The Algorithmic Public Sphere: How Recommendation Systems Reshape Selective Exposure Theory

Recommendation systems now curate much of the political information that citizens encounter online, yet selective exposure theory was developed before algorithmic curation existed. This paper conducts a systematic review of ten empirical, quantitative studies that test how recommendation systems and personalised algorithms shape exposure to political and news content. Guided by selective exposure theory and Habermas's theory of the public sphere, the review compares studies published between 2014 and 2020 in major communication, information-systems, and interdisciplinary journals. The reviewed studies use behavioural trace data, controlled experiments, platform audits, and cross-national surveys, which allows their outcomes to be compared directly. The synthesis shows that recommendation systems do not produce one uniform effect. Individual choice narrows exposure at least as much as algorithmic ranking does; recommender design, rather than personalisation itself, determines whether diversity rises or falls; and platform-level curation tends towards homogenisation around a small number of legacy outlets rather than towards fragmentation into isolated bubbles. This paper argues that these findings require selective exposure theory to be extended rather than replaced. It proposes the concept of algorithmic co-selection: recommendation systems act as a second, non-human selector that interacts with individual predispositions in platform-specific ways. The algorithmic public sphere that results is neither uniformly fragmented nor uniformly integrated, but differentiated by platform architecture and design choice.