Independent New Hampshire Journalism · Northfield, NHCorrection, Sept. 19, 2026, 3:09 p.m. EDT: An earlier version of this column quoted Mark Zuckerberg calling Meta's fact-checking program "a mistake-ridden system that acted as censorship by proxy," said Facebook deleted about 83 million accounts, 6.4 percent of users, in a 2012 "fake account cleanup," and said University of Washington researchers found that detection algorithms "fail 30 percent more often." In fact, we found no record of Zuckerberg using those words; the coverage the column cites quotes him saying the fact-checkers "have been too politically biased." In an August 2012 securities filing, Facebook estimated that about 8.7 percent of its 955 million active users, about 83 million accounts, were duplicate, misclassified or "undesirable" accounts; the filing did not describe a deletion. The University of Washington team reported that bots written with large language models reduced the performance of existing detectors by as much as 30 percent (ABC News/CNBC; University of Washington). We corrected those passages, sourced the claim about uneven enforcement to a 2024 study in Nature, removed an unsupported account count for Mexico's "Peñabots," and fixed citation titles and dates.
By Granite State Report
In the beginning, social media sold itself as a democratic miracle — a place where anyone could speak truth to power. But that promise rotted fast. Behind the memes and trending hashtags, an invisible war is raging between bots, algorithms, and human moderators armed with delete keys. It’s a battle not just for attention — but for control of the public mind.
The Bots That Never Sleep
Automated accounts are active across X (formerly Twitter) and Facebook. They don’t sleep, they don’t scroll, and they never stop posting. A 2017 study by Onur Varol and colleagues estimated that 9 to 15 percent of active Twitter accounts were bots. Bots carried an even larger share of link sharing: a 2018 Pew Research Center analysis found that two-thirds of the tweeted links it examined were shared by suspected bots.
Most bots aren’t Russian spies or digital saboteurs. Many are simple automation scripts pushing weather updates, stock prices, or sports scores. But hidden in the noise are networks with far darker motives — designed to manipulate conversation, boost propaganda, or bury inconvenient truths.
In August 2024, University of Washington researchers reported that bots using large language models reduced the performance of existing bot detectors by as much as 30 percent. The same team found that a detector built on a large language model outperformed existing systems by 9 percent. These aren’t clumsy spam accounts anymore — they’re digital ghosts engineered to blend in, influence discourse, and tip algorithms toward chaos.
Censorship Disguised as Moderation
Big Tech doesn’t just fight the bots; it controls the battlefield. What used to be “content moderation” — a technical process of enforcing terms of service — has evolved into something murkier: algorithmic censorship.
Facebook, YouTube, and X all maintain policies allowing them to suppress or “downrank” content that violates their community standards, even if it’s not illegal. That includes politics, public health, and satire. The platforms say it’s to prevent misinformation; critics say it’s to police thought.
Internal Meta emails from 2021, posted by House Judiciary Committee Chairman Jim Jordan in 2023, describe White House pressure on Facebook over COVID-19 content. One employee summarized what the White House wanted removed as including "humorous or satirical content that suggests the vaccine isn’t safe." In an Aug. 26, 2024, letter to Jordan, Mark Zuckerberg wrote that administration officials had "repeatedly pressured" Facebook for months to take down "certain COVID-19 content including humor and satire." In January 2025, Meta ended its third-party fact-checking program in the United States. Zuckerberg said the fact-checkers "have been too politically biased and have destroyed more trust than they created," and Meta's announcement said the program "too often became a tool to censor."
Fake accounts are not a new problem. In an August 2012 securities filing, Facebook estimated that about 8.7 percent of its 955 million monthly active users, roughly 83 million accounts, were duplicate, misclassified or "undesirable" accounts, the last group being accounts that break its rules, such as spammers.
The same story repeats across platforms: legitimate speech caught in digital dragnet operations, flagged as “bot-like” or “false” without explanation. The appeal process? Usually automated, opaque, and slow.
The Misinformation Wars
Both right and left accuse each other of being algorithmically silenced — and both are sometimes right. Researchers studying Twitter have found uneven enforcement. A 2024 study in Nature by Mohsen Mosleh, David Rand and colleagues examined 9,000 politically active Twitter users during the 2020 presidential election and found that users estimated to be pro-Trump or conservative were substantially more likely to be suspended. The same users also shared far more links to low-quality news sites, and the authors concluded that "political imbalance in enforcement need not imply bias on the part of social media companies."
Meta and X both deny ideological bias, but they also refuse to release full moderation datasets for external audit. The opacity itself is the tell. When the rules are secret, enforcement becomes politics by other means.
To make things worse, the bots are often blamed for the same chaos that censorship creates. Disinformation campaigns flood the zone with noise; platforms then overreact, tightening moderation in the name of safety. In the process, ordinary users — journalists, activists, scientists — get flagged as “inauthentic actors.” The cure becomes the disease.
The Political Weaponization of ‘Bot’ Accusations
Calling your critics “bots” has become the digital equivalent of calling them traitors. Governments from Mexico to the United States have used bot allegations to delegitimize dissent. In Mexico, analysts gave the name Peñabots to a suspected network of automated accounts used to spread pro-government messages under President Enrique Peña Nieto, flood hashtags critical of the government, and push smear campaigns against journalists and activists.
In the U.S., labeling opponents as “bots” or “Russian assets” became routine during election cycles. But as machine learning gets better at imitating human tone, distinguishing between authentic and synthetic discourse is nearly impossible without full transparency from the platforms — which is exactly what they refuse to give.
The New Gatekeepers
Here’s the uncomfortable truth: Silicon Valley didn’t kill free speech with censorship alone. It killed it with control. Algorithms decide what we see, what we share, and ultimately, what we believe.
If 10,000 real users post something critical of a government policy but an algorithm quietly de-ranks those posts as “low quality,” does that criticism even exist? When AI-driven bots can flood the zone with synthetic consensus, the line between public opinion and engineered perception collapses.
In that vacuum of trust, disinformation thrives — not because people are gullible, but because no one can tell what’s real anymore.
What Comes Next
Transparency is the only antidote. Independent audits, open moderation logs, and algorithmic explainability should be the new baseline for digital governance. Without them, we’re not citizens in a democracy of ideas — we’re data points in a behavioral experiment.
The platforms insist they’re fighting for safety and truth. But as the record shows, truth doesn’t need a censor, and safety doesn’t need secrecy.
What’s being protected isn’t the user. It’s the narrative.
The machines that shape your feed don’t just deliver information — they decide what’s allowed to exist. Inside the world of bots, shadow bans, and digital censorship — where free speech is filtered by code and consent is manufactured by algorithm.
Citations
Varol, O., et al. “Online Human-Bot Interactions: Detection, Estimation, and Characterization.” arXiv:1703.03107 (2017). arxiv.org
Pew Research Center. “5 things to know about bots on Twitter.” (April 9, 2018). pewresearch.org
University of Washington News. “Large language models can help detect social media bots — but can also make the problem worse.” (Aug. 28, 2024). washington.edu
PBS News. “Zuckerberg says the White House pressured Facebook to ‘censor’ some COVID-19 content during the pandemic.” (Aug. 27, 2024). pbs.org
NBC News. “Republicans target Biden White House’s emails with tech platforms.” (Aug. 16, 2023). nbcnews.com
Yahoo News. “Zuckerberg announces end to Facebook’s third-party fact-checking, admits model became a tool for censorship: ‘Too many mistakes’.” (Jan. 7, 2025). yahoo.com
Meta. “More Speech and Fewer Mistakes.” (Jan. 7, 2025). about.fb.com
ABC News / CNBC. “Facebook: About 83 million accounts are fake.” (Aug. 3, 2012). abcnews.com
Mosleh, M., Yang, Q., Zaman, T., Pennycook, G., and Rand, D.G. “Differences in misinformation sharing can lead to politically asymmetric sanctions.” Nature (Oct. 2, 2024). nature.com
Wikipedia: Peñabot. en.wikipedia.org
Corrections: Granite State Report corrects verified errors promptly and appends a dated note identifying what changed and when. The corrections policy and the corrections register are public, and correction notes are never removed.
