The front page of The New York Times on January 7, 2021, ran a single photograph above the fold: a Capitol Police officer with his hand on his holster, backing away from a mob in the building’s marble corridors. The headline read “Mob Storms Capitol; Trump Incites Chaos.” That same morning, a user scrolling Facebook in a suburb of Dallas might have encountered a video captioned “Patriots peacefully protest at the Capitol,” a post claiming antifa infiltrators had staged the violence, or nothing about the event at all because the algorithm had learned they preferred cooking videos and grievance threads about mask mandates. Three Americans, same country, same morning, same event, experienced three incompatible realities.
The political middleman has changed. The journalist who once stood between event and audience, applying judgment about significance, proportion, and context, has been displaced by a system that optimizes for something else entirely. The replacement is not a different editor with different politics. It is not an editor at all.
What Journalists Actually Did
The old gatekeeping function operated through visible rules. Editors weighed timeliness against proportion, prominence against proximity, conflict against consequence. A newspaper’s choices were inspectable: the front page announced its priorities openly, the editorial page carried signed arguments, and corrections ran in a standard box. The BBC‘s editorial guidelines run to 300 pages and are publicly available. The Times masthead names human beings who can be interviewed, criticized, or sued. This transparency did not eliminate bias, but it made bias legible. A reader could learn to read Le Monde against Figaro, understanding that each applied a recognizable filter.
The system had obvious failures. News values systematically overrepresented official sources, privileged drama over complexity, and treated politics as a spectator sport. But the failures were at least collective. When the entire profession misjudged the evidence for weapons of mass destruction in Iraq, the error was visible, debatable, and eventually documented. The shared baseline, however flawed, made disagreement possible.
What Algorithms Optimize For
Algorithmic curation answers to different masters. Engagement metrics, time-on-platform, and ad impression volume drive the sorting mechanism. The system does not ask whether a piece of political information is true, significant, or proportionate. It asks whether it will provoke a click, a share, a comment, or a return visit. Content that triggers strong emotional response performs reliably better by these measures, so the system systematically elevates what agitates over what informs.
The mechanism is opaque by design. The specific weighting of variables in Facebook’s ranking system, the exact signals that cause a YouTube video to surface or vanish, and the personal data points that shape an individual’s feed are proprietary trade secrets. The algorithm evolves continuously through A/B testing, its behavior shifting without announcement or explanation. A newspaper’s editorial line could be summarized and criticized, but no human can fully articulate why any particular user sees any particular post. Frances Haugen’s whistleblower testimony revealed that Facebook’s 2018 shift toward “meaningful social interactions” inadvertently amplified divisive content because outrage generates engagement. The company did not intend this outcome, but the algorithm’s logic made it predictable.
The personalization compounds the opacity. Two neighbors on the same street, connected to the same platform, receive divergent feeds shaped by their click histories, their demographic inferences, and their social networks. There is no universal “editorial line” to examine. Each user inhabits a private information environment that is unaccountable and, for practical purposes, unaccountable-to.
The Same Event, Three Different Movies
The January 6 Capitol attack illustrates the divergence with painful clarity. Traditional outlets framed the event through investigative reporting: documenting the violence, tracing the planning, examining the security failures, and reporting the political fallout for the certification of electoral votes. The narrative was contested in its interpretation but grounded in shared facts.
Algorithmic channels fragmented even the factual baseline. Some users encountered the “Stop the Steal” narrative, with claims of election fraud presented as established truth. Others saw the event downplayed as a peaceful protest with isolated bad actors. Others received the false flag theory, in which antifa or federal agents had staged the violence to discredit conservatives. The platforms’ design, with its emphasis on engagement, allowed each of these framings to find its audience and intensify within it. The algorithm did not create the conspiracy theories, but its logic of amplification and personalization ensured that contradictory versions of reality hardened in separate silos, each reinforced by repetition and social proof.
The Structural Difference, Not a Moral Scorecard
This analysis does not restore journalism to a pedestal it never occupied. Editorial gatekeeping produced its own distortions: the overrepresentation of powerful voices, the chase for sensationalism within professional constraints, and the blind spots of class and geography. The Daily Mail and RT operate through editorial selection too, and their criteria are no more publicly defensible than Meta’s engagement metrics.
The relevant distinction is structural. Journalistic curation, however flawed, operates through human judgment that can be examined, criticized, and held responsible. Algorithmic curation operates through automated optimization that resists examination by its nature. The first system makes its values visible even when they are bad values. The second system obscures its values behind technical complexity and commercial secrecy, making it difficult even to identify what is being optimized for, let alone to contest it.
The consequence for public understanding is severe. A society cannot deliberate together when its members receive incompatible accounts of what is happening. The erosion of shared factual reference points, the intensification of ideological segregation, and the declining trust in any information source that contradicts one’s feed follow not from any particular political bias but from the architecture of the distribution system itself. When the middleman is an algorithm, there is no one to ask why this story and not that one, no one whose judgment can be appealed, and no transparent process whose outcomes can be compared across communities.
The photograph on the Times front page was taken by a journalist who made a decision to be in that corridor at that moment. The cooking video that displaced it in someone’s feed was selected by a process no human directed and no human can fully explain. That is the shift. Everything else follows.