Filter Bubbles in the Age of GenAI - A Literature Review

Abstract

The term "filter bubble" refers to the possibility that online content customization, resulting from the use of algorithms, could isolate users from wider or even differing perspectives. Recommender systems, which are implemented in all digital platforms and rely on algorithms to anticipate users' preferences and recommend relevant items, are particularly vulnerable to this phenomenon. The rapid development of generative artificial intelligence (GenAI) has brought in focus how recommender systems narrow what users see and strengthen the filter bubble effect. This literature review examines the recent work on the impact of GenAI on search, social media news feeds and customized recommendations. The most direct evidence suggests that GenAI can amplify selective exposure when users make use of large language model systems in ways that tend to confirm their views and expectations, as results and outputs are personalized to existing preferences. At the same time, GenAI can also be used to interrupt the exposure to more of the same information, by directing users to diverging material and alternative viewpoints. Ultimately, GenAI seems to be less of a cause for filter bubble emergence and more of an amplifier. It can intensify narrowing through personalization and distilled content, but it can also be deployed to diversify exposure if platforms choose to optimize for diversity rather than pure user engagement. The review concludes by identifying further research needed to establish causal effects across social media, search, and e-commerce.