Do Online Health Information Bubbles Exist? An Analysis of Personalized Google Search Results

Abstract

Background: There is considerable debate focusing on the consequences of online health information, which is generally related to issues such as information quality and the necessary literacy to identify such information. However, the availability of information is a prerequisite to provoke positive outcomes. Search engines inherently prioritize information. In this study, we consider the nature of the health information that is prioritized by Google. Moreover, we consider the potential occurrence of health filter bubbles, due to algorithmic personalization.Such bubbles could restrict the availability of high-quality information, and amplify user selfconfirmation.Objective: The aim of this study was to examine the nature of the information sources that are retained as results to a standardized set of health-related Google search queries and to inquire the way in which these search results reflect personalization. Methods:We harvested the personal search results of 380 Google users on 16 standardized health-related search queries. In addition, we conducted a survey measuring socio-demographic and health behavior variables. To define the nature of the collected search results we undertook a content analysis of the collected search queries, coding entries for type of website, revenue model, information authorship, and source materials.Results: An initial Latent Class Analysis (LCA) of the unique search results, based on type of website, revenue model, information authorship, and source materials showed four clusters of information types: commercial health news (33%), health goods and services (30%), user contributions (21%), and health advocacy (16%). To identify personalization, we conducted a subsequent LCA of the entire set of search results per participant. Five patterns emerge, significantly differing for the information they are composed of (i.e., 11 to 41% difference). Still, tests of covariates did not reveal any significant relations (p < .05) with user characteristics (i.e., participants' gender, age, perception of current health, frequency of searching for online health information, and the age of their Google accounts). Conclusions:The results indicate that the overall majority of health information is commercial in nature, and often lacks proper information on authorship and source materials. Moreover, we find mild evidence for the personalization of health information. However, we were unable to isolate potential grounds for this personalization.Trial Registration: We received ethical clearance from the Social and Societal Ethics Committee (SMEC) at the KU Leuven (https://admin.kuleuven.be/raden/en/smec). The clearance is registered as G-2016 01 449.