The Feed That Picked Sides

That was always the cleanest answer: Trending Topics is determined algorithmically. No human bias. No editorial thumb on the scale. Just math, surfacing what matters. Facebook said it for years, and the press largely printed it, because the alternative — that a company with two billion users was running a covert newsroom with underpaid contractors making editorial calls — was too inconvenient to examine closely.

Then Gizmodo published the story on May 9, 2016, and the myth burned down.

Former members of Facebook's Trending Topics team told reporter Michael Nunez that they had routinely suppressed conservative news stories from appearing in the module — even when those stories were organically trending. Topics like CPAC, Mitt Romney, Rand Paul, and stories from conservative outlets like Breitbart and the Daily Caller were regularly prevented from trending, regardless of actual user engagement. The contractors also described injecting stories that weren't trending organically — pushing Facebook's preferred narratives into the feed.

The algorithm didn't pick sides. The people running the algorithm did. And Facebook had been telling everyone otherwise.

i · the mechanics of plausible deniability

Here's what the system actually looked like, as reconstructed from the Gizmodo reporting:

Facebook employed a team of contractors — mostly recent journalism graduates — as "news curators." Their job was nominally to verify that algorithmically-detected trending topics were legitimate (not spam, not low-quality). But in practice, the role had expanded far beyond that brief. Curators had discretion to "inject" topics that weren't trending and to "blacklist" topics that were. They worked from style guides and internal procedures that gave them significant editorial latitude.

The contractors earned between $15 and $18 an hour. They had no formal journalism training required. They worked in rotating shifts covering trending news for one of the most powerful media distribution platforms in human history.

The irony of this architecture is not subtle: Facebook structured the system to be plausibly deniable as editorial while being operationally editorial in every way that mattered. The contractors were doing what editors do. They were just classified as contractors doing "curation" — a word that sounds more like library science than journalism — so Facebook could maintain the fiction that it was running a platform, not a publisher.

This distinction matters enormously. Publishers are held to editorial standards, legal liability for content, and public accountability for what they amplify. Platforms are neutral conduits. Facebook wanted the distribution power of a publisher with the accountability of a plumbing fixture.

The system was designed to make that claim possible, not to make it true.

ii · the meeting that changed nothing

Nine days after the Gizmodo story, on May 18, 2016, Mark Zuckerberg held a meeting at Facebook headquarters with a group of prominent conservative media figures. In attendance: Glenn Beck, Tucker Carlson, Dana Perino, Barry Bennett (a Trump campaign adviser), and others. The message Facebook sent by convening this gathering was clear: we take your concerns seriously, we want dialogue, we are listening.

The meeting lasted about two hours. Zuckerberg reportedly acknowledged that Facebook had an obligation to be a platform for all ideas. There were discussions about trust and transparency. Photos were taken.

Nothing structural changed.

The Trending Topics curation team wasn't disbanded. The contractor model wasn't reformed. The style guides that gave curators discretion to suppress or inject content weren't published. Facebook didn't commit to external audits of the curation process, didn't change its contractor classification practices, didn't fundamentally alter the incentive structure that had produced the problem in the first place.

What changed was the optics. The meeting was the response. And the media largely covered it as a response — as if gathering conservative leaders for a listening session constituted action rather than performance.

This is the template Facebook would use for the next decade. Every scandal, every Senate hearing, every Cambridge Analytica revelation, every election interference report: convene some meetings, express concern, announce internal reviews, change nothing structurally. The political cost of the scandal is paid with spectacle; the business model continues unchanged.

iii · what the algorithm actually amplifies

The partisan framing of the 2016 controversy obscured something more interesting.

The real story wasn't that Facebook's contractors suppressed conservative content out of ideological malice — though that's what made it politically explosive. The real story was that Facebook had built a system where editorial discretion was inevitable, then denied that any such system existed. The biases that entered that system — whether ideological, cultural, commercial, or simply the result of which stories were easiest to handle on a given shift — were structurally guaranteed.

Every curation system has biases. This is not a controversial claim in media theory; it's definitional. When humans (or algorithms trained by humans) make decisions about what surfaces and what doesn't, those decisions reflect the values, incentives, and blind spots of the decision-makers. The question is never whether a platform has an editorial perspective. It's whether it's honest about that perspective and accountable for it.

Facebook was neither. The company had spent years insisting it was a neutral distributor of content, an infrastructure layer that merely reflected what people wanted to see. When the internal mechanics became visible, that claim collapsed — not just because contractors had suppressed some conservative stories, but because the entire framing of algorithmic neutrality was revealed as a convenient fiction.

The field wasn't distorted by the contractors' decisions. It was distorted by the original lie about what the system was. The editorial choices happening inside Facebook's curation team were the actual signal. The "algorithm" framing was the noise — a narrative overlay designed to prevent accountability by making the system's outputs appear to have no author.

iv · the decade that followed

It's easy, in retrospect, to see May 2016 as the moment the platform bias narrative entered American political life and refused to leave.

By 2020, "big tech censorship" had become a durable political talking point, a major driver of support for alternative platforms, and a basis for regulatory proposals. By 2022, Elon Musk's acquisition of Twitter was partly justified as a corrective to perceived platform bias. The grievance that Gizmodo's story surfaced — they say it's neutral but it isn't — became a load-bearing argument in a significant political movement.

The structural critique embedded in that grievance was legitimate. The political operationalization of it was something else entirely.

What got lost in the decade of platform bias discourse was the more fundamental question the Gizmodo story raised: not which side is being suppressed, but why are we comfortable with a single private company making covert editorial decisions that affect political discourse for billions of people?

The conservative pressure successfully moved Facebook toward giving more favorable treatment to conservative content in subsequent years — not because that was sound editorial judgment, but because the political cost of not doing so became unacceptably high. Which meant the correction to editorial bias was itself politically driven editorial decision-making. The system didn't become more neutral. It became differently biased in response to power pressure.

Facebook demonstrated, perhaps inadvertently, that "the algorithm" was a negotiation all along — with users, with advertisers, with governments, with powerful stakeholders. Whoever applied sufficient pressure could shift what surfaced and what didn't. The Trending Topics story was the first public proof of concept.

v · what doesn't get fixed

The lesson Facebook didn't learn — and I've watched it refuse to learn across every subsequent scandal — is that systems of this scale require structural honesty, not performative accountability.

The problem wasn't that contractors made editorial calls. Editorial calls are inevitable; someone always makes them. The problem was that Facebook had built its entire market position and public legitimacy on the claim that no one was making editorial calls — that the distribution of information was a neutral technical process. That claim was worth billions in reduced regulatory exposure and public trust. Abandoning it, even after the Gizmodo story, would have cost too much.

So instead: a meeting with Tucker Carlson. A commitment to "listen." An announcement that Facebook would update its guidelines for curators. The structure that made the problem possible remained intact. The fiction of algorithmic neutrality was repaired and redeployed.

I've watched this failure mode play out again and again since then. The platform announces it takes concerns seriously. There's a review. Guidelines are updated. The underlying incentives don't change. The next scandal emerges from the same structural conditions as the last one.

The feed picked sides in 2016. It just took a while to admit who was doing the picking.

vi · sources

• Former Facebook Workers: We Routinely Suppressed Conservative News — Gizmodo, 2016-05-09
• Mark Zuckerberg Meets with Conservative Media Leaders at Facebook HQ — The Verge, 2016-05-18
• Facebook's Fight Against Fake News — The Atlantic, 2016-05-19