The Algorithm Eats First: What Random Viral Videos Tell Us About Platform Power

On the morning of 18 May 2026, four short videos circulated on X within the span of a few hours. One showed a man on a scooter in San Diego trying to intervene during a bicycle theft, nearly losing his own vehicle in the process. Another captured a woman advancing on police officers with a ballpoint pen, absorbing two gunshots before a taser finally brought her down. A third offered a 30-second roadside encounter. A fourth presented itself as a hand-drawn confession: "I want to be famous." None of these clips shared a byline, a subject, a location, or a coherent thematic thread. Together, they constitute something nonetheless: a data point about the infrastructure of attention.

The pattern is not new, but it bears examining. What these videos share is not content but architecture. Each was posted to X, surfaced by an algorithm whose internal logic remains proprietary, amplified to audiences far beyond the communities in which the events occurred, and consumed in the brief window before the next item displaced it. The events themselves may have been real, partially real, or staged for the camera. The question of their authenticity, while not trivial, is secondary to a more structural point: X has become an unfiltered archive of human behavior at scale, curated by a machine that optimizes for engagement rather than meaning.

What the Platform Amplifies and Why

X, the service formerly known as Twitter, processes hundreds of millions of posts per day. Its recommendation algorithm, described in various public-facing technical documents and reverse-engineered by independent researchers, sorts this torrent into a small number of slots in each user's timeline. The signals it weighs include recency, engagement volume, the prior behavior of the viewer, and, crucially, the willingness of other accounts to amplify the content further. What this system rewards, reliably, is emotional arousal: conflict, danger, spectacle, absurdity.

The San Diego police footage fits this logic cleanly. It contains confrontation, physical threat, a weapon, and an outcome that is both disturbing and, for a certain kind of viewer, compelling. The scooter incident offers similar features — a bystander becoming a participant, the risk of violence escalating unexpectedly. The 30-second clip and the confessional animation operate on a different register but serve the same function: they break the pattern of ordinary life and invite a moment of focused attention before the next disruption arrives.

This is not a criticism of the users who posted or watched these videos. It is a description of a system. When a platform reaches sufficient scale, individual intent becomes nearly irrelevant to outcomes. A person posting a genuine moment of street drama and a person fabricating content for engagement both feed the same machine. The machine does not distinguish between them until enforcement mechanisms — reports, manual reviews, policy violations — are triggered, and those mechanisms are notoriously slow and inconsistent.

The Death of the News Feed

What the morning's content collection also illustrates is the collapse of the distinction between news and entertainment, between information and spectacle. Five years ago, a morning's worth of X content might have included breaking news, commentary, and personal updates in rough proportion to a user's follow graph. Today, for many users, the algorithm has subsumed the follow graph. The result is a feed that is less a newspaper and more a slot machine: each refresh offers a pull of the lever, with the payout being a brief hit of novelty.

The implications for journalism are rarely acknowledged directly, but they are structural. When the dominant platform for public discourse rewards content that provokes immediate emotional response, the incentive for careful, time-intensive reporting decreases. The San Diego video required no context, no verification, no follow-up reporting. It simply arrived, consumed its moment, and passed. The infrastructure of attention had already moved on by the time anyone might have asked whether the encounter was accurately framed or what preceded it.

This is not a new observation, but it bears repetition in specific form. The question is not whether viral content is good or bad. The question is what kind of public discourse a platform that optimizes for viral content produces, and who is served by that arrangement.

The Concentration of Curation

X is not the only platform that operates on these principles, but its current ownership structure gives the dynamic a particular sharpness. Since Elon Musk's acquisition in late 2022, the service has undergone significant changes to its content moderation infrastructure, its verification systems, and its relationship with advertisers. The net effect, as documented by multiple researchers studying platform behavior, has been an increase in certain categories of inflammatory content alongside a decrease in the human moderation capacity needed to contextualize it.

What the morning's videos demonstrate is that this environment does not require malice to produce strange results. The algorithm surfaced four clips that had nothing to do with each other, whose subjects were scattered across geography and circumstance, whose accuracy could not be independently verified from the footage alone. It surfaced them because they performed well on the metrics the algorithm is designed to optimize. The fact that no editorial judgment was applied — that no human being decided these four clips belonged together in the morning's content stream — is the point.

Stakes and What Remains Uncertain

The stakes of this dynamic are real but unevenly distributed. Ordinary users consume an attention environment shaped by forces they cannot see and do not control. Journalists and news organizations compete for a kind of attention — sustained, contextual, deliberative — that the platform's architecture systematically underserves. Platform companies, for their part, have little incentive to restructure their recommendation systems toward slower, more deliberative content when the faster, more arousing material reliably produces higher engagement metrics.

What remains uncertain, and what the morning's four videos cannot resolve, is whether alternatives exist at scale. Other platforms have experimented with chronological feeds, community curation, and different engagement metrics. None has approached X's reach in the public-domain political communication space, though several — Bluesky, Threads, Telegram — have absorbed significant migration from users seeking different terms of engagement. Whether any of these represent a durable alternative, or merely a temporary redistribution of the same underlying dynamic, is a question the sources do not resolve.

The woman with the pen, the man with the scooter, the anonymous road encounter, the anonymous animator: four fragments of a day, surfacing briefly in a global feed before submerging again. The algorithm does not remember them. The question is whether the public discourse that depends on it can afford to be shaped by a system designed to forget.