The Role of Social Media Algorithms in Shaping Political Polarisation in the United States

Human Rights Research Center
3 hours ago
15 min read

August 5, 2026

Social media is a central space where political polarisation is encountered, reproduced, and circulated in the U.S. Platforms like Facebook, Instagram, TikTok, YouTube, Reddit, and X are no longer sites designed for personal communication (Egan, 2026). They double as political information systems, having a hand in what users see, what they believe, and which voices are made visible in public debate (Liedke and St. Aubin, 2025).

This shift has raised questions about whether social media has deepened political division or opened access to more diverse political perspectives. While political polarisation in the U.S. existed before the emergence of online platforms such as Reddit and Instagram, it has intensified despite the rise of these platforms. On these platforms, algorithms rank and recommend content according to user behaviour, shaping what they see most frequently. This has sparked a discussion about whether the way algorithms work on these platforms is making the divide worse by controlling how people see political information.

(Gauthier et al., 2026, Guess et al., 2023).

Therefore, this article discusses how social media algorithms can influence political polarisation in the U.S. by selecting what is visible and how much exposure users are given to alternative perspectives. It will investigate the relationship between curation systems, recommendation structures, confirmation bias, and user behaviour in relation to amplifying polarisation. Furthermore, it will explain the human rights implications of algorithmic ranking, debates around freedom of expression, and overall impacts on democratic participation (OHCHR, 2024). This article argues that algorithmic systems should not be understood as neutral operations behind the screen. They are an integral part of the infrastructure in which democratic discourse is mediated and consumed. The power of algorithmic influence remains largely unchecked, unknown, and under-researched; this article seeks to address this contention.

(Metzler and Garcia 2023; DSA observatory, 2025).

The Mechanics of Social Media Algorithms

Social media algorithms work behind the scenes to categorise, rank, filter, and recommend content, influencing what emerges, receives visibility, and is suggested next. Recommendation algorithms rely on data on user behaviour, including likes, shares, comments, watch time, follows, searches, and previous engagement metrics (DSA Observatory, 2025). This indicates that users are not selecting from a neutral pool of information, but instead, they are navigating a political landscape that is already structured by rankings and personalised predictions. A chronological feed orders content by time, whereas an algorithmic feed organises it by predicted relevance and past behavioural responses. This distinction matters because political content is made visible according to the reactions it generates. Posts likely to attract attention, emotion, and debate are more likely to be prioritised than material that is slower and less emotive (Milli et al., 2025; DSA Observatory, 2025).

This doesn’t mean that all algorithms are harmful or that every user is automatically pushed towards charged content. The problem is that engagement can become a poor substitute for public value. A post may gain popularity because it evokes anger or disagreement, not because it is accurate or democratically useful.

In political contexts, this situation allows sensationalist and antagonistic content to gain greater visibility compared to more measured forms of political and educational material (Brady et al., 2021; Milli et al., 2025). This results in an imbalance formed by algorithmic reward and can be detrimental to combatting political polarisation. Consequently, a post that portrays opponents as a threat or simplifies complex issues may attract more viewers and seem more important than it actually is (Yuan and Chang, 2026).

This uneven visibility goes beyond simply privileging one post over another. It can influence the wider political environment by making some viewpoints appear more common or legitimate than they are. When users repeatedly encounter the same kinds of political narratives, exposure starts to appear as consensus. Through this, algorithmic recommendation moves beyond content ranking and starts to reinforce the social conditions that cement rigid political identities and enclose users in a set belief (Gauthier et al., 2026; Ibrahim et al., 2026).

How Ideology is Solidified

In a discussion about algorithmic polarisation, “echo chambers” and “feedback loops” are often mentioned. However, it is crucial to differentiate them in context with social media algorithms. A filter bubble arises from algorithmic curation of information, whereby algorithms predict what users want to see, resulting in limited exposure to diverse sources. An echo chamber describes how users form a selective social environment, primarily interacting with others who reaffirm their existing ideology while filtering out opposing viewpoints (Ahmmad, 2025; Hartmann, 2025)

Users also reinforce their digital environment by following accounts that align with their beliefs and sustain their desired feed through repeated engagement. Algorithms were designed to learn from these behaviours. Therefore, the process is not just technological or psychological; rather, it is a feedback loop between platform architecture and user behaviour (Metzler et al., 2023).

This feedback loop makes algorithmic polarisation particularly difficult to mitigate. If policymakers focus only on platform systems, they may overlook the individual’s role of selective exposure and partisan identity. However, if they specifically focus on individual users, they overlook the mechanisms that platforms use to design predictions and monetise attention. The problem is not that users are passive victims of the algorithm, nor that algorithms are harmless reflections of user preferences. The central dilemma is the interaction between the two, and the current relationship is polarising political perspectives.

Evidence of Algorithmic Influence

Recent research contests the claim that algorithms solely reflect user preferences. A 2026 Nature study on X examines the political effects of the platform’s feed algorithm by randomly assigning active US-based users to either an algorithmic,or a chronological feed. Over seven weeks, the study found that turning on X’s algorithmic feed substantially shifted political attitudes towards more conservative opinions on policy and current affairs, though it did not majorly impact partisanship or polarisation (Gauthier et al., 2026). The algorithm further increased engagement on the platform and promoted more content from conservative activist accounts, meanwhile reducing the visibility of traditional news sources (Gauthier et al., 2026).

The finding reflects that algorithmic ranking holds more powerful roles than just organising content; it also actively influences political attitudes by determining what information users repeatedly encode. Switching from an algorithmic to a chronological feed did not produce a comparable reversal, suggesting that prior algorithmic exposure leaves lasting imprints on users' habits and beliefs.

Research on TikTok presents similar findings. A large-scale audit of TikTok during the 2024 U.S. election used hundreds of “sock puppet” accounts to gain insight into how the platform’s "for you page” recommended partisan political content (Ibrahim et al., 2026). This study identified systematic imbalances in political exposure: Republican- acqua received more ideologically aligned recommendations than Democratic ones, while Democratic accounts were exposed to more opposite-party recommendations on average. These disparities cause deep concern about platform neutrality during election periods and its power to sway voters (Ibrahim et al., 2026).

A study on Facebook and Instagram on the 2020 election revealed that replacing algorithmic feeds with reverse chronological feeds altered the content users experienced. Although there was a decline in participation, the study found no significant shifts in affective personalization or political opinions during the study period (Guess et al., 2023).

When taken together, these studies suggest a more precise conclusion: social media algorithms are not the sole cause of US political polarisation, but they can absolutely shape the informational filters through which polarisation develops, deepens, and becomes politically influential.

Affective Polarisation and Political Hostility

Perhaps the most consequential form of polarisation in the U.S. does not arise from policy disagreement alone but from affective polarisation which refers to the growing tendency for citizens to view differing perspectives as illegitimate, wrong, and immoral. That’s where the algorithmic ranking becomes especially significant. Content that provokes fear, outrage, hostility, or moral indignation often performs strongly across engagement metrics. Political posts that are accusatory, inflammatory, and emotionally charged may often generate more attention than posts explaining institutional processes or policy trade-offs (Brady et al., 2021).

Piccardi et al. (2025) tested whether changing the amount of anti-democratic attitudes and partisan-hostile content in user’s X feeds affected their feelings towards political opponents. Using a browser extension to rerank users’ feeds, researchers found that decreased exposure to this type of content reduced hostile feelings.

The reranking of exposure to this content significantly influenced affective polarisation, leading to users experiencing heightened emotions towards political outgroups, while reduced exposure resulted in users feeling more positively towards the opposition. The effects of affective polarisation was resemblant in both in-feed experiences and post-experience surveys. Piccardi et al. (2025) referenced this change is comparable in size to 3 years of change in the United States. These findings suggest that affective polarisation is directly linked to algorithmic design. If content can be intensified and reduced through feed ranking, then a solution lies in algorithmic intervention itself (Piccardi et al. 2025).

Human Rights Implications

The human rights concern is not only that algorithms may influence political opinion but also that they can distribute information unequally. The right to share and interact with information is recognised by international human rights law and is fundamental for just democratic participation (OHCHR, 2024). If algorithmic systems constantly prioritise certain viewpoints while also diverting exposure to others, users may not encounter the range of information necessary to develop an independent opinion on political matters. Research further supports that algorithmic ranking leads to reproduction of existing inequalities by reinforcing hierarchies and reducing interaction with diverse viewpoints (Ahmmad, 2025).

This is especially significant in the U.S. because political participation is closely aligned with digital news, social media, and access to electoral debate. If misleading or hostile partisan perspectives are dominantly promoted and repeatedly prioritised online, citizens may make political conclusions from incomplete and unbalanced information. Disinformation does not just cause individual misunderstanding but can shape electoral participation, lower trust in journalism, and compromise productive public communication (Kim, 2026; Chavda, 2025; Kofi Annan Foundation, 2023).

Another issue arises concerning freedom of expression. When feed ranking prefers particular identities, others are pushed down to make room. In this regard, expression in the digital sphere is shaped by what is buried and what is amplified. In the U.S. context, this gives private platforms significant power in controlling the political narrative they choose, a power that could influence elections (Ibrahim et al., 2026), and the future of U.S. politics. A journalist, activist, or marginalised demographic may still be permitted to speak; however if their content is not made publicly discoverable, they cannot participate in political discussions fairly. As a result, a gap emerges between the right to express oneself and the right to receive fair hearing. (Article 19, 2023; Schaffner et al., 2024).

Disinformation widens the human rights concern even further. Sensationalist content, meaning content that is designed to evoke reaction, can spread quickly across platforms that reward engagement, even if it is untrue. The Kofi Annan Foundation says that social media can heighten extreme content on user feeds, which adds to a sustained hostile rhetoric and production of dangerous ideologies (Kofi Annan Foundation, 2023). In a democratic society the detriment of misinformation travels beyond just public debate; it could undermine elections and generate hostility towards political oppositions, making democratic participation less authentic.

Therefore, the major human rights consequence is that algorithmic bias can damage public access to balanced public debate and distort the conditions that people rely on to exercise their democratic rights. Citizens still have the right to speak, share, vote, and access information, but these rights are compromised as the political environment becomes increasingly shaped by algorithms designed by private platforms who share their own interests rather than public consensus (OHCHR, 2024).

Democratic Participation and Inequality

Democratic participation depends on multiple variables besides voting. It requires access to information, balanced exposure to opposing ideologies, and the ability to deliberate in a shared public sphere. Algorithmic polarisation undermines these conditions by fragmenting political discourse into filtered categories of content. This asymmetry jeopardises not just electoral outcomes but raises questions about the ways users encounter electoral information. If recommendation systems consistently promote misleading, partisan content then this could impact how users view candidates, parties, and the act of voting itself (Ibrahim et al.,2026; Kim, 2026).

Once users encounter politics primarily through online feeds designed to maximise engagement, public debate suffers. It can become more reactive, emotionally charged, and hostile. Complex political issues are flattened into monetised quick-form content, designed to provoke rather than inform. Political opponents are presented as enemies, and trust in institutions like journalism erodes when users are repeatedly shown content that confirms their suspicions, regardless of accuracy.

This does not mean social media should be a scapegoat for all polarisation in the United States. Structural inequalities, declining media trust, and entrenched electoral division all contribute to polarisation. However, social media platforms increasingly mediate how political conflicts are presented at a mass scale (Gauthier et al., 2026; Ibrahim et al., 2026; Guess et al., 2023). Algorithms do not create political tension from nothing. Instead, they amplify existing divisions and make them harder to bridge.

Addressing Algorithmic Polarisation

Addressing this polarisation requires more than simply asking platforms to remove harmful content. Blanket moderation can threaten legitimate expression, while a lack of regulation can allow harassment, hate speech, and misinformation to spread. The challenge is to regulate algorithmic visibility without compromising democratic speech as a matter of state or corporate control.

One response is greater algorithmic transparency. Platforms should be required to provide clear information about how their recommendation systems rank political content and what factors determine the promotion or demotion of certain narratives. Transparency alone is insufficient, but without it, researchers, regulators, and users cannot assess whether platforms are causing structural democratic harm.

A second response is independent auditing. These platforms control access to their own data; therefore, external researchers often struggle to study the full extent of algorithmic effects. Public interest audits could help identify how algorithms systematically amplify extreme content, misinformation, and unequal political exposure (Meßmer & Degeling, 2023). The European Union’s Digital Service Act (DSA) office offers one model, requiring large platforms to assess the algorithmic risks and improve transparency around their recommendation systems (European Commission, 2026; Regulation (EU) 2022/2065, 2022). While the U.S. does not have an equivalent framework, the DSA demonstrates how algorithmic accountability can be done with transparency and thorough risk assessment. Without a comparable legislative framework, the U.S. leaves equal information accessibility largely dependent on voluntary regulation and policy by private platforms. Advocacy for laws on transparency and independent audits is an essential next step for combatting political polarisation in the U.S.

A third response is user agency, meaning the ability of users to control what political content they are exposed to. Platforms could offer users more meaningful control over how political content is categorised and ordered, including options for chronological feeds and reducing political recommendations. These options should not be buried in obscure settings but offered transparently as a way to balance exposure. Users should be able to access practical ways to mitigate what political information they see (Article 27; DSA Observatory, 2025).

Finally, increased algorithmic literacy remains necessary, though it cannot be treated as a substitute for platform accountability. Algorithmic literacy refers to the understanding of how algorithms work and their biases. Social platforms should provide their users with the necessary information to identify manipulative content, verify the sources, and analyze the impact of engagement metrics on the content they view. However, education on algorithms alone will not solve the problem created by concentrated platform power. Users can be encouraged to think critically and learn about algorithms, but the main responsibility lies with the platforms that design and profiteer off these recommendation systems.

Conclusion

Social media algorithms are powerful architects of the contemporary information landscape. By personalising content to maximise engagement and deliberately steering users toward specific ideologies, they have the power to shape what narratives are heard and which are silenced. The evidence shows that social media algorithms do not solely create political polarisation in the U.S. but can further shift policy preferences and shape attitudes towards political debates as well. The result: an asymmetric promotion of partisan material inadvertently interacts with user behaviour, cementing echo chambers and feedback loops.

The key concern is not whether Americans are free to express their political opinions online; it is who controls the mechanism that determines which political opinions are made public. In an age where democratic discourse is increasingly validated or silenced through private platforms, algorithmic power cannot be underestimated and must be treated as an imminent human rights concern. Addressing this issue will require more than individual media coverage, literacy, research studies, and voluntary platform acknowledgment. It will need absolute transparency, independent auditing, and serious pushes for legislative reform to hold private platforms liable for their role in unjustly shaping public discourse.

Glossary

Affective Polarisation: A form of political division where people feel strong dislike or hostility toward those with opposing political views.
Algorithm: A step-by-step procedure or set of rules used to solve a problem or complete a task. In social media, algorithms are used to sort, rank, and recommend content.
Algorithmic Curation: The process by which platforms select, organise, rank, and recommend content to users through automated systems.
Algorithmic Literacy: Knowledge of how algorithms work and their potential biases.
Algorithmic Ranking: The ordering of posts, videos, or recommendations depending on what a platform predicts a user is most likely to engage with.
Algorithmic Transparency: Clear information about how platforms rank, recommend, promote, or reduce the visibility of content.
Algorithmic Visibility: The level of attention or exposure content receives because of how a platform ranks or recommends it.
Chronological Feed: A feed that shows posts in the order they were published, rather than according to predicted relevance or engagement.
Confirmation bias: people’s tendency to process information by looking for, or interpreting, information that is consistent with their existing beliefs.
Content Moderation: The process by which platforms remove, restrict, label, or reduce the visibility of content that breaks their rules or causes harm.
Democratic Participation: The ability of citizens to take part in political life, including voting, public debate, civic engagement, and access to political information.
Digital Services Act: A European Union law that places rules on large online platforms, including requirements around transparency, risk assessment, and recommender systems.
Disinformation: False or misleading information that is deliberately created or shared to deceive or manipulate people.
Echo Chamber: A social environment where people mainly encounter views that confirm their existing beliefs and avoid opposing perspectives.
Engagement Metrics: Data showing how users interact with content, such as likes, comments, shares, clicks, watch time, and reposts.
Feedback Loop: a process in which information about the output of a system, process, or activity is returned to the input to modify or reinforce the actions being taken.
Filter Bubble: A personalised information environment where algorithms repeatedly show users content similar to what they have already engaged with.
Freedom of Expression: The right to hold opinions and to seek, receive, and share information and ideas.
Independent Audit: An external review of a platform or algorithm to assess how it works and whether it causes harm.
Information Exposure: The political information, viewpoints, and sources that users come across online.
Personalisation: The process of tailoring content to individual users based on their past activity, interests, and predicted preferences.
Platform Neutrality: The idea that platforms treat content, users, or political viewpoints equally, without bias. This is often questioned when algorithms appear to favour certain content or perspectives.
Political Polarisation: The widening division between political groups, especially when disagreement becomes tied to identity, distrust, or hostility.
Reaffirming phenomena: Patterns where users repeatedly encounter information that confirms their existing beliefs, making those beliefs feel stronger or more widely accepted.
Recommendation Systems: A system used by platforms to suggest posts, videos, accounts, or other content to users.
Shared Public Sphere: A space in which citizens come across common information, debate public issues, and take part in democratic life.
Sock Puppet Account: A fake or controlled account used for research purposes to test what content a platform recommends to different types of users.
User Agency: The ability users have inside digital apps to control or influence what they see online, including options such as chronological feeds or reduced political recommendations.

References

ARTICLE 19. Content Moderation and Freedom of Expression Handbook. 2023.
Ahmmad, Mukhtar, et al. “Trap of Social Media Algorithms: A Systematic Review of Research on Filter Bubbles, Echo Chambers, and Their Impact on Youth.” Societies, vol. 15, no. 11, 30 Oct. 2025, p. 301, https://doi.org/10.3390/soc15110301.
Brady, William J., et al. “How Social Learning Amplifies Moral Outrage Expression in Online Social Networks.” Science Advances, vol. 7, no. 33, Aug. 2021, https://doi.org/10.1126/sciadv.abe5641.
Cooper, Alice and Peter Chapman. “Making Recommender Systems Work for People: Turning the DSA’s Potential into Practice - DSA Observatory.” DSA Observatory , Knight-Georgetown Institute, 19 May 2025, dsa-observatory.eu/2025/05/19/making-recommender-systems-work-for-people/.
Chavda, Janakee. “How Americans’ Trust in Information from News Organizations and Social Media Sites Has Changed over Time.” Pew Research Center, 29 Oct. 2025, https://www.pewresearch.org/short-reads/2025/10/29/how-americans-trust-in-information-from-news-organizations-and-social-media-sites-has-changed-over-time/.
DSA Observatory. “Making Recommender Systems Work for People: Turning the DSA’s Potential into Practice - DSA Observatory.” DSA Observatory - a Hub of Expertise on the DSA Package., 19 May 2025, https://dsa-observatory.eu/2025/05/19/making-recommender-systems-work-for-people/.
Egan, Jim. “Overview and Key Findings of the 2026 Digital News Report.” Reuters Institute for the Study of Journalism, Reuters Institute, 16 June 2026, https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/dnr-executive-summary.
European Commission. “The Digital Services Act.” Shaping Europe’s Digital Future, 2022, digital-strategy.ec.europa.eu/en/policies/digital-services-act.
Gauthier, Germain, et al. “The Political Effects of X’s Feed Algorithm.” Nature, 18 Feb. 2026, www.nature.com/articles/s41586-026-10098-2,
Guess, Andrew M, et al. “How Do Social Media Feed Algorithms Affect Attitudes and Behavior in an Election Campaign?” Science, vol. 381, no. 6656, 28 July 2023, pp. 398–404, www.science.org/doi/10.1126/science.abp9364, https://doi.org/10.1126/science.abp9364.
Ibrahim, Hazem, et al. “Systematic Partisan Content Skews in TikTok during the 2024 US Elections.” Nature, 28 May 2026, https://doi.org/10.1038/s41586-026-10447-1.
Kim, Young Mie, et al. “Targeted Digital Voter Suppression Efforts Likely Decrease Voter Turnout.” Proceedings of the National Academy of Sciences, vol. 123, no. 5, 26 Jan. 2026, 10.1073/pnas.2519944123.
Kofi Annan Foundation. “Navigating the Age of Polarization: How Social Media Fuels Divides.” Kofi Annan Foundation, 15 Sept. 2023,
www.kofiannanfoundation.org/news/navigating-the-age-of-polarization-how-social-media-fuels-divides/.
Liedke, Jacob, and Christopher St. Aubin. “Social Media and News Fact Sheet.” Pew Research Center, 17 Sept. 2024, www.pewresearch.org/journalism/fact-sheet/social-media-and-news-fact-sheet/..
Meßmer, Anna-Katharina, and Martin Degeling. “Auditing Recommender Systems – Putting the DSA into Practice with a Risk-Scenario-Based Approach.” https://arxiv.org/abs/2302.04556.
Metzler, Hannah, and David Garcia. “Social Drivers and Algorithmic Mechanisms on Digital Media.” Perspectives on Psychological Science, vol. 19, no. 5, 19 July 2023, pp. 735–748, journals.sagepub.com/doi/full/10.1177/17456916231185057, https://doi.org/10.1177/17456916231185057.
Milli, Smitha, et al. “Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media.” PNAS Nexus, vol. 4, no. 3, 5 Mar. 2025,, https://doi.org/10.1093/pnasnexus/pgaf062.
OHCHR. “International Covenant on Civil and Political Rights.” OHCHR, 2024, https://www.ohchr.org/en/instruments-mechanisms/instruments/international-covenant-civil-and-political-rights?utm.
Yuan, Ni Annie, and Ho-chun Herbert Chang. “Motivation, Attention, and Visual Platform Design: How Moral Contagions Spread on TikTok and Instagram in the 2024 United States Presidential Election.” ArXiv, Computers and Society, 2024, arxiv.org/abs/2602.02479.
Schaffner, Brennan, et al. “‘Community Guidelines Make This the Best Party on the Internet’: An In-Depth Study of Online Platforms’ Content Moderation Policies.” arXiv (Cornell University), 11 May 2024, 10.1145/3613904.3642333.
Tiziano Piccardi, et al. “Reranking Partisan Animosity in Algorithmic Social Media Feeds Alters Affective Polarization.” Science, vol. 390, no. 6776, 27 Nov. 2025, 10.1126/science.adu5584.
Yuan, Ni Annie, and Ho-chun Herbert Chang. “Motivation, Attention, and Visual Platform Design: How Moral Contagions Spread on TikTok and Instagram in the 2024 United States Presidential Election.” arXiv, Computers and Society, 2024, https://arxiv.org/abs/2602.02479.

HRRC accepts NO funding from governments for our research or campaigns. At every step of the way, our impact is made possible because of people like YOU. Make your contribution now!