On 8 August, the world’s wealthiest man wrote on his own social network that Hitler was a left-wing socialist. It was neither a slip nor a first occurrence: in January of last year, he had already endorsed that thesis while conversing with the leader of Alternative for Germany, who maintained in his presence that Hitler was a communist.

The assertion is false, and not in the manner of opinions we do not share. The United States Holocaust Memorial Museum encyclopedia defines the Nazi party as a far-right, racist, and antisemitic party, and explains that it sought to attract German workers by distancing them from socialism and communism. Britannica answers the question of whether the Nazis were socialists with a resounding no: not in any meaningful sense, and certainly not after 1934. The regime persecuted communists and social democrats, and dissolved free trade unions. There is no open historiographical debate. There is an established fact that is being denied, and denied repeatedly after having been corrected.

That repetition matters, because it distinguishes error from operation. An analysis by the Center for Countering Digital Hate, a monitoring organisation with a stated methodology, counted eighty-seven posts by Musk regarding the 2024 US elections that fact-checkers classified as false or misleading, with two billion cumulative views. None carried a visible community note. A debunked piece of data that is republished ceases to be a mistake.

However, if the objective were to persuade, one would have to say that the method functions poorly. And here I wish to present the evidence that runs counter to what many of us take for granted, because to conceal it would be to indulge in the very vice this text denounces.

In 2023, two experiments were published in Science and Nature that intervened in the feeds of tens of thousands of users over several months. One replaced the algorithm with a chronological order; the other reduced exposure to like-minded sources. Neither found significant effects on ideological or affective polarisation, nor on eight pre-registered attitudinal indicators. Previous research had found that unreliable websites accounted for a small portion of the average information diet, concentrated among users who were already seeking such material. Exposure changes; attitudes, in those measurements, do not.

Thus, the algorithm does not appear to be changing what people think. And yet something has shifted, and we all perceive it. The best-documented explanation does not lie in persuasion.

Vicente Valentim argues, in a book published by Oxford, that the growth of radical behaviour does not stem from individuals who have changed their minds, but from individuals who already held those positions and did not act upon them because they believed them to be socially unacceptable. What is modified is not the belief: it is the cost of expressing it.

Read this from the perspective of someone who wishes to advance a position that society rejects. If they do not need to convince anyone, their objective changes entirely. They do not have to win the debate. It suffices for them to be admitted into it as one party among others. The moment their position appears treated as a legitimate side against another, the social cost of maintaining it collapses, and those who already shared it cease to remain silent. Legitimation is not a byproduct of the strategy. It is the strategy.

And there exists a procedure that produces exactly that, the effect of which is measured under experimental conditions.

In 2016, Derek Koehler presented a group of participants with questions of varying degrees of expert consensus: some were given information balanced between two positions, while others were given the full account of opinions. Those in the first group ceased to distinguish clearly between issues where there was consensus and those where there was not. His conclusion is that false balance distorts the perception of expert judgement even when the reader has all the information necessary to correct it.

Three years earlier, Dixon and Clarke had measured it where the consequences are not abstract. They provided an article to read that presented the supposed controversy between vaccines and autism in a balanced manner. Those who read it were less certain that vaccines were safe, believed that experts were less certain, and declared a lower willingness to vaccinate their future children. The article did not contain a single lie. It presented two sides of something that has none.

This is where I wish to pause, because the name we give to this is performing work that does not belong to it.

We call this phenomenon equidistance. But equidistance describes a position: being at an equal distance from two extremes. And that presupposes that the two extremes exist and are where they are said to be. When someone presents as equivalent two positions that are not, they are not situating themselves in the middle of anything real. They are manufacturing the middle. They decide who sits at the table, with how much time, under what framework, and with what contradiction. The operation does not consist of positioning oneself: it consists of positioning others.

Calling it equidistance grants the person executing it the description that suits them best, because it sounds like an excess of caution, like a poorly calibrated professional scruple, like a character flaw. It sounds like something that happens to someone, not something that someone does.

With that distinction in mind, one sees two different things that produce the same effect. One is the inertia of the profession: the journalistic norm of presenting both sides, applied mechanically. The classic study by Boykoff and Boykoff demonstrated in 2004 that this norm, in the coverage of climate change, produced a significant divergence between public and scientific discourse. There is no bad faith there; there is a procedure that fails.

The other is the deliberate exploitation of that very norm. And here we are not dealing with a hypothesis regarding hidden intentions: academic literature has spent years documenting communication strategies of the far right on platforms, with systematic reviews of the field, and studies that have identified the use of new accounts and coordinated bots to disseminate identifiable discourses. Coordination is not inferred: it is measured. And those who coordinate know what they are doing.

The decisive factor is that the latter uses the former. The professional norm is there, it is public, it is predictable, and it produces a known effect: whoever demands to be heard as an equal does not appeal to anyone's generosity, they appeal to a procedure they know will work. That is why the framework of the debate has become filled with invitations to pluralism and complaints about censorship. These are not arguments about the substance. They are keys to open a door whose mechanism is well-studied.

The fact that a door is opened by someone acting in good faith does not alter the outcome. It does, however, shift where the responsibility lies, and both aspects must be addressed: there are those who manipulate and those who allow themselves to be used, and the latter possess the means to discern the situation. The professional duty is clearly defined. The BBC’s editorial guidelines establish that minority positions or those less supported by evidence should not necessarily be granted prominence or weight comparable to those that are better documented or enjoy greater support. The British regulator further clarifies that due impartiality does not equate to an equal division of time. This has been documented and accessible for years.

There is a domain where this ceases to be a discussion regarding journalistic procedures. The obligations derived from international human rights treaties are not political opinions: they are legal obligations that State parties are bound to respect, as formulated by the Office of the High Commissioner. Presenting compliance with these as merely one stance among others is not pluralism; it is a category error, and when done knowingly, it is something more grave. Accuracy, conversely, has no relation to this: an arrest warrant from the International Criminal Court is issued when there are reasonable grounds to believe a crime has been committed and does not equate to a conviction; stating this with precision is rigour, not counterbalancing. Precision and equidistance are opposites: one refines, the other equates that which is not equal.

The role of platforms remains to be addressed, and not to absolve them. It is not proven that they persuade. What is proven is worse, as it does not rely on any hypothesis regarding anyone’s state of mind.

Last December, the European Commission fined X one hundred and twenty million euros for failing to meet its transparency obligations: misleading verification marks, an insufficient advertising repository, and a lack of access to data for researchers. The sanction is not for disinformation; it is for preventing external verification of what occurs internally, while the investigation into systemic risks continues separately. With TikTok, the Commission opened a procedure in December 2024 regarding electoral integrity following the Romanian presidential elections, the first round of which the country's Constitutional Court had just annulled. The procedure remains open and it cannot be claimed that the platform caused the annulment, but the fact that a European election is annulled and the regulator investigates the role of a social network is unprecedented.

And then there is what the companies knew. Among the internal Meta documents published in 2021 is the finding by their own data scientists in 2019: posts that elicited angry reactions disproportionately contained disinformation, toxicity, and low-quality information. Frances Haugen summarised it under oath before the Senate: the company repeatedly found conflicts between its profit and safety, and systematically resolved them in favour of profit. And in the most serious case, Amnesty International concluded in 2022 that Meta's algorithms and its reckless pursuit of profit contributed substantially to the atrocities committed against the Rohingya population, and that the company knew or should have known that its systems were supercharging the spread of harmful content.

That is not algorithmic persuasion. It is distribution, awareness of harm, and omission. And it is substantiated.

I shall add a clarification: Cambridge Analytica did not win Brexit. What the British Electoral Commission sanctioned was electoral expenditure: an associated campaign enabled Vote Leave to exceed its legal limit by nearly half a million pounds. Furthermore, the data protection authority concluded there was no evidence to modify its assessment that said consultancy was not involved in the referendum campaign. Repeating the conventional narrative does not strengthen the denunciation; it weakens it, as the first informed reader dismantles it and brings everything else down with it. When manipulation is denounced, accuracy is not an embellishment: it is the sole defence.

I shall summarise what I perceive to be occurring. There are those who disseminate historical falsehoods from positions of maximum visibility and fail to rectify when corrected. There are organised structures that amplify these messages in a coordinated and documented manner. And there is a procedure, known and studied, which converts an inadmissible position into one side of a debate without the need to convince anyone: it suffices to seat it opposite another and allow the viewer to infer that the truth must lie somewhere in between. That midpoint did not exist before someone constructed it.

And when what is normalised in this manner is hatred, calling it equidistance ceases to be a vocabulary error. It is participation.

On open conversation

This text belongs to the series on infrastructure that I am developing in this notebook, with a conclusion that partially corrects what I myself anticipated: the evidence that platforms distribute unequally, were aware of harms, and failed to mitigate them is robust; the evidence that they independently alter what people think, is not. The mechanism with an experimentally measured effect is older and less spectacular, and consists of presenting as two sides that which has none. I maintain that calling this operation equidistance absolves the one executing it, because it describes as a position what is, in fact, an act. Should anyone wish to intervene from the perspective of journalistic deontology, research into disinformation, political communication, or international human rights law, this notebook remains open, and I am particularly interested in being pointed towards evidence I may not have considered.

Sources

Center for Countering Digital Hate. Analysis of Elon Musk's electoral posts on X, 8 August and 4 November 2024.

Barrie, Christopher. “Did the Musk takeover boost contentious actors on Twitter?”. Harvard Kennedy School Misinformation Review, 29 August 2023.

European Commission. Procedure against X under the Digital Services Act (18 December 2023; preliminary findings, 12 July 2024; 120 million euro sanction, 5 December 2025). Procedures against TikTok (19 February 2024; commitments regarding TikTok Lite, 5 August 2024; electoral integrity, 17 December 2024).

Constitutional Court of Romania. Annulment of the presidential electoral process, 6 December 2024.

United States Holocaust Memorial Museum, Holocaust Encyclopedia, entry on the Nazi Party. Encyclopaedia Britannica, “Were the Nazis socialists?”.

FBarchive (Harvard Kennedy School). Internal Meta documents published in 2021. Testimony of Frances Haugen before the United States Senate, 4-5 October 2021.

Amnesty International. The Social Atrocity: Meta and the right to remedy for the Rohingya, 2022.

Electoral Commission (United Kingdom). Investigation into Vote Leave, BeLeave and AggregateIQ, 2018. Information Commissioner’s Office, investigation into the use of data analytics in political campaigns, 2018.

Koehler, Derek J. “Can journalistic ‘false balance’ distort public perception of consensus in expert opinion?”. Journal of Experimental Psychology: Applied 22(1), 2016. DOI: 10.1037/xap0000073.

Dixon, Graham and Christopher Clarke. Health Education Research 28(2), 2013. DOI: 10.1093/her/cys110.

Boykoff, Maxwell T. and Jules M. Boykoff. “Balance as bias”. Global Environmental Change 14(2), 2004. DOI: 10.1016/j.gloenvcha.2003.10.001.

BBC Editorial Guidelines, section on due impartiality. Ofcom, Broadcasting Code.

Office of the United Nations High Commissioner for Human Rights, International Human Rights Law. Rome Statute of the International Criminal Court, Article 58.

Valentim, Vicente. The Normalization of the Radical Right. Oxford University Press, 2024.

Guess, Andrew et al. Science 381 (2023). Nyhan, Brendan et al. Nature 620 (2023). Guess, Nyhan and Reifler. Nature Human Behaviour (2020).