How platforms rank content has become a central issue in the digital information ecosystem. Algorithms determine what millions of users see each day, shaping which voices gain visibility and how political information spreads online. Despite the intensity of debate around these systems, there has been limited experimental evidence showing how algorithmic feeds influence political attitudes in real-world settings.

A new study examining X offers fresh insight into the question. The research finds that algorithmic feeds increase engagement compared with chronological feeds while also shifting certain political attitudes and altering the mix of political content users encounter. The findings help clarify how recommendation systems shape the information environment in which audiences consume news and commentary.

The study tracked nearly 5,000 active U.S. users of X over a seven-week period. Participants used either an algorithmic feed or a chronological feed, enabling researchers to compare how each format affected engagement, content exposure, and survey responses related to policy priorities.

Researchers randomly assigned participants to one of the two feed experiences. Throughout the study, they analyzed survey responses, the content appearing in each user’s feed, and behavioral signals such as likes, reposts, and comments.

Patterns emerge in engagement and content exposure

Posts surfaced by the algorithm generated substantially more interaction than those appearing in chronological feeds. On average, recommended posts received about five times more likes and several times more reposts and comments. The higher level of engagement reflects how algorithmic systems elevate posts that spark strong reactions or conversation.

Participants who moved from chronological feeds to algorithmic feeds were also more likely to maintain or increase their use of X. In practice, the recommendation system steered attention toward posts that generate ongoing engagement, reinforcing activity on the platform.

The algorithm also altered the composition of content appearing in users’ feeds. Recommendation-driven feeds contained more political posts overall and a greater share of content from political activists. At the same time, posts from traditional news organizations appeared less frequently.

Across users with different political affiliations, the algorithmic feed increased the share of conservative political content appearing in feeds. As exposure shifted, so did the issues users emphasized. Participants became more likely to prioritize policy topics commonly highlighted by Republicans, including immigration, crime, and inflation.

Algorithms shape information networks

Public discussion about social media algorithms often focuses on how platforms rank individual posts. Earlier research examining Facebook and Instagram during the 2020 election suggested that ranking alone may not significantly alter political attitudes. In those experiments, removing algorithmic feeds did not produce measurable changes in users’ views.

The new research on X suggests that the political effects of algorithms may emerge through a different mechanism. Recommendation systems influence which accounts users discover and choose to follow, gradually shaping the networks that define their information environment.

The study finds that the most noticeable changes occur when users first move from a chronological feed to an algorithmic one. Exposure to recommendations encourages users to follow new accounts, particularly those run by political activists. Once those accounts become part of a user’s network, their content continues appearing in the feed even if the ranking system changes.

As a result, turning an algorithm off does not necessarily reverse the earlier effects. The recommendation system has already reshaped the user’s information network, influencing which voices appear regularly in the feed. In that sense, algorithms operate not only as ranking systems but also as engines of network formation.

The study also indicates that algorithmic exposure may influence politics indirectly. Rather than shifting party identification, recommendation systems appear to affect how users interpret events and which policy issues they view as most important.

The authors note that the findings apply specifically to X and to the time period examined in the experiment. Algorithms evolve frequently, and different platforms may produce different outcomes. Even so, the research provides rare experimental evidence showing how recommendation systems shape political information flows online.

For publishers and policymakers alike, the implications extend beyond a single platform. As algorithmic feeds increasingly mediate access to news and public debate, understanding how those systems influence engagement, exposure, and information networks remains essential.