Websites such as Facebook and Google use information about our previous behaviour to prioritize and display information for us. This phenomenon is sometimes called a 'filter bubble' because it means we are less likely to find information that is novel or challenging. Filter bubbles may reinforce anti-science health beliefs and make it harder to disseminate evidence-based information to the people that need it most.
To explore the effect of filter bubbles in Facebook in the context of a controversial health belief and to consider the implications for evidence-based health information.
A Facebook account was created and a search for 'vaccine harms' was performed. Groups and pages suggested by Facebook were browsed and the first results were 'liked', as were the suggestions. After the timeline was populated, a relevant article was 'liked' and suggestions were classified into pro-vaccines or anti-vaccines. Screen capture software was used to record the experience. The process was repeated with different accounts to enable a comparison between different 'experiences' of the same website according to: pro-versus anti-vaccine search terms; UK versus American user; English language versus Spanish language user (location controlled).
The search for 'vaccine harms' was performed in January 2014. Within three clicks of the search results we identified three anti-vaccine pages with over 125,000 'likes'. Very quickly, our timeline was populated with anti-vaccine information. After 'liking' an article, four more anti-vaccination suggestions appeared, all with over 10,000 'likes'. Our comparisons revealed a different Facebook experience according to opinion (pro- or anti-vaccine), location and language.
Personalization of online searching makes it more likely that people will find information that they already agree with and less likely that they will find information that challenges their views. As personalization becomes the norm in online services, research is needed to assess the impact on public health and the prospects for interventions to counter their impact.