Who Controls the Medium of Democracy?

I. The People Have the Right to Be Wrong

Imagine that a country holds a genuinely free vote. The rules are public, citizens can speak, opponents are not repressed, and the count is honest. Yet the proposition before the electorate is almost absurd in its extremity: Should the country deliberately set itself on a path toward destruction? A majority votes yes.

The result would plainly be bad. It might be foolish, cruel, and irreversible. Nevertheless, if it were truly the free collective decision of the people, the government would have no right to annul it merely because officials believed it to be wrong. Before the vote, the government could argue, warn, publish evidence, and describe every foreseeable danger. After an authentic decision, however, refusing to obey on the ground that “the public is irrational,” “experts know better,” or “the long-term national interest overrides the majority” might preserve the state while ceasing to preserve democracy in its full sense.

Democracy is not a machine guaranteed to produce correct answers. Its legitimacy rests first not on the quality of the outcome, but on the authorship of the decision. A political community has the right to determine its own fate and to bear the consequences. The people have the right to choose wisely, but also to choose shortsightedly, mistakenly, even self-destructively. To say that “the people have the right to be wrong” ² is not to celebrate error. It is to reject the permanent substitution of a supposedly wiser, more competent, or more virtuous minority for the people as the final author of public decisions.

This does not make every majority decision morally right. A decision may be democratically authentic and substantively unjust at the same time. It may violate the rights of minorities and therefore remain subject to constitutional and moral opposition. This essay is not a complete theory of justice. It isolates a prior question: When a decision is called “the will of the people,” was it actually made by the people?

That distinction matters. The gravest danger to democracy is not necessarily that citizens make a bad choice. A greater danger may arise when a choice is substantially shaped, filtered, or foreclosed by another power, yet still acquires legitimacy under the name of popular will. A mistaken judgment that genuinely belongs to the public still preserves democratic authorship. Another actor’s judgment passed off as public will strips citizens of both agency and responsibility.

II. The Delegation We No Longer Notice

Modern democratic states rarely practice continuous direct democracy. Their scale is too large and public affairs too complex for every citizen to decide every day on budgets, foreign policy, courts, public health, energy, transport, and the countless details of administration. Representative government is therefore a mechanism of efficiency. Citizens entrust much of the daily work of politics to legislatures, executives, courts, and professional administrations, and then attempt to control those institutions through elections, oversight, law, and public criticism.

“Delegation” is more accurate here than permanent surrender. Sovereignty remains with the people in principle, even though most public power is exercised in daily life by representatives and institutions. The difficulty is that modern citizens are born into a system in which this delegation has already been institutionalized. We do not witness an original compact, nor do we usually examine each authority exercised in our name. Governments regulate conduct, collect taxes, police borders, allocate resources, and define legal obligations; most of the time, these activities appear as the natural background of political life. The boundary of delegation becomes visible only when a rule directly constrains one’s own freedom.

Most citizens therefore possess a real but indistinct understanding of popular sovereignty. They know that governmental authority should ultimately derive from the people, yet they may not recognize how many decisions the state makes on their behalf or how much responsibility follows from being the ultimate sovereign. Politics is often compressed into a vote every few years. Between elections, citizens return to private life.

Hayek’s warning is useful here: good people do not necessarily wish to participate in government. Many would rather entrust public decisions to someone they consider more competent. That preference is not inherently dishonorable or undemocratic. It simply reveals that citizens in a democracy do not maintain a constant and equal intensity of political engagement. Representative institutions accommodate that fact, but over time they can also deepen the habit behind it: we assign difficult matters to institutions, then gradually forget that institutional power is exercised in our name.

The internet alters precisely this relationship between delegation and withdrawal.

III. How the Internet Informally Simulates Direct Democracy

The internet has not abolished legislatures, nor has it distributed formal legislative authority to each citizen. It has, however, dispersed a range of political capacities that were once concentrated in political parties, established media, and formal organizations. Ordinary individuals can now speak publicly, organize, mobilize, monitor officials, set agendas, raise money, associate across distance, question public figures, and exert sustained pressure on candidates, corporations, and governments.

A video recorded at the scene of an event can force an official response within hours. An issue ignored by established institutions can enter national debate through widespread sharing. Individuals who have never met can form an action network without a traditional organization. Online petitions, crowdfunding, hashtag campaigns, and real-time public reactions can alter candidates’ positions, newsroom priorities, and administrative agendas. Citizens no longer need to wait for the next election before their judgments can produce political consequences.

This is not direct democracy in the legal sense. It is, however, a functional simulation of some of direct democracy’s effects. Representative institutions continue to operate at the center of the state, while a continuous, high-frequency, informal layer of digital politics grows around them. Every day, the public registers approval and opposition in ways that resemble voting without becoming votes. Attention becomes pressure; pressure enters formal institutions. Governments may not be legally bound by online opinion, but they find it increasingly difficult to behave as though that opinion does not exist.

In this sense, the internet has “returned” part of political power to citizens. The power in question is not merely the liberty to speak. It is the capacity to produce political consequences: to shape agendas, scrutinize government, coordinate action, transform electoral environments, and compel representatives to reconsider decisions.

That return is valuable in itself. The problem is not that ordinary people have regained too much political capacity, nor that online participation is inherently inferior to institutional politics. The problem is that these capacities are never exercised in a vacuum. To reach a public audience, they are mediated by user accounts, ranking systems, recommender systems, commercial rules, and privately controlled infrastructure. Power appears to have returned to the citizen, but on its way back it passes through a new class of intermediaries.

A question that once seemed less urgent therefore becomes unavoidable: Who controls the channels through which citizens regain political capacity?

IV. Authentic Will, Public Expression, and Final Choice

A commitment to popular sovereignty does not require a romantic view of the public.

Citizens may be poorly informed, emotionally driven, susceptible to group pressure, or unwilling to spend enough time on difficult questions. Political opinions are often unstable. The same person may change position because of new evidence, a change in material circumstances, the attitudes of friends, or a sudden event. These facts cannot justify taking final authority away from citizens, but they must form the starting point for any honest account of how democracy works.

I nevertheless believe that people possess something that can reasonably be called an authentic political will. They form genuine, if revisable, judgments about their interests, values, fears, and desired futures. Yet those inward judgments never enter public space in a complete and unmediated form. We need to distinguish at least three things: what a person inwardly believes, what that person is willing to say in front of others, and what that person ultimately chooses politically.

These three can diverge. Even in a society strongly committed to freedom of speech, people consider employment, family, friendships, reputation, and group belonging. They may conceal a belief for fear of losing a job, being ostracized, acquiring a permanent stigma, or undergoing a kind of social death. Costless candor is almost nonexistent. Law can prevent the state from imprisoning someone for an opinion, but it cannot remove every social consequence of expression.

The internet lowers the cost of speaking while also magnifying those consequences. A sentence can be detached from its original context, copied without limit, made searchable, and preserved indefinitely. The appearance of an overwhelming majority on a platform can then intensify self-censorship. Silence is not necessarily consent; loud expression is not necessarily the speaker’s most stable judgment.

A deeper transformation occurs before expression, at the stage when opinions are formed. Platforms are not neutral repositories of information. They determine the order in which information appears, the frequency with which it is repeated, the emotions with which it is associated, and the material that may never enter a particular user’s field of vision. Research has found that false news can travel farther and faster than truthful reporting, that language charged with moral emotion spreads especially readily, and that exposure to opposing views does not automatically reduce polarization and may, in some circumstances, intensify it.^3–5

At the same time, the evidence does not support a crude theory of algorithmic remote control. An experiment involving consenting Facebook and Instagram users during the 2020 U.S. election found that, over roughly three months, replacing default ranking with reverse-chronological feeds substantially changed time spent on the platforms and patterns of information exposure, but produced no statistically significant change in several political attitudes measured by the study. The article was corrected in 2024 and again in 2026 for issues involving deleted Instagram accounts and the treatment of carousel posts. It should therefore be read as a limited example rather than a universal claim: changing an information environment does not necessarily produce a detectable short-term change in political attitudes.⁶

The most dangerous condition is therefore not complete ignorance. It is the possession of a large volume of selected information accompanied by the belief that one has seen the whole. People may consume endless political content while remaining unaware of what was never presented. They may feel that they are reasoning independently while being unable to observe the mechanisms that define the boundaries of their attention. Information abundance does not necessarily create epistemic autonomy. It can also create an illusion of completeness.

V. From Influencing Democracy to Controlling It

To describe the internet as merely “influencing” democracy can be too mild. Advertisements influence consumers, speeches influence audiences, and friends influence one another. If platforms simply placed additional opinions before citizens, their political role would differ only in scale from older forms of communication.

Digital platforms do more than persuade. They can control the gateways to public attention, rank visibility, decide what can spread, what is demoted, what receives a warning label, and what disappears from search or recommendation. They define the conditions under which accounts continue to exist and can, at decisive moments, remove a participant from a principal arena of public discussion.

I therefore retain the stronger expression “control democracy,” but it must be defined carefully. Control does not mean determining every ballot. It does not turn citizens into machines without will, and it does not imply that an algorithm can manufacture any result at command. It means structural control over the conditions under which democratic judgment is formed:

• control over what the public notices first;
• control over which voices can reach mass audiences;
• control over which emotions are repeatedly intensified;
• control over what appears to be the majority;
• control over which alternatives seem imaginable, acceptable, or unspeakable;
• control over who is permitted to remain in the public arena.

One need not control every drop of water in a river. The durable power to reshape the riverbed can still alter the direction of the flow. Likewise, a platform need not determine each citizen’s final choice. If it can continuously structure attention, agenda, visibility, and the boundaries of participation, it possesses the capacity to control the operating conditions of democracy.

I therefore continue to defend a claim from the original version of this essay: If democracy must operate through a medium, whoever controls that medium acquires the possibility of controlling democracy. The internet has become one of the principal media through which modern democracy operates.

The “who,” however, is not a single actor. States influence platforms through law, enforcement, and sometimes informal pressure. Platforms govern users through rules and algorithms. Politicians, advertisers, influencers, and organized movements attempt to exploit those systems. Users, in turn, constrain states and platforms through exit, protest, and public criticism. Power circulates among these actors and becomes difficult to locate. Jack Balkin ⁷ describes digital speech as a triangle involving states, private infrastructure, and speakers; Kate Klonick ⁸ calls large platforms “new governors” of online expression.

The danger of electronic democracy is therefore not reducible to a false post or a disputed suspension. It lies in an entire structure of power that already produces political consequences while lacking a settled name, procedures, and forms of accountability commensurate with its role.

VI. Brexit: Democratic Choice and Belated Understanding

The 2016 referendum on the United Kingdom’s membership in the European Union is an instructive case, but it must be used carefully.

The official result recorded 17,410,742 Leave votes (51.9 percent) and 16,141,241 Remain votes (48.1 percent). The difference was 1,269,501 votes, approximately 3.8 percentage points, on a turnout of 72.2 percent. ⁹ This was not a negligible result. Leave won a majority of the votes, and the outcome had to be treated as an authentic democratic decision.

A claim frequently attached to the referendum—that six million people demanded another vote immediately after learning the result—is inaccurate. The parliamentary petition that eventually received 6,103,056 signatures called for the revocation of Article 50 and continued EU membership. It took place in 2019, almost three years after the 2016 referendum; it was not an immediate demand to rerun the vote. ¹⁰ Petition signatures are also not equivalent to a regulated, secret national ballot. The petition’s signatories were not necessarily the same people who had voted in the referendum, and a society confronting the consequences of a decision three years later may possess different information and interests.

Correcting the chronology does not make the case irrelevant. It clarifies a temporal difference between electronic and formal democracy. A referendum asks citizens to make a binary choice on a specified date. Digital politics does not end when the ballot boxes close. Negotiation details, economic expectations, identity conflicts, and practical consequences continue to emerge; public understanding and opinion continue to change. The formal decision is episodic. Digital public deliberation is continuous.

Brexit also demonstrates the compressive force of a binary referendum. “Leave” and “Remain” folded questions of trade, borders, sovereignty, immigration, regulation, Northern Ireland, party competition, and national identity into a single binary choice. No voter could possess complete knowledge of every consequence. The online information environment then repackaged that complexity as slogans, images, emotional appeals, and highly targeted messages. A later parliamentary inquiry examined data analytics, political microtargeting, and disinformation, and raised serious concerns about the lack of transparency and accountability in platform-mediated political communication.¹¹

None of this proves that “the internet caused Brexit.” Nor does a close result by itself establish that any particular stream of information changed the outcome. Such a causal claim exceeds the evidence. A more defensible conclusion is that when an electorate is divided nearly in half, even mechanisms operating at the margins of attention and information distribution may become consequential; when those mechanisms are invisible and unauditable, society has difficulty judging how far the result reflects autonomous public choice.

Even then, voters remain responsible for their ballots. Incomplete information is a permanent condition of politics. Being persuaded, inflamed, or later regretful does not automatically transform one’s vote into someone else’s act. Unless unlawful interference sufficient to invalidate the result can be established through legitimate procedures, a democratic result should not be set aside retrospectively merely because it has become unpopular. The proper response is not to remove the public from future decisions. It is to build a more transparent, pluralistic, and auditable information environment before the next consequential choice.

The enduring question raised by Brexit is therefore not simply whether the people chose wrongly. It is this: How can we know that the mistake—if it was a mistake—truly belonged to the people?

VII. A Privately Owned Public Square

Large social platforms usually exist as private companies. Their servers, software, brands, and business models are privately owned, and users enter after accepting contractual terms. Under traditional private-law logic, a company must have substantial authority to manage its own product.

Yet “product” becomes an inadequate description when political leaders address a nation there, news organizations distribute reporting there, social movements organize there, and citizens use the space to determine what matters. Global platforms remain private in ownership while performing some functions of a public square, communications infrastructure, editorial institution, and political mobilization system. They are not states, but their decisions have public consequences.

The U.S. Supreme Court has described social media as among the most important places for the exchange of views in the modern era and invoked the image of the “modern public square.”¹² That language does not mean that private platforms automatically become government bodies. Under the American state-action doctrine, the First Amendment primarily constrains the state; providing a forum for speech does not by itself transform a private entity into a state actor.¹³ At the same time, Moody v. NetChoice emphasized that, depending on the platform function at issue, selecting, ranking, and organizing third-party content may constitute the platform’s own constitutionally protected editorial judgment.¹⁴

Taken together, these propositions reveal the difficulty. Social platforms can resemble public squares in social function, remain private actors in constitutional classification, and possess their own expressive and editorial rights. We cannot simply say, “The company owns the servers, so it may decide everything without restraint.” Nor can we mechanically apply every constitutional rule governing the state to every platform decision.

The result is not a complete legal vacuum but a constitutional and institutional gap: platform power has acquired public consequences without yet being matched by equally robust forms of public accountability.

Private ownership should not erase obligations created by public function. The harder a platform is to replace, the more power it has to allocate public visibility, and the more readily its decisions produce political consequences across borders, the greater its need for political awareness, institutional responsibility, the capacity to act, and restraint. It must be able to respond to violence, illegality, and coordinated abuse. It must also refrain from quietly transforming commercial rules into unreviewable political judgments.

VIII. The Suspension of Donald Trump: A Question of Power and Procedure

On 8 January 2021, twelve days before the end of Donald Trump’s presidential term, Twitter (now X) permanently suspended his account, citing “the risk of further incitement of violence.” Facebook and Instagram also restricted his ability to post. The Facebook Oversight Board later upheld the immediate suspension but found that an indefinite penalty, lacking both a defined endpoint and a place within Facebook’s existing penalty system, was inappropriate. It required Facebook to issue a new decision grounded in published rules, necessity, and proportionality. Twitter restored the account in 2022, and Meta ended its suspension in 2023; the issue examined here is the power and procedure exposed by the decisions made in 2021.^15–16

Public debate over the case broadly divided into two positions. One side treated the suspension as an expansion of private censorship; the other emphasized the platform’s responsibility to address an immediate risk of real-world violence. This essay takes no position on Trump’s politics, nor does it assume that the speech at issue was harmless. Public office does not place a user above published rules; private ownership, however, does not automatically confer an unlimited power of political exclusion.

The question is who may impose a decision of such political consequence, and by what procedure.

The suspension did not literally render a president silent. Trump retained the White House press room, traditional media, and other means of communication. Yet the decision excluded a sitting U.S. president from one of the most important digital public arenas of the time and severed a direct channel connecting him to tens of millions of followers. This was not an ordinary account-service dispute. It was a decision with major political consequences.

An imminent danger may require immediate platform action. If a credible risk of real-world violence exists, waiting for lengthy litigation before intervening can make governance ineffective. But the power to act temporarily in an emergency is not the same as the authority to impose a permanent and unreviewable political exclusion. The more urgent the initial action, the more clearly its duration should be limited. The more consequential the sanction, the greater the need for published reasons, an evidentiary record, an avenue of appeal, and external review.

The objection, then, is not that a platform may never restrict Trump’s account. It is that a private company, acting through internal procedures alone, can decide for an extended period whether a person holding exceptional public status may continue to participate in a de facto public square. When that power lacks clear boundaries, stable standards, and independent review, it begins to resemble a political power of exclusion generated outside public law.

In one case, such power may be used against a nationally prominent political figure at the center of intense controversy. In another, it may be used against a minority candidate, an anti-war organizer, a whistleblower, or a group that neither the state nor the platform wants to hear. The legitimacy of an institution cannot depend on whether the immediate target is popular or unpopular. It must depend on whether the power itself is bounded by clear rules, fair process, and independent review.

IX. The Manipulated Citizen Remains a Political Agent

To recognize that platforms can control the conditions under which democratic opinion forms is not to reduce voters to innocent victims of algorithms.

“Manipulation” is a matter of degree. Contemporary information environments are unequal, but they are rarely absolutely closed. A person may encounter the same narrative repeatedly, be moved by fear or anger, and be constrained by group relationships. Yet that person may still search for contrary evidence, suspend judgment, consult others, or admit ignorance. So long as that possibility has not been eliminated, political agency remains.

This demand on citizens may be severe, but it is inseparable from democracy. A citizenry that claims final authority cannot, whenever a decision produces harm, simultaneously describe itself as wholly incapable of thought. A ballot has authority precisely because it is treated as the act of someone capable of judgment. If every misled choice ceases to belong to the voter, nearly every real democratic decision can be reclassified as the product of some influence operation or communications system. Popular sovereignty would then be hollowed out in the name of protecting the people.

Three forms of responsibility must therefore coexist. Manipulators bear responsibility for intentional deception and incitement. Platforms bear structural responsibility for the design, ranking, and enforcement systems through which political communication operates. Citizens bear responsibility for what they ultimately accept, believe, and choose.

Structural control and individual responsibility are not contradictory. Platforms alter the environment of choice; citizens complete the choice. We cannot let an algorithm sign the ballot in place of a voter. Nor can we use the voter’s signature to absolve the algorithm and the platform.

X. What Constraints Does a Digital Public Space Require?

If major platforms are neither ordinary private businesses nor state agencies, governance should not allow any one actor to acquire absolute control. Platforms cannot abandon content moderation. Governments must not construct a ministry of truth in the name of saving democracy. Courts cannot preapprove the vast volume of material posted each day. The practical task is to subject platform powers with significant public consequences to clear procedures, graduated sanctions, independent review, and enforceable accountability.

At least ten principles follow.

1. Rules should be published in advance, not invented for a particular target

Platforms should explain in clear language what conduct is prohibited, how sanctions escalate, and what conditions trigger emergency action. Comparable conduct should receive comparable treatment. An undefined appeal to corporate “values” cannot by itself justify exclusion from public participation.

2. Major decisions require reasons and evidence

The deletion of spam and the permanent suspension of a major political figure should not generate the same minimal notice. The more severe the sanction and the greater its public effect, the more complete the explanation should be. Evidence sufficient for later review should be preserved.

3. Appeals must be meaningful

Internal review cannot merely ask the same automated system to confirm itself. Users must be able to provide context and contrary evidence. Major cases should also be eligible for review by a body institutionally independent of the platform’s commercial leadership. The European Union’s Digital Services Act requires reasons for specified content removals and account restrictions, together with internal complaint procedures and out-of-court dispute settlement. It offers at least one example of how private governance can be made more procedural.¹⁷

4. Emergency measures must actually be temporary

When content may immediately produce violence or grave harm, platforms should be able to restrict distribution quickly. But emergency authority must carry an automatic expiration and enter a more demanding review within a short period. Exceptional circumstances cannot become a formula for indefinite punishment.

5. Accounts central to democratic life require heightened procedural safeguards

This is not a speech privilege for politicians. It reflects the fact that excluding them affects the capacity of large numbers of citizens to receive information, monitor authority, and participate in elections. Long-term or permanent exclusion of incumbent national leaders, major candidates, public institutions, and other accounts of exceptional public interest should automatically trigger independent review.

6. Courts should be able to review major long-term political exclusions rapidly

Courts need not approve every deletion and should not become the daily editor of a platform. But where a private decision effectively excludes a sitting leader, major candidate, or another participant whose exclusion would have systemic democratic consequences for an extended period, legislation should provide an expedited avenue of judicial review. The court’s role would not be to decide the truth of every disputed sentence. It would examine whether the rules were public, the evidence adequate, the measure necessary and proportionate, and the procedure fair.

7. Large platforms should bear a limited duty of public access

Such a duty would not require platforms to host threats, harassment, illegal material, or incitement to violence. Nor would it require algorithms to promote every viewpoint equally. It would mean that a platform controlling a crucial public space could not arbitrarily deny basic access merely because a political viewpoint was unpopular or controversial. Political viewpoints must be distinguished from illegal or dangerous conduct.

8. Users should be able to see and choose their information environment

Recommender systems should not be treated as a natural and unchallengeable order. Platforms should disclose the main principles affecting ranking, provide a meaningful option not based on personal profiling, and help users understand why particular political material reaches them. Transparency cannot eliminate manipulation, but it can reduce the power of what remains unseen.

9. Political advertising, microtargeting, and public-interest research must be auditable

Citizens should know who paid for political communication, which groups were targeted, and what different versions were shown. Qualified researchers and regulators need privacy-preserving access to sufficient data to determine whether platforms systematically amplify particular risks. Power that cannot be observed can scarcely be governed democratically.

10. State pressure on platforms must also be transparent

Platforms are not the only institutions capable of manipulating public space. Governments may use enforcement threats, private communications, or regulatory bargaining to induce platforms to suppress speech that the state would prefer not to prohibit openly. Government requests affecting political expression should leave an auditable record and be subject to legal and public scrutiny. Otherwise, apparent platform autonomy may become outsourced state censorship.

These principles cannot eliminate conflict. Platforms possess expressive and editorial freedoms of their own, and government regulation can itself become abusive. Serious institutional design does not pretend that risk can be reduced to zero. It prevents any single actor from monopolizing rulemaking, fact-finding, punishment, and final appeal at once.

XI. We Are Already Inside It

Electronic democracy is not a system that future society can simply choose to adopt or reject. We cannot withdraw from it by decree. Closing digital political participation would be both unrealistic and destructive of the capacities for expression, organization, and public scrutiny that the internet has already returned to ordinary citizens.

Nor has electronic democracy replaced representative government. Legislatures, executives, courts, parties, and elections remain the core of democratic politics. Digital technology is creating a new political layer around those formal institutions. It has no unified constitutional framework and no settled boundary, yet it can structure agendas, amplify collective emotion, allocate visibility, coordinate action, and compel formal authority to respond.

The power that most urgently requires restraint is therefore not the political capacity citizens have regained, but the intermediary power that decides how that capacity is seen, connected, and amplified. The state must not reclaim final authority from the people on the pretext of saving them from error. Platforms must not govern public life invisibly on the pretext of private ownership. Citizens, for their part, cannot transfer all responsibility to the information environment when their choices produce consequences.

Democracy can fail. Citizens can regret their decisions. A political community may even choose to harm itself. We may criticize that choice, attempt to persuade the public to change it, and build better education, media, and procedures so that more facts are visible before the decision is made. But if we continue to believe in democracy, we cannot allow an intermediary lacking democratic authorization to take control of the process by which the people decide.

The task of electronic democracy is not to protect the people from making mistakes. It is to make sure that the mistake truly belongs to the people.