is American, but the underlying dynamic is not bounded by geography. Health is already among the most common uses of chatbots — OpenAI has said hundreds of millions ask health questions weekly — and the NHS and public-health bodies face the same challenge of patients arriving with AI-shaped assumptions. The lesson is less “ban the bots” than the need for accuracy safeguards and clear routes to qualified human advice.

Looking forward

As chatbots become a default first stop for health questions, the variation between models becomes a public-health variable in its own right. How AI firms tune responses to contested medical topics — and whether regulators treat health misinformation as a safety issue — will determine whether these tools inform or mislead at scale.