When violent unrest swept across cities in the summer of 2024, attention quickly turned to the streets. But much of the real action was unfolding elsewhere: on X, the platform formerly known as Twitter. As rumours went viral, protest routes were crowdsourced in real time and doctored videos fuelled tit-for-tat reprisals, the crisis exposed the extent to which a single social media company could shape – and distort – democratic life.
In this article,researchers from the London School of Economics and Political Science dissect how X functioned not merely as a mirror of public anger,but as an accelerant and organiser of it. Drawing on digital trace data, platform policy documents and first-hand accounts, they identify four key mechanisms through which X undermined democratic norms during the 2024 summer riots: amplifying disinformation, eroding trust in institutions, enabling coordinated harassment, and incentivising polarising content.
Far from an certain by-product of online communication, these dynamics were closely tied to design choices, moderation regimes and business incentives. Understanding them is essential,the authors argue,not only for evaluating X’s role in the 2024 unrest,but for confronting the broader question of how platform governance can protect,rather than imperil,democracy in moments of crisis.
How opaque algorithms fanned the flames of unrest during the 2024 summer riots
At the height of the disturbances, the platform’s recommendation systems acted like accelerants on dry kindling, auto-curating a parallel facts universe in which the most alarming, polarising and emotionally charged posts were pushed to the top of feeds. As these systems operate as black boxes, neither users nor regulators could see why a misleading video of a burning building went viral while verified corrections languished unseen. Rather of surfacing calm, contextual reporting, the algorithm disproportionately rewarded engagement signals – rapid comments, quote-posts and repeat shares – which are structurally biased towards outrage. In practise, this meant that a small cluster of incendiary accounts could, with minimal content, dominate the attention economy of entire cities.
The opacity of these systems also obscured meaningful accountability. Officials could not know whether critical safety information was being throttled, local journalists could not understand why their on-the-ground updates were buried, and affected communities had no tools to contest the amplification of content that misrepresented them.Behind the scenes, automated moderation models struggled to distinguish between documentation and incitement, often removing testimonies from residents while leaving decontextualised clips untouched. The result was a feedback loop in which rumours were rewarded and nuance penalised, systematically privileging those willing to post first and verify later.
- Emotion over accuracy: Posts tagged with anger and fear travelled faster than verified updates.
- Speed over context: Real-time clips outran slower, fact-checked explanations.
- Engagement over safety: High-risk content was algorithmically promoted if it kept users scrolling.
| Content type | Typical reach | Democratic impact |
|---|---|---|
| Verified local reporting | Low-medium | Informed, slower reactions |
| Unverified riot footage | Very high | Escalated fear and retaliation |
| Official safety notices | Low | Limited crowd de-escalation |
Coordinated disinformation campaigns and the erosion of public trust in democratic institutions
What made the 2024 summer riots uniquely volatile was not just the speed of false claims but their orchestration. Networks of anonymous accounts-some automated, some coordinated by small, well-organised groups-amplified identical narratives within minutes of each other, exploiting the platform’s algorithmic preference for outrage. Fabricated videos of “state-sanctioned violence”, doctored screenshots of leaked government memos and pseudo-legal threads claiming that emergency regulations were “secretly suspended” circulated concurrently in multiple languages. The result was a manufactured sense of consensus: citizens did not merely see rumours; they saw what looked like overwhelming confirmation of those rumours from countless,apparently unrelated sources.
This atmosphere of synthetic unanimity eroded confidence in the very idea of neutral democratic referees. When fact-checkers, public broadcasters or electoral commissions attempted to intervene, they were swiftly rebranded-via coordinated hashtags and viral quote-posts-as partisan actors in a rigged game. On X, users were nudged to choose between “the people” and “the institutions”, a framing that made democratic checks and balances look like obstacles rather than safeguards. The feedback loop was brutal:
- Every correction was cast as elite censorship.
- Every institutional delay was reframed as proof of a cover-up.
- Every call for calm was depicted as an attempt to silence “real citizens”.
| Disinformation Tactic | Targeted Institution | Public Effect |
|---|---|---|
| Fake legal “explainers” | Courts & judiciary | Perception of biased justice |
| Edited protest footage | Police & local authorities | Heightened fear and anger |
| Fabricated poll numbers | Parliament & parties | Belief that votes “no longer count” |
| Impersonation of officials | Election regulators | Mistrust in official announcements |
The failure of platform governance and content moderation in moments of political crisis
As the streets filled with protestors and counter-protestors,the platform’s governing systems buckled under pressure,exposing how fragile its safeguards really were. Emergency escalation channels malfunctioned,regional trust and safety teams were understaffed,and automated detection tools were tuned for “normal” times,not for the velocity and volatility of a nationwide riot. Moderation queues overflowed, leaving reported content – including calls to violence and doxxing – untouched for hours, precisely when speed mattered most. Internally, risk teams had long warned that centralised decision-making, erratic leadership, and opaque policy changes had created a governance vacuum. When the crisis hit, there was no coherent framework for prioritising harms, only improvised reactions and last-minute reversals that signalled confusion rather than control.
- Policies rewritten on the fly,without clear public announcements
- Safety staff sidelined in favour of political and PR calculations
- Appeals channels clogged,leaving harmful posts visible for days
- Patchwork enforcement varying wildly by language and location
| Issue | Platform Response | Democratic Impact |
|---|---|---|
| Incitement to violence | Delayed takedowns | Escalated street clashes |
| Election rumours | Inconsistent labels | Eroded trust in results |
| Targeted harassment | Minimal intervention | Silenced civic voices |
The net effect was a form of algorithmic abdication: recommendation systems continued to amplify sensational and polarising posts while human oversight faltered. Civic groups struggled to get verified information boosted, while fringe accounts exploited loopholes to monetise outrage and conspiracy. The platform’s own design incentivised speed over accuracy, rewarding early, inflammatory narratives that were almost unachievable to correct later. In this vacuum, opportunistic actors – from organised extremists to opportunistic politicians – used the site as a megaphone to redefine the riots in real time, shifting blame, stoking ethnic tensions, and normalising suspicion of electoral institutions. Governance failed not only because rules were weak or badly enforced,but because the architecture of attention was fundamentally misaligned with democratic resilience.
Policy reforms and accountability measures to safeguard democracy from future platform-driven unrest
Preventing a repeat of the summer’s turmoil requires moving beyond voluntary “trust and safety” pledges toward enforceable obligations that reflect social media’s infrastructural power. Legislators could tie platform responsibilities to risk thresholds-as an example, stricter rules for services that amplify real-time political discourse at scale-while mandating self-reliant algorithmic and data audits overseen by public-interest bodies rather than industry-kind panels. This might include legally binding crisis protocols: time-limited throttling of virality features during verified emergencies, rapid de-boosting of content flagged by trusted fact-checking partners, and mandatory openness reports within hours, not months. Crucially, reforms should protect free expression by focusing on process (how content is amplified, ranked, and targeted) rather than policing opinions, ensuring that accountability does not slide into censorship.
Effective safeguards also demand new lines of accountability that extend beyond opaque corporate boards.Governments could require major platforms to publish democracy impact assessments before rolling out product changes likely to affect political information flows, and to co-fund independent research hubs with guaranteed access to anonymised data. A practical framework might combine regulatory sticks with incentives, rewarding firms that demonstrably reduce harmful amplification while preserving pluralistic debate.
- Statutory transparency on moderation rules, appeal rates, and political content reach
- Independent audits of recommender systems and advertising tools during electoral cycles
- Public-interest APIs giving accredited researchers real-time access to aggregate data
- Civil and financial liability for reckless design choices that predictably fuel offline violence
| Measure | Primary Goal | Key Actor |
|---|---|---|
| Real-time transparency dashboards | Expose viral unrest narratives | Platform + regulators |
| Mandatory crisis playbooks | Slow escalation on the streets | Platform |
| Democratic oversight panels | Scrutinise algorithmic risks | Civil society |
| Sanctions for repeat failures | Deterrence and culture change | Regulators |
Final Thoughts
As the dust settles on the summer riots of 2024, one lesson stands out with disquieting clarity: the infrastructure of our public conversation is neither neutral nor peripheral. It is central to how crises unfold.
The case of X shows how, in moments of instability, a single platform can amplify falsehoods faster than facts, reward outrage over accuracy, and blur the line between authentic voices and coordinated manipulation. The four dynamics explored in this article are not glitches in an otherwise sound system; they are features of a business model that monetises attention, regardless of its democratic cost.
For policymakers, regulators and civil society, the implications are stark. Treating social media as a mere conduit for speech underestimates its power to set agendas, shape perceptions and tilt the playing field of democratic contestation. The question is no longer whether such platforms influence political outcomes, but whether liberal democracies can afford to let their core institutions be mediated by systems optimised for engagement rather than truth.
If the 2024 summer riots are a warning, they are also an opportunity. They invite a rethinking of platform governance, transparency and accountability, and a renewed focus on the public-interest obligations of digital intermediaries. Whether that leads to meaningful reform will depend not only on what governments and companies do next, but on how societies choose to value – and defend – the fragile conditions that make democracy possible.
