Professor Jason Arday and the anatomy of a digital pile-on

There is an old saying, often attributed to Mark Twain, that history does not repeat itself, but it rhymes. When it comes to the representation of Black people in the media, the technology may have changed dramatically, but there is something uncomfortably familiar about the tune.

For decades, researchers and campaigners have questioned the disproportionate representation of Black people in parts of the traditional media, particularly when stories concern crime, misconduct, failure or social disorder. The argument was never that wrongdoing involving Black people should not be reported. It was about proportionality, which stories were selected, how prominently they were presented, how frequently particular images and narratives were repeated, and whether the misdemeanour of an individual was allowed to become representative of something much larger.

Research commissioned by the Sir Lenny Henry Centre for Media Diversity illustrates the problem. Its examination of 275 documentaries found that race and racism were the leading subjects when Black people featured, with crime the second most common theme. Researchers warned that repeatedly framing Black lives through racism, criminality and trauma risked presenting a narrow picture of Black experience and reinforcing troubling associations between Black communities, fear and crime.

This matters because representation is not simply about whether something is true. It is also about frequency, prominence and proportion. A particular story may be factually justified, yet the cumulative effect of repeatedly selecting the same kinds of stories about the same groups can produce a distorted picture of reality.

Traditional media exercised this power primarily through editorial selection. Editors determined what made the front page, which photograph accompanied a story, how long a controversy remained in the news cycle and when the public should move on.

The digital age has not necessarily eliminated that problem. It may have given it a new form, a new shape and, crucially, a new distributor.

Over the last several years, the debate over social media took another turn. Technology companies faced criticism from political figures and free-speech advocates who argued that centralised fact-checking and content moderation had become politically biased, overly interventionist and too willing to determine what people should be allowed to see. There was, and remains, a legitimate debate to be had about freedom of expression, political neutrality and the extraordinary power private technology companies exercise over public discourse.

But there was another question we perhaps did not ask loudly enough:
What happens when some of the guardrails come down, while the engagement machine remains?

Removing or weakening moderation does not create a neutral public square. Beneath every social media feed sits a collection of ranking and recommendation systems making millions of decisions about what receives attention, what disappears and what people encounter next. These systems are not principally designed to establish truth. They learn from behavioural signals: what we watch, click, share, reply to, search for and spend time looking at.

In September 2024, the Guardian documented an increasingly toxic economy developing on X in which controversial and extreme content could translate engagement into visibility and, for some users, revenue. Researchers and anti-extremism experts warned that previously deplatformed figures had returned while enforcement weakened, creating an environment in which deliberately provocative content could become commercially rewarding.

The consequences were not confined to the screen. In the aftermath of the Southport murders, false claims about the attacker’s identity, ethnicity and religion spread rapidly online before racist disorder erupted across Britain. Hope Not Hate described X as a central hub in the creation and distribution of content surrounding the unrest. One man later convicted of inciting racial hatred had reportedly been earning about £1,400 a month from his activity on the platform.

This matters because it exposes the weakness in describing the problem simply as “free speech”. When provocative speech is algorithmically amplified, financially rewarded and distributed at enormous scale, the platform is no longer merely providing somewhere for people to speak. It has created an economy around which kinds of speech attract attention.

This is where the economics of outrage becomes important. A social media algorithm does not need to believe a racist talking point. It does not necessarily need to establish whether it is true. It needs to recognise that the talking point is generating a reaction. Those agreeing with it may enthusiastically share it. Those appalled by it may quote it to condemn it. Journalists may report on it, campaigners may challenge it and thousands of people may enter the comments to argue.

Their motivations could not be more different. Computationally, however, they may produce many of the same signals. Agreement generates engagement, but so does disagreement. Outrage generates engagement, but so does counter-outrage.

When attention becomes a proxy for relevance, controversy can acquire the appearance of importance, repetition can acquire the appearance of consensus and visibility can acquire the appearance of truth.

The attention economy discovered outrage

This distinction is particularly important when considering racism and the far right. It would be too simplistic to claim that social media algorithms are simply designed to promote far-right ideas. The more plausible problem is an interaction between two forces, an engagement system that rewards material capable of producing intense human reaction, and political actors, influencers, anonymous accounts and opportunists who have become increasingly sophisticated at producing precisely that reaction. The more troubling possibility is that engagement-driven systems can disproportionately reward some of the characteristics on which extremist and racialised narratives thrive on issues and codes like immigration, invasion, grievance, fear, identity, humiliation, resentment, perceived threat and moral outrage.

The algorithm does not have to share the ideology for the ideology to benefit from its architecture.

This has helped create an economy of attention in which bots, anonymous accounts, influencers, commentators and professional clout-chasers compete for the same scarce resource, our eyeballs! The incentives are obvious. Nuance rarely travels as quickly as accusation. Restraint struggles against outrage. Being first can become more valuable than being right. A measured explanation may attract hundreds of interactions while an inflammatory accusation attracts tens of thousands.

Provocation is no longer merely speech. It can be a growth strategy.

We have already seen evidence pointing towards this problem in Britain. Research examining activity on X surrounding the 2024 disorder that followed the Southport murders found a strong relationship between engagement and visibility, while visual posts promoting racist conspiracy theories received substantially greater amplification than other material examined by the researchers.

Amnesty International subsequently argued that X’s design and policy choices had created fertile conditions for inflammatory anti-Muslim and anti-migrant narratives to spread.

Southport therefore matters because it takes the argument beyond the hypothetical. False and racialised narratives can enter an engagement ecosystem, be repeated by networks of users, attract enormous reaction and acquire visibility wildly disproportionate to their factual foundations.

The bad actor does not necessarily need to hack the algorithm. They need to understand its incentives. They need to know that anger travels, accusation invites response and racial grievance generates both supporters and furious opponents. They need to understand that a provocative question can attract more attention than a measured answer, and that when enough accounts discuss the same subject at sufficient velocity, journalists, commentators and ordinary users may begin responding not only to the original claim but to the apparent scale of the conversation itself.

The algorithm then encounters what looks like highly successful content. This creates a disturbing relationship. The provocateur supplies the outrage, opponents supply counter-outrage, influencers supply commentary, journalists supply further attention and ordinary users supply clicks and replies. The ranking system supplies distribution.

It also complicates the argument about platform guardrails. When moderation policies are relaxed, fact-checking mechanisms weakened or controversial accounts restored, the significance is not merely that additional speech becomes permissible. Such changes can alter the environment in which the engagement engine operates. Actors previously constrained by platform rules may find greater freedom to experiment with the kinds of material that generate attention.

The question is therefore not simply whether removing guardrails increases racism. It is whether a more permissive environment, combined with an engagement-ranking system, creates an exploitable opportunity for those who already understand how outrage travels.

Whoever controls the algorithms and platforms consequently exercises extraordinary influence over our online reality, not necessarily by determining what is true, but by determining which version of reality reaches the most eyeballs.

Algorithmic overrepresentation

This is where an older problem of media representation begins to rhyme with a new one.

Where traditional media could overrepresent Black people in particular narratives through repeated editorial selection, today’s information environment creates the possibility of overrepresentation through repeated algorithmic selection. We might call this algorithmic overrepresentation.

The principle is deceptively simple: something does not have to occur more frequently to appear more frequently. A controversy does not have to become more important to become more visible. It merely has to become more engaging.

A Black public figure can legitimately make a mistake, face an allegation or become involved in a controversy. The initial reporting may be entirely justified. But once the story enters an engagement-driven environment, its prominence is no longer determined solely by editors or its underlying news value. It is also influenced by how effectively the subject generates clicks, comments, shares, watch time, arguments and reactions.

The historical problem was partly about who controlled representation. The contemporary problem is also about what controls distribution.

Diane Abbott provides an important bridge between those two eras. In 1987, she became Britain’s first Black woman MP. Three decades later, Amnesty International examined approximately one million tweets mentioning 177 women MPs during the six weeks preceding the 2017 general election. Abbott alone received about 45 per cent of all the abusive tweets directed at the women MPs studied. She received roughly ten times more abuse than any other woman MP in the analysis. Even when Abbott was removed from the figures, Black and Asian women MPs continued to receive disproportionately more abuse than their white colleagues.

That does not mean criticism of Diane Abbott is racist. Politicians should expect robust scrutiny, particularly those occupying prominent positions. It is that one Black woman could become the focal point for such an extraordinary concentration of hostility, much of it racialised and gendered. The crucial word is volume. Volume is precisely what digital platforms are exceptionally capable of manufacturing.

Nor was Abbott an isolated case. In the run-up to the 2024 general election, Labour’s Dawn Butler received a torrent of racist abuse after posting a light-hearted campaign video in which she rapped over So Solid Crew’s “21 Seconds”. Among the responses were images depicting monkeys, which were reported to the Metropolitan Police. Butler said Black women in public life were treated as “easy targets” for abuse. Research cited alongside her experience, conducted by the anti-online-abuse organisation Glitch, analysed almost one million messages across five social-media platforms and found that one in five posts about women was highly toxic, with Black women bearing the brunt of the most extreme material.

Butler’s experience is revealing precisely because of how ordinary the trigger was. This was not a major scandal or allegation of wrongdoing. A Black woman politician posted a campaign video and the racial abuse arrived anyway. It reminds us that the phenomenon cannot be explained entirely by the seriousness of the underlying controversy. For some Black public figures, visibility itself can be enough to attract racialised hostility.

Evidence published by Goldsmiths, University of London in 2025 broadens the picture further. Researchers surveying more than 800 Black and racially minoritised people aged 16 to 24 across the UK found that 95 per cent had encountered violent or abusive racist content online. Sixteen per cent encountered racist content daily and 38 per cent at least weekly. This is therefore not simply a story about what happens to famous Black politicians. Racist material has become a recurring feature of the digital environment encountered by many ordinary young people.

Taken together, Abbott, Butler and the Goldsmiths findings suggest a continuum. At one end is routine exposure to racist content. At the other is the extraordinary concentration of abuse that can descend upon a visible Black individual.

The question for the algorithmic age is what happens when systems designed to detect engagement encounter existing racial hostility and, by responding to that engagement, increase its visibility, velocity and reach. Volume, after all, is precisely what digital platforms are exceptionally capable of manufacturing.

And that brings us to Professor Jason Arday.

Then came Professor Jason Arday

Criticism of a Black person is not automatically racism, and scrutiny of a Black academic is not automatically discrimination. The concern is not whether Black public figures should be scrutinised, but whether that scrutiny is conducted fairly, proportionately and with dignity and whether they are subjected to a level, character and persistence of treatment that others would not reasonably be expected to endure.

Professor Jason Arday’s became fragmented into an extraordinary number of separately consumable stories about his qualifications, appointment, plagiarism allegations, childhood, disability, school grades, race and DEI. One individual increasingly became the focal point through which much larger political and cultural arguments could be conducted.

The news section on X, 8th August 2026 4pm, every one of the five visible news topics on screen concerned Professor Arday.

These were not five individual posts. They were five separately presented news stories about one Black academic. One topic displayed about 75,000 posts. Several others showed only hundreds.

Seventy-five thousand posts about one Black academic should itself give pause.

Professor Jason Arday was not a politician, celebrity or elected public official. He had committed no crime. He faced legitimate allegations concerning academic integrity, but what followed extended far beyond scrutiny of those allegations: racialised attacks, malice, threats, intensely personal abuse, ridicule and attempts to dehumanise him, alongside an extraordinary volume of commentary about almost every aspect of his life.

The scale raises an uncomfortable question. Was this what harmful algorithmic amplification looks like when the brakes fail? Not an algorithm inventing racism, but a system encountering a controversy saturated with racial grievance and outrage, detecting extraordinary engagement and continuing to accelerate its visibility rather than applying sufficient friction.

Recommendation systems do have brakes. Platforms can downrank material, remove posts, suspend accounts, restrict recommendations, demonetise creators and intervene against coordinated or harmful behaviour. The question raised by Arday is therefore not whether intervention is technically possible. It is why, amid such an extraordinary concentration of hostility around one individual, the information environment appeared to keep accelerating.

That is what requires investigation. How did one academic controversy generate tens of thousands of posts? How much was organic? How much came from repeat or coordinated accounts? How much was driven by news coverage and paid promotion? How much additional reach came from recommendation systems? And crucially, at what point did the platforms recognise that legitimate scrutiny had become something qualitatively different and what did their algorithms do then?

If the answer is that the systems simply continued rewarding whatever kept people engaged, then the Arday case exposes something more serious than an online pile-on. It exposes the possibility of an amplification system with brakes that failed to engage precisely when they were most needed.

Then the story became something to sell

The Arday controversy did not circulate solely through organic posts, news coverage and recommendation systems. Some news outlets promoted through paid social-media advertising.

The consequences become more complicated when paid promotion enters an already accelerating controversy. A publisher produces a story. Readers engage with it. Social platforms detect the interest. Commentators react. Searches increase. Publishers see heightened demand and commission further coverage. Some of that journalism is then promoted through paid advertising, creating additional impressions and potentially introducing the controversy to audiences who might not otherwise have encountered it.

The story can consequently move through several stages:

from reporting to trending, from trending to virality, and from virality to paid amplification.

At some point, the prominence of the story begins generating demand for the story. People click because they want to understand why Arday is everywhere. Their clicks demonstrate further audience interest. That interest makes further coverage commercially attractive. Further coverage makes Arday still more visible. Nobody has to intend to create a pile-on for the economics of the system to intensify one.

The platform wants engagement. The publisher wants readers and subscribers. The influencer wants reach. The commentator wants attention. The opponent wants to challenge what is being said. The curious user wants to understand what everyone else is talking about. Each has a different motivation, yet almost every action pushes in the same direction, more visibility.

There is some evidence that paid promotion of Arday stories occurred. Social-media posts have subsequently drawn attention to paid advertisements promoting coverage about him, including complaints that such advertising remained visible around the time of his death. That deserves further investigation rather than speculation about the motives behind individual campaigns.

What matters for this argument is the structural effect. Organic recommendation and paid distribution are different mechanisms, but to the person holding the phone they can produce the same experience Jason Arday everywhere on my feed

Then came the caricatures

The controversy also began changing form.

Alongside malicious comments, social media post, AI-generated or manipulated images mocking Arday, including representations of him as Pinocchio and other caricatures started to emerge. Political caricature is hardly new. What is new is the ease with which generative AI can industrialise it.

A finite collection of allegations can now produce an almost limitless supply of content. One photograph becomes a meme. The meme becomes an AI-generated image. The image becomes a reaction post. Someone produces another caricature. Another account adds a caption. A commentator turns it into a video.

The underlying information may barely have changed.

The packaging has.

This matters because every new format can renew attention around a story that might otherwise be approaching exhaustion. The attention economy requires novelty, and generative AI makes synthetic novelty extraordinarily cheap. There is something particularly powerful about visual ridicule. Someone who would never read 1,500 words about an academic plagiarism dispute can understand the insinuation of a Pinocchio caricature almost instantly. A complicated allegation has been compressed into a visual proposition: this man is a liar.

It requires almost no context to consume and almost no effort to share. But it demonstrates another way in which a finite controversy can continually be refreshed for an engagement economy.

Different content. Same person. Another opportunity to react and be incentivised.

The modern pile-on has an engine

This brings us to the question that deserves much greater investigation, not simply who participates in an online pile-on, but what makes a pile-on capable of reaching such extraordinary scale, speed and persistence in the first place.

Once political actors, commentators, publishers and opportunistic content creators recognise that a particular individual has become an unusually productive source of engagement, there are obvious incentives to keep producing material. If one Arday story performs strongly, another angle can be found. His academic record becomes one engagement object, his appointment another, his childhood another, his disability another, his school grades another and his race and DEI another. When conventional posts lose their novelty, memes and synthetic imagery can refresh the controversy. Paid advertising can extend its reach further still. But focusing only on those feeding the system risks overlooking the most powerful actor in the chain, the system itself.

Recommendation and ranking algorithms are not passive conduits through which human behaviour happens to flow. They are designed systems making billions of consequential decisions about visibility.
What appears in a feed, what trends, what is recommended next, which conversations are surfaced and how widely particular content travels. Their objectives, safeguards and thresholds are choices made by the companies that build and govern them. Lets be clear those choices can produce harmful outcomes.

A harmful algorithm does not require racist intent written into its code. Harm can arise when a system optimised around engagement repeatedly discovers that inflammatory, dehumanising or racialised material generates exceptional reaction and consequently gives that material greater visibility than it would otherwise have received.

That is algorithmic amplification, and it should be named when the evidence supports it. The distinction matters. A racist account may originate an attack. A provocateur may deliberately manufacture outrage. A publisher may produce another story. An advertiser may purchase additional reach. Users may click, condemn, defend and share. But the extraordinary scale at which these actions can become visible is not simply the spontaneous behaviour of a digital crowd.

There is an amplification architecture sitting between the crowd and what the crowd sees.

And technology companies control it.

This is why the familiar defence that a platform merely reflects what users are interested in is inadequate. Recommendation systems do more than reflect interest. They measure it, predict it, rank it and redistribute it. In doing so, they can turn hundreds of interactions into thousands of impressions and thousands into millions. What begins as human prejudice can therefore acquire computational scale.

Professor Arday’s treatment makes this distinction particularly important. The extraordinary volume of material surrounding one individual, multiple trending stories, commentary, racialised attacks, apparently AI-generated caricatures, memes and paid promotion, does not by itself prove that a particular algorithm intentionally targeted him. Establishing precisely how much additional reach recommendation systems produced requires access to platform data that outsiders largely do not possess.

But that opacity should itself concern us.

When a platform can determine what millions of people see while researchers, journalists and the public cannot adequately inspect why particular material was amplified, the absence of evidence about the algorithm cannot continually be treated as evidence that the algorithm played no meaningful role.

Harmful algorithms exist. Algorithmic discrimination has been demonstrated in moderation, advertising, image systems and other automated technologies. Recommendation systems therefore deserve the same scrutiny. Where evidence shows that a ranking system has significantly amplified racist, dehumanising or otherwise harmful material, it should be identified as an algorithmic failure rather than explained away as merely unfortunate user behaviour. Responsibility is shared, but it is not equal.

The provocateur may supply the outrage. Influencers may supply commentary. Publishers may supply stories. Advertisers may purchase additional reach. Generative AI may supply endless variations. Users may supply clicks, searches and replies.

But the platform built the engine.

It determines what the engine optimises for, what it accelerates, what it suppresses, what safeguards surround it and when intervention occurs.

That is why the modern pile-on is fundamentally different from the mob of an earlier media age. The crowd is no longer limited by geography, printing presses or broadcasting schedules. It can be assembled, personalised and continuously supplied with material by computational systems operating at enormous scale. Those seeking to exploit racial grievance may learn what to feed the machine.

But those who built the machine cannot escape responsibility for what it learns to amplify.

Playing to the tune of the algorithm

Strip away, for a moment, the sentiment of individual posts and another feature of the Arday controversy becomes visible, the mechanics of attention.

One highly critical post published after Arday’s death was long, provocative and subsequently pinned to the author’s profile. At the time captured, X displayed 3.4mn views, 30,000 likes, 4,800 replies and 4,700 reposts.

Those figures demonstrate precisely the kind of engagement pattern that deserves examination. The post does not simply communicate an opinion. Its length demands attention. Its language invites agreement and furious disagreement. Replies extend the conversation, reposts carry it into other networks and quote-posts create further arguments. Pinning it gives the post continuing prominence to anyone arriving at the account. In other words, the argument itself becomes an engagement object.

There is also a financial dimension. X’s Creator Revenue Sharing programme allows eligible creators to earn money from their content and explicitly says earnings are influenced by verified Home Timeline impressions, who views the content and its format. X says it rewards content that drives “meaningful interactions and conversation”.

There is no public evidence that this particular post generated revenue, and raw views cannot be converted into a reliable estimate of earnings. But the architecture matters, the same engagement being generated by an inflammatory controversy can, for eligible creators, carry financial value.

This is important because outrage does not require consensus to perform. Someone who agrees may like or repost. Someone appalled may reply, quote-post or share it in condemnation. Journalists and campaigners may reproduce it as evidence of the wider debate. Their intentions are completely different, yet each can contribute attention to the same object.

That is why engagement should never be confused with endorsement. And in an attention economy, outrage does not have to persuade everyone to succeed, it simply has to keep enough people looking.

Conclusion

Racism has never existed only in words or individual acts. It has travelled through institutions, systems and processes, shaping who is represented, who is believed and who is disproportionately scrutinised. The digital age did not invent that history. It has given it new infrastructure.

Algorithms do not hate. But neither do they operate outside society. They measure human behaviour, classify it, rank it and redistribute it. When racial grievance generates anger, argument and counter-outrage, an engagement-driven system can learn that the controversy is valuable without understanding the racism within it. The algorithm does not have to be racist for racism to become algorithmically advantageous.

Technology companies cannot present these systems as forces of nature. The digital town square is privately governed. Platforms design the algorithms, set the guardrails and determine what is moderated, recommended, monetised and amplified. Those choices are also made within a political environment in which governments, regulators and powerful interests exert pressure over the boundaries of acceptable speech.

The danger is what happens when the guardrails weaken while the amplification machinery remains. Freedom of expression matters, but a platform cannot credibly claim neutrality while its systems can transform racial hostility into valuable engagement. A report button or fact-checking note does not resolve the deeper problem when harmful content remains visible, reproducible and capable of further amplification.

Professor Jason Arday’s treatment should force difficult questions about that system. Legitimate scrutiny became entangled with racialised attacks, intensely personal commentary, paid promotion, memes and AI-generated caricatures. What requires investigation is how far algorithmic amplification contributed to turning a legitimate public-interest story into something approaching hyper-sensationalisation and digital dehumanisation i.e. Digital Lynching.

This is the historical rhyme. Where previous generations confronted the disproportionate representation of Black people through editorial selection, this generation must confront algorithmic overrepresentation, old prejudices travelling through systems capable of giving them unprecedented visibility, velocity and reach.

The algorithm did not need to hate Professor Jason Arday. It did not need to understand racism at all. It needed only to discover that the controversy surrounding race, identity and outrage kept people looking.

That is the danger. Racism once needed a voice. In the digital age, it has acquired an amplifier.

Rest in peace, Professor Jason Arday