In contemporary democracies, the right to free speech is a foundational safeguard of political legitimacy and public participation. Yet, in the digital age, this right is increasingly mediated by opaque and profit-driven algorithmic systems that determine what speech is visible, what circulates and what is ultimately suppressed. As social media platforms become the primary forums for discourse, the underlying logics of algorithmic content curation – designed to maximise engagement rather than uphold democratic values – are quietly reshaping who gets heard and on what terms. This transformation is not merely a matter of changing communication technologies; it marks a profound shift in the structure of the political and public sphere and the conditions under which speech can function as a democratic tool.

This article argues that algorithmic content curation on social media platforms undermines the legitimacy of democratic discourse, particularly for marginalised speech that serves to challenge dominant norms. This gives rise to both epistemic and political harms. Epistemically, it facilitates epistemic injustice by privileging popular dis- and misinformation while suppressing the capacity of marginalised groups to contribute to and rectify shared understanding. Politically, it erodes the legitimacy of democratic decision-making by undermining the principle of equal participation, excluding marginalised voices from the processes that give majority mandates their political weight.

Although ‘democracy’ is a broad concept, this article adopts a flexible definition to reflect the global reach of social media and its effects across diverse democratic contexts. Rather than committing to one specific model – liberal, deliberative or participatory – it focuses on three widely recognised democratic principles: majority rule, fundamental rights and legitimate decision-making (Barroso & Van Brussel Barroso 2023). Majority rule enables collective decisions; rights protect individuals and marginalised groups from majoritarian overreach; and legitimacy depends on inclusive public discourse through which decisions are publicly justified. This framework, drawing from both liberal and deliberative traditions, highlights how democratic legitimacy requires more than formal procedures. It depends on the ability of all citizens, especially marginalised citizens, to participate meaningfully in public debate. It is from this perspective that the article examines how algorithmic curation disrupts the conditions for democratic speech.

For the purposes of this article, ‘marginalised’ is used rather liberally to refer to one of potentially three things. These are numerical minorities, politically disadvantaged groups and economically marginalised communities. The reason for this is twofold. Firstly, the purpose of the article is to identify the impact of algorithmic content curation on those latter two groups who, as the article will explain, rely on the democratic function of free speech in bringing about legal protection for their moral pursuits. These groups are marginalised in the political sense because of the baseline establishment of their lack of legal equality and protection. Secondly, the result of majority rule is that it establishes numerical minorities which may or may not also reflect the population of those politically disadvantaged groups. Democratic outcomes cannot be taken to be reflective of very particular political questions; however, where legal protection cannot be achieved through these democratic processes, the reality for those groups or communities is nonetheless numerical.

This article is divided into four parts. Part one establishes the conceptual framework for understanding the value of speech and public discourse. Drawing on political and communication theories of political and free speech, it illustrates the importance of public discourse within the broader context of democratic principles, particularly focusing on the role and inclusion of marginalised groups. It examines how speech functions within a democracy and explores the benefits of equal participation in democratic decision-making processes for marginalised groups.

Part two examines how social media platforms use algorithms to curate content and structure information flow. It explains how these systems prioritise engagement over democratic values, often leading to the formation of isolated ‘echo chambers’ that marginalise counterspeech and limit exposure to diverse perspectives. By contrasting this with the assumptions underlying democratic theories of public discourse, the section shows how algorithmic curation has reshaped – and ultimately disrupted – the conditions necessary for inclusive and reasoned debate.

Part three draws on recent examples to illustrate how algorithms have suppressed marginalised voices. It considers the shadowbanning of Black Lives Matter content on Instagram and TikTok, Meta’s suppression of pro-Palestinian speech and Facebook’s promotion of anti-Rohingya hatred. These cases are analysed through the lens of democratic principles, highlighting how algorithmic interventions can distort public discourse, marginalise minority participation and ultimately undermine the legitimacy of democratic outcomes. In doing so, it considers the empirical and normative implications of algorithmic curation for marginalised groups.

Part four reflects on broader normative concerns about the transformation of public discourse by social media algorithms. It asks whether contemporary platforms can meaningfully support deliberative democracy, and whether the ideal of public discourse can survive within an attention-driven economy. It also considers the future of democratic communication – particularly for marginalised groups – when social media grants unprecedented visibility to diverse and often conflicting moral perspectives.

Harari (2024:62) emphasises that a defining characteristic of democracy is its recognition of the fallibility of both government and its citizens. Whereas authoritarian regimes centralise information flow in order to project an image that leadership is always correct, democratic regimes are built on the assumption of fallibility. This means that leaders (and the voting public) can make mistakes, and systems need checks and balances to correct them (Harari 2024:62–64). Democracies are therefore sustained by self-correcting mechanisms to rectify the errors of those in power. These mechanisms include regular elections, the protection of press freedom and the separation of executive, legislative and judicial organs of the state (Abromeit & Stoiber 2007). To inform these mechanisms, information flow is decentralised and democratic societies are connected by a wide array of independent and interlinked ‘nodes’ through which people can communicate freely and directly. These include legislative bodies, opposition political parties, the media, corporations, local communities, non-government organisations and individual citizens (Harari 2024:62–63). This networked structure enables public discourse through which citizens can engage with, challenge and mediate tensions with one another.

The idea that free speech and public discourse perform a vital function in democratic societies is well established in both political and speech theory (Dworkin 2009; Habermas 1996; Rawls 1996). Whitten (2022), for instance, offers an instrumentalist account, arguing that free speech and public discourse support democracy by promoting what values a democracy is meant to realise. If democracy’s purpose is understood as fostering a state of non-domination among relationally egalitarian citizens, then free speech can be defended on the grounds that it facilitates precisely that kind of relationship. Another line of reasoning holds that free speech and public discourse are not merely supportive of democracy’s aims but constitutive of democracy itself. On this view, expressed by theorists such as Gutmann and Thompson (2008), democratic legitimacy depends not solely on the act of voting, but on an ongoing process of open deliberation among free and equal citizens. Here, majority rule is legitimised only insofar as it is embedded in a broader commitment to transparent, inclusive dialogue (Dworkin 1977). As Habermas (1996) argues, when the public sphere allows for an unrestricted flow of information and argument, the best ideas can be tested and prevail, thereby grounding democratic decisions in rational deliberation.

These arguments help clarify the distinction between majority rule as a procedural mechanism and the broader normative principles that underpin democratic legitimacy. While majority rule is frequently identified as a defining feature of democracy, it does not, in itself, constitute the entirety of democratic governance. Rather, as Ober (2007:2) explains, democracy is fundamentally about the capacity of its citizens to act – to deliberate, to participate and to vote, capacities that are safeguarded by core civil and political rights, including the freedoms of speech and assembly (Dworkin 1977). Consider, for instance, a scenario in which a majority uses its power to revoke the voting rights of a dissenting minority. In such a case, although majority rule has technically been exercised, its democratic character collapses, as the essential safeguards that constrain and legitimise majority power are undermined. Rights such as voting, freedom of the press and freedom of assembly function as critical self-correcting mechanisms, enabling citizens to challenge and rectify errors – whether made by elected officials or fellow citizens.

It is in this context that Barroso and Van Brussel Barroso (2023:54) argue that democracy rests on a triad of ‘votes, rights, and reasons’. Votes represent the procedural dimension of democracy, enabling collective decision-making through majority rule; rights provide substantive guarantees that protect individuals regardless of electoral outcomes; and reasons, expressed through free and open public discourse, confer legitimacy on democratic decisions by anchoring them in justification rather than mere aggregation. Therefore, the restriction on public discourse, free speech and information flow would constitute ‘at least a partial suspension of democracy’ (Rawls 1996:254) that resembles the centralisation of information flows of authoritarian regimes.

While public discourse is often celebrated as a means of legitimising majority rule, the inescapable reality is that majority decisions cannot yield equal outcomes for all citizens. This is a point of tension between theorists with respect to defining the legitimacy of majority rule. Waldron (2006), for example, argues that this inequality is acceptable provided the processes that produce these outcomes are procedurally legitimate, such as through mechanisms like public discourse. From this perspective, the legitimacy of democratic outcomes hinges on the integrity of fundamental rights, such as the right to vote, as conditions enabling genuine deliberation. Waldron thus defends fundamental rights not because they ensure substantive success for marginalised groups in achieving legal protection for their own moral interests, but because they ensure equal participation in the democratic process. In contrast, Dworkin (2006) argues that procedural legitimacy alone is insufficient. He maintains that marginalised groups must be entitled to private spheres in which they can pursue their moral convictions independently of majority approval. Although a full analysis of this debate lies beyond the scope of this article, it underscores an important tension: public discourse and free speech, while essential, do not, by themselves, provide a clear pathway to the recognition or creation of more specific rights. Nor do fundamental rights, despite their contribution to the triad of democratic principles, necessarily translate into more precise civil equalities. This is thus a distinction between procedural and substantive definitions of legitimacy (Cohen 1996).

This is not to suggest that public discourse and free speech are without value for marginalised groups. On the contrary, they can serve as powerful tools for advancing minority interests; however, this possibility must be understood through the lens of procedural legitimacy. It is crucial to recognise that majority rule does not refer to static, demographically fixed groups bound by permanent political alignments. Public opinion is fluid: individuals can change their views over time, and they may choose to support policies or rights from which they do not personally benefit, but which they believe are just or beneficial to others. In this way, public discourse creates the conditions for persuasion, solidarity and moral progress, allowing minority positions to gain broader traction within the democratic process. The ‘majority’ therefore does not refer to a resistant, oppositional group, but rather a political mandate determined by collective action. The premise of collective intelligence is closely linked to the marketplace of ideas theory of speech protection (Blasi 1977; Post 1995; Schauer 1982). While the theory is primarily concerned with the pursuit of truth (and elections, as Harari (2024:65) notes, are not fundamentally about the pursuit of truth), the marketplace of ideas nonetheless promotes the idea that speech protections can empower marginalised groups within democratic systems (Blasi 2004). In the marketplace model, expression is shielded from government interference such as regulation, censorship or punishment. The underlying assumption is that, through open and transparent competition, society can distinguish between strong and weak ideas. Robust arguments gain traction and shape public consensus, while weaker ones are discredited and fade from influence. In this way, the marketplace facilitates not only epistemic progress but also political inclusivity by allowing marginalised perspectives to challenge dominant views.

This model of speech protection finds strong support in political philosophy. For example, Habermas (1996) conceptualises the public sphere as a domain of social life in which individuals act as citizens rather than private persons, and participate in rational-critical debate to shape public opinion. Within this sphere, discourse is governed by principles of equality, openness and reason, where arguments are assessed on their intrinsic merits rather than on the authority, status or identity of the speaker. The legitimacy of the public sphere depends on the freedom and equality of participants, fostering a deliberative environment where ideas can be contested and collective will can emerge through reasoned discourse.

The marketplace of ideas therefore depends on the existence of counterspeech. Counterspeech operates as a response to contested ideas, enabling individuals to challenge their validity, present alternatives and offer the public a broader range of options from which to decide what they find most persuasive (Nunziato 2019). In this way, counterspeech serves as the primary remedy for harmful or ‘bad’ ideas, making the marketplace a self-correcting mechanism for the errors of government or fellow citizens. It supports a democratic ideal of self-determination by ensuring that no authority monopolises the process of truth-seeking. The marketplace of ideas therefore helps safeguard the conditions for democratic engagement, namely, ‘public debate, open dialogue, and the foundations of democracy itself’ (Parsons 2020:2159).

While the metaphor of the marketplace is often described as a mechanism for separating good ideas from bad ones, its true democratic value lies in its commitment to allowing all ideas to be heard. The freedom to agree or disagree with those ideas – rather than their immediate popularity or correctness – is what gives the marketplace its normative strength as a safeguard against censorship. Majority decisions, particularly when formalised through voting, may have adverse consequences for marginalised groups if dominant groups are unwilling to support their interests. Yet the marketplace is not a metaphor for elections, but for a free and dynamic exchange of ideas. The fact that an idea begins as unpopular or marginal does not preclude it from gaining traction over time. In this regard, the enduring freedom to pursue an idea is arguably more consequential than the freedom to denounce it, especially in the long term. History demonstrates that persistence in advocating for minority rights has often succeeded when the marketplace is allowed to operate freely over time. For example, consider the gay rights movement in the United Kingdom. Over the past 60 years, the law has shifted from criminalising homosexual activity to recognising same-sex marriage. This transformation was not the result of moral self-reflection by opponents, but of sustained activism by LGBTQ+ individuals and allies who challenged dominant norms through counterspeech. A BBC News review highlighted that between 1983 and 2016, the proportion of people who believed same-sex relationships were ‘not wrong at all’ rose from 17% to 64% (Schraer & D’Urso 2017). Marriage equality emerged from this shift in public and political attitudes, driven by growing societal support and global momentum for reform. In short, marriage equality was achieved through a democratic process shaped by the dynamics of the marketplace of ideas.

While the success of movements like marriage equality highlights the potential of public discourse to advance the rights of marginalised groups, it is difficult to ignore that the marketplace of ideas rests on an idealised vision of collective intelligence (Brietzke 1996; Ho & Schauer 2015). The hope that marginalised groups can build support through open dialogue depends on several fragile assumptions, namely that participants engage in good faith, are receptive to opposing views and are free from entrenched bias. The fact that democracy assumes fallibility ironically means that resistance to new ideas may also be the result of fallibility. Indeed, paradoxically, when democracy functions properly – where all voices are included – it often becomes harder to reach consensus (Dahl 1989; Fung 2003). Greater inclusion brings a wider array of groups, perspectives and competing interests into public discourse. For example, in the 1820s United States, more people were enslaved than there were eligible white male voters – the only group permitted to participate politically. At that time, policies reinforcing gender roles and racial inequality were easily enacted, largely because public discourse was confined to a narrow segment of society. For example, since the abolition of slavery, women’s suffrage and the civil rights movement, the United States has faced greater difficulty in forging broad agreement, as a more inclusive marketplace of ideas has introduced complex and often conflicting claims on justice and policy (Harari 2024:66, 86–87).

At the other end of this issue, Fraser (1992) contends that without participatory parity, speech cannot be considered truly free unless individuals have the conditions necessary to be genuinely heard. This requires dismantling structural barriers – such as institutional discrimination, social exclusion and material inequality – that constrain the discursive power of marginalised groups. While the marketplace of ideas idealises a level playing field, in reality such conditions can only be approached by first leveraging that very marketplace to advocate for structural reform. In this sense, the process is yet again paradoxical: the marketplace must be used to push for the changes required to make it function as intended.

Nevertheless, the enduring strength of the marketplace of ideas, as well as arguments concerned with the value of public discourse and free speech, lies not in their empirical perfection but in their normative ambition that embraces the fallibility of government and its citizens. The marketplace of ideas offers visions of a political order where error can be acknowledged, power can be challenged and even marginal voices can shift the trajectory of public discourse. Recognising the limitations of public discourse does not undermine its value. It invites a more critical, inclusive and realistic understanding of the conditions under which democracy might thrive. In an age of algorithmic content curation and fragmented digital communities, the challenge is not to discard these ideals but to improve their realisation by addressing the inequalities that distort speech, access and participation.

Social media initially appeared to offer a powerful remedy to some of the limitations of the marketplace of ideas, particularly the structural barriers that have historically excluded marginalised voices from public discourse. By lowering the cost of entry to mass communication, social media platforms promised to democratise access to speech, allowing individuals to bypass traditional gatekeepers such as mainstream media, political elites or academic institutions (Magarian 2021). In theory, this technological shift offered a level of discursive equality that more closely aligned with the ideal of participatory parity described by Fraser (1992). For historically silenced or underrepresented groups, social media created new opportunities to be heard, to organise and to challenge dominant narratives through empowered and transparent counterspeech (Joseph 2012; Karami et al. 2020; Okonkwo et al. 2024). Movements like the Arab Spring were emblematic of this potential: grassroots campaigns that gained traction, not through institutional endorsement but by mobilising public support across decentralised networks of individuals. The Arab Spring was a series of pro-democracy anti-government protests that spread across much of the Arab world in the early 2010s in response to political corruption and economic stagnation. Although the precise role of social media in the outcome of the initial protests has been debated (Alkhouja 2015), social media platforms like Facebook and YouTube undoubtedly played a significant role in facilitating communication among participants of political protests and raising global awareness for the purpose of the protests (Salem & Mourtada 2011).

Furthermore, social media’s architecture – based on virality, interactivity and asymmetrical discourse – offered a new mechanism for public deliberation that was faster, more participatory and less hierarchical than traditional models of discourse, and opened up opportunities for new and marginalised ideas to gain traction (Joseph 2012). In this sense, social media held the promise of realising a more inclusive marketplace of ideas by giving marginalised communities platforms to amplify their perspectives and demand political and cultural recognition (Zimmer et al. 2019). However, as subsequent developments have shown, this promise was neither universally fulfilled nor without consequence.

Arguably, the most significant oversight in early optimism about social media’s potential as a marketplace of ideas was the failure to recognise that these platforms are fundamentally profit-driven businesses (Kim 2017). Their financial model is built on maximising user engagement, a metric encompassing actions such as viewing, liking, commenting, sharing and saving content. The longer users remain engaged, the more advertising they are exposed to, increasing the platform’s value to advertisers. As advertisers seek the highest number of impressions possible, platforms are incentivised to design algorithms that prioritise content likely to capture attention and sustain user activity. In essence, the commercial logic of social media monetises attention, often to the tune of billions in revenue (Ross 2024), and this creates a structural misalignment between democratic ideals of discourse and the economic interests of the platforms that host it.

In the early 2010s, where movements like the Arab Spring brought about some success, social media platforms were dominated by chronological feeds, making it easy for activists to coordinate and go viral without platform interference. Early optimism was justified because moderation was minimal, meaning minimal algorithmic downranking. This also meant that most users saw the same posts in the same order, enabling mass visibility. This began to shift around 2013, where platforms sought to maximise user engagement by relying more heavily on algorithmic content curation designed to optimise the user experience (Berman & Katona 2020). These algorithms analyse vast amounts of impression data to predict what individual users are most likely to find interesting, including those ‘passive’ users, who make up approximately 90% of social media audiences. Although users may not interact directly with content, their screen time and scrolling behaviour provide crucial signals of interest (Kim 2017:147–151). For example, Instagram’s ranking algorithm was introduced to ensure users ‘see the moments [they] care about first’, responding to the platform’s claim that 70% of content was previously going unnoticed by users (Instagram 2016). While this move ostensibly aimed to improve user satisfaction, it also served to increase the likelihood that users would return, thus driving up ad impressions and, in turn, revenue. Similar systems exist across platforms: TikTok and X (formerly Twitter) deploy ‘For You’ feeds that deliver content algorithmically aligned with prior engagement, while YouTube’s autoplay feature queues videos based on a user’s watch history. These techniques do not simply respond to user preferences. Rather, they actively shape them and reinforce patterns of consumption that serve platform profitability over democratic discourse.

While mass media and the press have long been credited as essential ‘nodes’ in the flow of information, social media platforms, which have succeeded traditional newspapers as a more popular source of news, differ significantly in their orientation towards truth, accuracy and credibility. Despite their own ideological biases, newspapers historically derived their authority from journalistic norms and a readership that regarded them as credible sources of news (Anderson 2008). Their financial viability depended on maintaining public trust and attracting a loyal audience willing to return. In contrast, the rise of social media marked a shift from a world of information scarcity to one of information abundance, giving rise to a new economy built around capturing and monetising user attention. In this environment, attention becomes the most valuable commodity, and social media companies are incentivised not to prioritise reliable information but rather to maximise engagement – even if doing so means amplifying sensational, misleading or polarising content. As a result, social media companies operate under a business model fundamentally misaligned with the epistemic and civic responsibilities historically associated with traditional journalism. Johnson, Norton and Skover (2018) described this bluntly:

Falsity has indeed shown to be popular on social media. One notable example occurred during the women’s boxing tournament at the 2024 Summer Olympic Games. Following Angela Carini’s forfeit just 46 seconds into her second-round match, the eventual winner, Imane Khelif, became the subject of intense online scrutiny and speculation. Social media users, reacting to her previous ineligibility at the 2023 Women’s World Boxing Championships, circulated claims that Khelif was either male or a transgender woman, arguing that she should not have been allowed to compete because of a presumed physical advantage. These claims unfolded within a broader socio-political climate of heightened tensions over transgender rights, particularly the participation of transgender women in women’s sports. The resulting media storm – fuelled by both social and mainstream outlets – reflected a deep polarisation. Despite the fact that Khelif is not transgender, and the International Olympic Committee had confirmed her female identity across all legal and sporting documentation, misinformation spread rapidly. Inflammatory posts – such as J.K. Rowling’s claim that Khelif was ‘enjoying the distress of a woman he’s just punched in the head’ and Elon Musk’s assertion that ‘men don’t belong in women’s sports’ – amplified confusion and hostility (BBC News 2024). The virality of these narratives, rather than verifiable facts, dominated online discourse, illustrating how social media platforms often privilege emotional intensity and ideological outrage over accuracy, particularly on contentious identity-related issues (Brady et al. 2021).

This algorithmic structuring of discourse disproportionately disadvantages marginalised groups, as it shifts attention away from the substance of their moral or legal claims and towards the perceived injustices those claims are thought to inflict on majority groups. In such contexts, the focus is not on the rationale behind why transgender women may seek equal legal recognition and protection, but rather on the controversy surrounding their inclusion. This was true in the case of Khelif’s participation in the 2024 women’s boxing tournament. Here, transgender equality was framed not in terms of justice or inclusion but as an invasive threat to the rights and spaces of biological women. This algorithmic amplification of outrage tends to distort the interests of marginalised communities, casting them in a morally suspect light that does not accurately reflect their intentions. For instance, transgender women are frequently mischaracterised as having antagonistic or dismissive attitudes towards biological women’s rights, when in reality their advocacy is often grounded in appeals to fairness, dignity and safety (Morgenroth, Axt & Westgate 2024). Yet these nuances are lost in the polarised and attention-driven logic of social media platforms, which reward emotive over reasoned discourse. The result is a public discourse in which the moral legitimacy of minority pursuits is overshadowed by reactionary framings, undermining the potential for a rational-critical debate.

But, the challenges of algorithmic content curation extend well beyond the amplification of false or sensational content. Social media platforms do not function as a unified marketplace of ideas but as a fragmented network of ideologically siloed communities – often described as echo chambers or filter bubbles – where users are primarily exposed to content that aligns with their existing beliefs (Nguyen 2020). This fragmentation is driven by algorithms, such as those created for TikTok’s and Twitter’s ‘For You’ feeds, which cluster users based on engagement patterns that often reflect political or ideological preferences. Berman and Katona (2020) refer to this phenomenon as ‘cyber-balkanisation’, in which the digital public sphere splinters into isolated groups. Within these groups, ideas circulate without meaningful opposition, counterspeech is diminished, and artificial polarisation is intensified because there is a lack of exposure to other ideas conducive to constructive dialogue (Garimella et al. 2018; Sheeks 2023). As a result, the kind of open, adversarial exchange that the marketplace of ideas relies on is increasingly absent. Although platforms may host vast user bases, algorithmic sorting creates new structural barriers to public discourse – barriers perhaps even more rigid than those found offline. This dynamic has serious consequences for public discourse, especially for marginalised voices, whose perspectives may be further marginalised in these fragmented digital spaces.

Firstly, echo chambers fundamentally undermine the equal participation that marginalised groups require to succeed in the marketplace of ideas by restricting their visibility and distorting their representation. Democratic legitimacy relies on open dialogue among free and equal citizens, but echo chambers create significant asymmetries in who is heard and how. Marginalised voices, especially those advocating for progressive or contested rights, often find themselves confined to sympathetic communities, isolated from the broader public forums where dominant perspectives are shaped and challenged. This exclusion not only diminishes their ability to engage with wider audiences but also precludes their arguments from being subjected to serious public deliberation, thus reinforcing the very idealistic limitations of public discourse that democracy aims to overcome.

Secondly, and relatedly, echo chambers critically weaken the effectiveness of counterspeech (Napoli 2018). Within opposing ideological echo chambers, minority positions are frequently misrepresented, caricatured or framed as threats to prevailing social norms. For instance, platforms like Facebook, Instagram and Twitter have been shown to host significant volumes of anti-trans content in response to reports of transgender athletes in sports (Miller 2024), including deliberate deadnaming, misgendering, casual use of discriminatory slurs, patronising dismissal of transgender experiences and claims that transgender athletes intentionally participate in sports competitions to exploit perceived physical advantages, thereby undermining the legitimacy and rights of their cisgender peers (ISD 2022). These distortions fuel misunderstanding and deepen social polarisation, further marginalising minority groups from equal participation in discourse. This dynamic is evident when two sides of a political debate come together, such as in social media controversies like the Khelif case, where entrenched misunderstandings prevent constructive dialogue and reconciliation of competing ideas. Counterspeech in such contexts loses its corrective power, not because it is absent, but because it is unable to cross the algorithmically enforced and user-reinforced ideological boundaries. Therefore, echo chambers do more than reduce the effectiveness of minority advocacy. They structurally erode the channels of information flow that are essential for a genuinely democratic public sphere.

Thirdly, and resultantly, echo chambers undermine the democratic principle of fallibility. By filtering out disagreement and reinforcing preexisting beliefs, they foster epistemic closure rather than openness to revision. This disrupts the conditions of public discourse that democracies depend on, namely, the possibility of reasoned deliberation among diverse viewpoints. Without a shared epistemic ground, mutual justification becomes difficult, and legitimacy weakens. Echo chambers thus impair a democracy’s capacity for self-correction: they isolate groups from corrective information, intensify political polarisation and marginalise dissenting voices, all of which entrench epistemic insularity and make it harder for societies to recognise and respond to their own errors and injustices.

Social media algorithms thus pose a direct challenge to democratic principles related to the role of public discourse. Even setting aside concerns about elections, the capacity for constructive political dialogue is severely undermined by algorithms designed to prioritise engagement within an economy centred on capturing and monetising user attention. This issue is amplified by the sheer popularity of social media, which now boast over 5.3 billion users globally, making these media the most widely used form of online communication and source of news. Consequently, the problems caused by these algorithms have far-reaching implications for marginalised groups as they disrupt the democratic mechanisms that assure them equal participation to legitimise the democratic decision-making processes. Given the inevitability of majority rule, the inability to challenge ideas that represent the status quo, to share lived experiences of grounding their moral pursuits in a justified sense and potentially to cultivate support in pursuit of legal protection disenfranchises marginalised groups in a way that perhaps even traditional means of public discourse in the physical world do not.

Not only does the use of algorithms at least inadvertently control the dynamics of public discourse by favouring engagement over the quality of information, but they can be programmed to deliberately favour certain content and, in turn, remove content that the platform does not want users to see. This controversy exists on a sliding scale of censorship. At one end of the spectrum are instances of soft censorship through content filtering that have been introduced above. Here, users are shown only what the algorithm predicts they want to see, effectively hiding content deemed less relevant or interesting to them. While this can serve benign purposes – for example, excluding female modelling content from the feed of a user primarily interested in male sports – it also carries significant political implications regarding access to diverse political speech. A case in point is Twitter’s ‘For You’ feed, introduced under Elon Musk’s leadership. This feature aims to personalise content by recommending posts from accounts users do not follow but are predicted to enjoy. The stated goal is to broaden the user’s experience by exposing them to new content beyond their immediate network. However, the reality is that for users heavily engaged with political content, the algorithm tends to reinforce their existing political interests, populating their feeds with similar viewpoints rather than diverse or opposing perspectives. While one might argue that social media platforms are not obligated to provide neutral political coverage, this supposed expansion paradoxically narrows users’ awareness by limiting exposure to contrasting or conflicting viewpoints.

At the other end of the spectrum lies more overt censorship, where platforms actively remove or restrict content deemed ‘unacceptable’. However, what constitutes ‘unacceptable’ is highly subjective, particularly in the political sphere, as it involves deliberate judgements about the legitimacy of certain ideologies, the social arguments they produce and the degree to which those arguments should be amplified or suppressed. These decisions are inherently perspectival: what one person considers offensive or dangerous, another may view as valid or necessary political expression. As a result, the shaping of political discourse through content moderation can be seen either as a protective measure or as a form of ideological distortion, depending on one’s particular standpoint. Indeed, debates about algorithmic content curation often merge with broader political controversies concerning the use of social media to shape public discourse in line with the preferences of platform owners. This phenomenon became particularly apparent following Elon Musk’s acquisition of Twitter in 2022 (Rohlinger et al. 2023). For instance, a computational analysis published in November 2024 identified signs of algorithmic bias on Twitter favouring Republican supporters (Graham & Andrejevic 2024). Examining over 56 000 posts, researchers noted a marked shift around 13 July 2024 – the same day Musk publicly endorsed Donald Trump in the 2024 US presidential election. After this date, Musk and other Republican commentators experienced increased visibility, with Musk’s posts dominating the feeds of users who did not even follow him. While the researchers cautiously suggested that this ‘could imply an algorithmic adjustment’, official confirmation of this particular example has yet to be provided (Graham & Andrejevic 2024). It has nevertheless been cited as a major factor in Trump’s eventual victory, particularly given that momentum for his second campaign – amid ongoing legal troubles that highlighted concerns about his moral character, and widespread apprehension within the political sphere regarding a return to the rhetoric and policies of his first term – once appeared weak (Reid & Lange 2024).

Twitter is not the only platform to have been susceptible to such criticism. In an analysis of Meta’s platforms since the beginning of the Gaza War, Human Rights Watch and the American Civil Liberties Union documented how pro-Palestinian content was disproportionately flagged or removed under community standards enforcement, often under vague categories like ‘spam’ or ‘hate speech’ (Human Rights Watch 2023). Slogans such as ‘Free Palestine’, ‘Ceasefire Now’, ‘Stop the Genocide’ and ‘From the river to the sea, Palestine will be free’ were repeatedly targeted, and in dozens of cases, content was removed for prolonged periods. Users also reported instances of ‘shadowbanning’, a form of covert content moderation that involves making posts or accounts less discoverable by restricting their appearance in hashtag searches, recommendation algorithms or follower feeds, often without informing the user (Risius & Blasiak 2024). This particular phenomenon was also observed during the height of the 2020 Black Lives Matter protests, where multiple activists and users reported that hashtags such as #BlackLivesMatter, #BlackOutTuesday and #BLM were being hidden or demoted on Instagram and TikTok’s recommendation feeds (McCluskey 2020). While Meta claimed these were unintentional glitches or enforcement of content policies against misinformation, critics argued that the impact was a silencing of vital political expression and grassroots mobilisation efforts. This algorithmic opacity creates an uneven digital playing field where minority political speech is structurally disadvantaged.

As Roberts (2019:33–72) explains, content moderation is a fraught and often opaque process, shaped by a combination of automated systems and human decision-making. This results in inconsistent enforcement and unpredictable outcomes, frequently replicating offline inequalities within digital spaces. Research increasingly supports this, revealing systemic biases embedded within algorithmic moderation practices that can reproduce or even amplify social inequities (Noble 2018). Regardless of whether one accepts social media companies’ claims that such outcomes are the result of system errors rather than intentional interference, the effect remains the same: political discourse is being shaped and constrained. Artificial intelligence, in this context, operates as a gatekeeping mechanism, filtering speech and access to information in ways that function as de facto censorship. Given that social media platforms now serve as primary forums for public debate, their unilateral power to define the rules of speech, control information flows and set the terms under which discourse occurs represents a serious challenge to democratic principles. As Barroso and Van Brussel Barroso (2023:58) put it, ‘[t]he ability to participate and be heard in online public discourse is currently defined by the … algorithms of a couple major technology companies’.

A deeper structural challenge to democracy lies in the way social media platforms shape information flows, undermining not only the fairness of majority rule but also the capacity of marginalised groups to participate equally in democratic processes. These information flows increasingly resemble the opaque, centralised control more commonly associated with authoritarian regimes (Harari 2024:62). While it may seem self-evident that social media companies lack the formal authority of governments, the comparison is not without merit. Given the extent to which communication, including that of political parties, relies on these platforms, there is a strong case to be made that they exert a form of influence over political discourse that even many governments cannot match (Gillespie 2010). This influence is evident in the persistence of the marketplace of ideas in principle, but not in practice. Despite the appearance of openness, the promise of decentralised public discourse is often denied through algorithmic and corporate decisions that, whether intentionally or not, narrow the scope of visible, legitimate speech. In contrast to authoritarian regimes that overtly restrict fundamental rights such as speech, assembly and voting, social media platforms achieve similar outcomes more subtly, through the quiet, but powerful filtering of information.

The comparisons with authoritarian regimes may also be less subtle. Just as authoritarian governments often blame external forces to deflect accountability for their internal failings (Rosenfeld & Wallace 2024), social media companies frequently shift responsibility for the harmful consequences of their algorithms. And even if these platforms sincerely claim no intent to interfere in democratic processes, the outcomes remain deeply troubling. Harari (2024:90) highlights a striking example from 2016, when Facebook’s algorithm played a pivotal role in amplifying anti-Rohingya hatred, ultimately enabling ethnic violence in Myanmar. In response, Facebook issued a carefully worded statement acknowledging that it ‘wasn’t doing enough to help prevent [its] platform from being used to foment division and incite offline violence’. While this may appear as a concession, Harari argues that it functions more as a strategic deflection – shifting the blame onto users by characterising the platform’s failure as one of inaction rather than culpability. This framing obscures the active, algorithmic promotion of inflammatory content that contributed to the violence. In doing so, Facebook’s role was not merely passive negligence, but a form of structural complicity in the events that unfolded.

The reality of algorithmic content curation and its evident consequences for public discourse make the continued denial of responsibility by social media companies deeply concerning, particularly for marginalised groups. Their refusal to acknowledge the structural harms caused by profit-driven platform design reflects a troubling disregard for the democratic ideals they claim to support. Platform capitalism, by privileging engagement over equity, enacts a form of deliberate epistemic injustice against marginalised groups. It systematically excludes them from fair participation in democratic processes and distorts the public recognition of their experiences and claims. While traditional mass media did not guarantee democratic inclusion, it, at least, provided clearer institutional pathways for minorities to assert their voices and gain visibility. Indeed, the evolution of democratic legitimacy has depended on the inclusion of previously excluded groups, achieved through ‘votes, rights, and reasons’ for all. To suggest that social media algorithms now pose a threat to these foundational principles may seem hyperbolic at first glance, but the evidence increasingly supports this claim. The architecture of contemporary digital discourse does not simply fail to deliver on the promise of democracy for minorities – it actively undermines it.

In closing, it is important to emphasise that democracy is inherently complex – a reflection of the ambitious aims it strives to achieve. In a pluralistic society marked by moral diversity and competing interests, the prospect of achieving consensus on matters of law and policy through decentralised information flows is, in many ways, improbable. This complexity underscores the indispensable role of fundamental rights, which serve as safeguards for political legitimacy and public participation, and ensure that democratic legitimacy does not come at the expense of minority protection. Equally vital are the principles of public discourse and free speech, which function as mechanisms for empowerment, self-correction and potentially progress. While democracy may not be simple, its commitment to the inclusion and protection of marginalised groups must remain clear, uncompromising and grounded in the enduring values of equal participation and reasoned deliberation that define democratic life and distinguish it from other forms of governance.

This article has explored the tensions between the profit-driven nature of social media platforms and the democratic principles that underpin public discourse. In doing so, it has raised the point that public discourse has transformed online, to the extent that the marketplace of ideas might not exist anymore, at least as it was constituted in the theories that legitimise democratic mandates through public debate. This raises urgent questions about the future of speech, the participation of marginalised groups and the theoretical legitimacy of democratic mandates moving forward. As the Western world undergoes a period of intense political polarisation – with the resurgence of right-wing populism contrasting with vocal and mobilised liberal contingents advocating for the legal recognition of marginalised communities – these questions become increasingly relevant. This challenge moving forward lies in reconciling the democratic promise of open, inclusive discourse with the commercial logic of platform capitalism that so often undermines it.

A central problem is that the primary vehicle through which public discourse is now conducted is fundamentally misaligned with the normative ideals of deliberative democracy. While it is tempting to place blame squarely on social media companies for the erosion of public discourse, it is important to recognise that platforms like Facebook, which began in 2004 as little more than an improvement on email, did not initially emerge with the intention of shaping democratic engagement. In this context, the deeper failure arguably lies with governments and regulators who have been slow to adapt to the rapidly evolving influence of social media and have failed to implement effective safeguards. Ironically, though, their inaction has enabled an environment where digital communication remains largely unregulated – which is exactly the kind of public sphere that democratic societies demand.

In theory, algorithms hold the potential to enhance deliberative democracy, particularly in a way that supports the participation of marginalised groups. If designed with democratic values rather than profit maximisation in mind, they could be used to promote transparency, diversity and balance in public discourse. This might involve curating user feeds to expose individuals to a broader range of perspectives or integrating multiple information sources to produce more neutral, inclusive news coverage. In this way, algorithms could not only sustain a digital marketplace of ideas but also improve upon the often-unequal dynamics of the physical world. If algorithmic curation were genuinely neutral, it would no longer obstruct users’ efforts to engage in public debate. Instead, it could provide a space for constructive dialogue where the asymmetries of social media discourse make competing ideas visible to all users – including passive observers – thereby fostering a more balanced, real-time exchange on social issues in which marginalised voices are not just present, but meaningfully heard. Of course, not all structural barriers would be eliminated. For instance, access to the internet remains a fundamental obstacle – one that persists regardless of how inclusive or equitable online public discourse may become. It could nevertheless promote the normative ambition of public discourse in a way that social media algorithms do not currently.

More broadly, achieving a genuinely deliberative digital public sphere would require social media companies to prioritise epistemic and civic values over their core identity as profit-driven enterprises. This shift appears unlikely, as it would entail a fundamental restructuring of the business models that make these platforms viable. So long as user engagement remains the primary driver of revenue, platforms would need to be confident that open, transparent communication – some of which may interest users, and much of which may not – can still generate the levels of attention they depend on. An optimistic view suggests that users might ultimately appreciate exposure to unfamiliar content, discovering perspectives or topics beyond their usual preferences, just as the marketplace of ideas would hope. Yet this premise is questionable. Given the sheer volume of social media traffic, the notion that users would consistently engage with a transparent, less personalised and more predictable feed seems tenuous. Even if such a model attracted new users, whether they would remain engaged with a stream of disconnected, unfiltered content remains uncertain. Although the decentralisation of information channels is appealing to democracies because of its supposed transparency, it does not presume that users want to engage with all those channels. The nature of social media platforms is such that those channels would inevitably collide and overlap, either to the dismay of those seeking personalised experiences or the disappointment of those who are not. Therefore, the question of whether social media platforms can support deliberative democracy seems almost self-evident and a choice that lies with the companies themselves.

Perhaps the paradox at the heart of this challenge lies in the fact that, while social media platforms do not explicitly seek to foster democratic engagement – TikTok, for instance, prohibits political advertising despite its clear role in contemporary political discourse – they nonetheless champion the unregulated nature of their platforms as a defence of free speech. Klonick (2018) argues that platforms strategically present themselves as neutral intermediaries, invoking a ‘marketplace of ideas’ framework that assumes harmful speech will be met with counterspeech rather than actively removed. This stance enables platforms to distance themselves from direct accountability in content moderation decisions, allowing them to promote a generalised free speech ethos while sidestepping the political risks associated with removing specific content, including discriminatory or hateful speech (Gillespie 2018). Aswad (2018) highlights the ambiguities in Twitter’s hate speech policies, which lack sufficient specificity to effectively define the boundaries of harassment and incitement to violence. Despite policy updates following Elon Musk’s acquisition of Twitter, such critiques remain relevant – perhaps even more so – given the increased visibility and impact of algorithmically amplified hostile speech. Of course, these arguments depend on the acceptance that social media companies do not, in fact, use algorithms to interfere with speech on their platforms.

There is therefore a tension between what free speech means in the broadly democratic sense of public discourse and equal participation, and what free speech means in respect to the proclaimed lack of intervention on the part of social media companies. The first definition refers to the parity of processes that benefit marginalised groups because it fosters a deliberative and transparent environment where their voices and ideas can be heard. The second definition refers to a lack of regulation, at least in the overt sense that ignores the realities of content curation. However, it is this lack of regulation that compromises both definitions harmonising with one another. The future of public discourse is therefore uncertain. Because of the popularity of social media platforms, it would seem that what is happening right now with the transformation of political dialogue is likely to continue if social media companies remain financially constrained by their own profit-driven models of engagement maximisation. Social media companies certainly have the ability to foster decentralised and neutral forms of political dialogue, but the manifestation of such a promise lies squarely in their hands.

This article has argued that algorithmic content curation on social media platforms undermines the legitimacy of democratic discourse, particularly for marginalised speech that serves to challenge dominant norms. In practice, the act of curation constitutes a form of censorship or interference that centralises information flows and contradicts the principles of non-governmental interference in ordinary public discourse. In principle, the act of curation constitutes a denial of the fallibility of government and voting citizens – a principle which underpins the conceptual need for checks and balances, gives speech its democratic function and serves to distinguish democracy from authoritarianism. The purpose of this article is to demonstrate how the structure of public discourse is transforming in ways that contradict the practices and principles that help to distinguish the two methods of governance. This is problematic not just because they represent contradicting means of governance, but because echo chambers and algorithmic censorship deny the democratic mechanisms that give marginalised groups the spaces to be heard and the power to change the hearts and minds of fellow citizens.

As Harari (2024:68) writes, ‘[d]emocracies die not only when people are not free to talk but also when people are not willing or able to listen’. In today’s algorithmically structured public sphere, the ability to be heard – and to be listened to – is shaped by systems that are not transparent, accountable or designed with democratic principles in mind. If democracies are to remain legitimate in the digital age, they must take seriously not only the right to speak, but also the structures that govern information flow, deciding whose speech matters, who is heard and how disagreement and correction can take place. Protecting democratic communication in this context means ensuring that all voices – especially those at the margins – can meaningfully participate in shaping public understanding.