The algorithms that most social media platforms like Facebook, X, and TikTok use today might be contributing to political conflict and polarization. But a new study from the University of Copenhagen suggests that simple changes to how posts are sorted in our feed can increase consensus and factual understanding among users.
Most of us rarely think about why we see the posts, videos, or articles that appear in our social media feeds. But the way content is sorted and presented not only decides what we see; it also greatly influences what we believe, know, and think.
In a new study, researchers from the University of Copenhagen, Dresden University of Technology, and the Max Planck Institute for Human Development demonstrate that even minimal changes to the algorithms that sort content for users can shift both the degree of polarization in groups and the accuracy of their assessments of reality.
"The platforms will say that their algorithms are just designed to help their users find the content they want to see. But our study shows how the way the algorithms work can have potentially harmful effects on how users form beliefs about the world," says Jason William Burton, Assistant Professor at the Department of Psychology and the Center for Social Data Science, and lead author of the study. He adds:
"Our results are, on the one hand, concerning because the social media platforms have become such dominant fora for civic discussion in our society, but our study could, on the other hand, be interpreted in a positive light as we show that it is possible to design algorithms that help users find more mutual understanding on these platforms."
Engagement feels good but may mislead
Most major social media platforms rank content based on engagement, such as likes, reactions, or shares. The new study suggests that while this approach gives users content they find appealing, it may also have unintended consequences.
"Algorithms optimized for engagement can feel informative and satisfying, but that doesn't mean they help people understand the world more accurately," says Jason Burton.
He and his colleagues found that a personalized, engagement‑based ranking algorithm similar to those used by platforms like Facebook and X led participants to form more polarized and less accurate beliefs than any other algorithm tested.
Ironically, these same feeds were also rated by users as more "insightful" and received more positive feedback.
Testing alternatives to today's algorithms
To explore whether social media could be designed differently, the researchers compared engagement‑based ranking with two alternative approaches - the so-called bridging‑based ranking, which prioritizes posts that receive approval from people across political divides; and intelligence‑based ranking, which prioritizes content likely to improve the overall accuracy of collective judgments.
"The bridging algorithm increased consensus between liberal and conservative participants in some cases while the intelligence algorithm improved the accuracy of their factual judgments compared with both random and engagement-based ranking," explains Jason Burton and adds:
"Our results thus suggest that it is indeed possible to design feeds that don't divide or mislead people."
Good for business or democracy?
According to the researchers, the study raises an important question: What values should social media platforms support? Today, platforms primarily optimize for engagement because this is what is best for business.
"Our experiment shows that we should explore more types of algorithms because these can have positive effects on deliberative outcomes on social media. It challenges the idea that engagement-driven social media is the only viable model," concludes Jason Burton.
The platforms themselves may hesitate to use alternative algorithms if they pose a threat to their business model so Jason Burton and his colleagues suggest that political regulation may be necessary to change the current situation.
The study "Simple changes to content curation algorithms affect the beliefs people form in a collaborative filtering experiment" is published in the Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, where it received an honorable mention for best paper.
About the study
In order to measure how algorithms affect polarization and the accuracy of group beliefs, the researchers conducted a controlled two-stage online experiment with residents of the United States, half liberals and half conservatives.
Stage 1
Five hundred participants evaluated 72 short, argumentative posts across six political and societal topics (e.g. social media regulation, religion, and concrete societal forecasts). Each participant upvoted, downvoted, or ignored each post, allowing the researchers to map which types of content liberals and conservatives tended to support or reject.
Stage 2
A new group of 1,000 participants was asked to state their beliefs on the same six topics, then they were shown short three-post "feeds," and then they we asked to re-state their beliefs. The posts that appeared in a participant's feed depended on which condition they were assigned to, with each condition involving a different approach to algorithmic content curation:
- random ranking,
- engagement-based ranking (content that received the most engagement of any kind),
- personalized engagement-based ranking (content approved by your political in-group),
- bridging-based ranking (content approved by both liberals and conservatives)
- and intelligence-based ranking (content expected to improve collective belief accuracy).
The researchers then measured whether participants' beliefs converged or polarized (consensus); whether group-level judgments became more or less accurate; and how participants perceived the content (e.g. as civil, emotional, or insightful).
