Remove the Material
The invaders were Chinese until the edit. The Pentagon's contract says "remove the material". A jacket lost two flags, then got them back.
A troll factory with a daily quota, 448 million posts that never argue and an Army sign-up form open to 12-year-olds: how states take the room by filling it
According to a former paid Russian troll, the shifts ran 12 hours, around the clock, and each troll had a daily quota of 135 comments of at least 200 characters. That detail sits in a 2016 RAND paper, “The Russian ‘Firehose of Falsehood’ Propaganda Model”, a page before this line:
Quantity does indeed have a quality all its own.
Two sentences later it says why. High volume, RAND writes, “can consume the attention and other available bandwidth of potential audiences, drowning out competing messages.”
This is the last of three posts on how states sell a story. The first was about borrowing other people’s names. The second was about editing what already gets made. This one is about the method that needs neither, the one behind the Kremlin’s twenty-year project: saying it so often, on so many channels, that nothing else fits. I assumed the research would show the flood changing minds. I got that half wrong. The volume is real and it is cheap. The evidence that it persuades anyone is thin. What it does take is attention, and somebody is paid to take it.
The Quota
RAND’s paper is careful about its own status. The subtitle is “Why It Might Work”. It describes four features of Russian propaganda since about 2008: high volume across many channels, speed and repetition, no commitment to objective reality, and no commitment to consistency. Then it explains, from the psychology literature, why each might be effective. It is a hypothesis from 2016, not a measurement.
Some of the evidence it gathers is very good. RT, it notes, “claims more than a billion page views. If true, that would make it the most-watched news source on the Internet.” And it describes a leaked video in which a “reporter” is supposedly on a battlefield in Donetsk until someone switches on the light, and she is revealed “to be in a darkened room with explosion sounds playing in the background”.
1. The quota
“the trolls are on duty 24 hours a day, in 12-hour shifts, and each has a daily quota of 135 posted comments of at least 200 characters.”
One former paid troll's account, as cited by RAND.
RAND, The Russian "Firehose of Falsehood" Propaganda Model, 2016
2. Why volume
“First, high volume can consume the attention and other available bandwidth of potential audiences, drowning out competing messages.”
RAND's subtitle is "Why It Might Work". The paper argues from the psychology literature and measures nothing itself.
3. The shifts
“The specialists were divided into day-shift and night-shift hours and instructed to make posts in accordance with the appropriate U.S. time zone.”
United States v. Internet Research Agency et al., indictment, 2018
A five-year-plan production board, drawn by me. The figures are one former troll’s account as cited by RAND; “day-shift and night-shift hours” and the clippings are from the 2018 indictment of the Internet Research Agency.
China’s version is bigger and quieter. In 2017 Gary King, Jennifer Pan and Margaret Roberts published an analysis of a leaked archive from one district propaganda office in Jiangxi province, and scaled it up. They estimate that “the government fabricates and posts about 448 million social media comments a year.” That is an estimate from 2013 data, and they say so. The posts, they found, avoid arguing with critics altogether. Most “involve cheerleading for China, the revolutionary history of the Communist Party, or other symbols of the regime.” The paper’s title calls it “strategic distraction, not engaged argument”.
The operation has a famous nickname, the 50-cent army, after the rumoured fee per post. The researchers “found no evidence that 50c party members were actually paid 50 cents or any other piecemeal amount.” The posts seem to come from government employees doing it part time, outside their regular jobs. The 50-cent army is not paid 50 cents.
1. About 448 million a year
“We estimate that the government fabricates and posts about 448 million social media comments a year.”
An estimate, scaled up from a leaked archive of one district propaganda office in Jiangxi, 2013 data.
King, Pan and Roberts, American Political Science Review, 2017
2. Cheerleading
“most of these posts involve cheerleading for China, the revolutionary history of the Communist Party, or other symbols of the regime.”
3. Change the subject
“the goal of this massive secretive operation is instead to distract the public and change the subject”
4. Fifty cents
“We also found no evidence that 50c party members were actually paid 50 cents or any other piecemeal amount.”
五毛, wǔ máo, fifty cents: the operation's nickname, after a rumoured fee per post.
A wall of big-character posters, drawn by me. The characters are 五毛, “fifty cents”, the operation’s nickname; every English line is from King, Pan and Roberts.
Of those 448 million posts, about 212 million go onto commercial social media, into a stream the authors put at around 80 billion posts. That is about a quarter of one per cent of everything posted there.
What the Flood Buys
This is where I expected the frightening part, and mostly didn’t find it.
The best measured case is Russia’s campaign on Twitter during the 2016 US election. A team led by Gregory Eady matched a panel of 1,496 Americans to what they actually saw, and published the result in Nature Communications in 2023. Exposure was heavily concentrated: “only 1% of users accounted for 70% of exposures”, mostly people who already identified strongly as Republicans. The Russian posts were “eclipsed by content from domestic news media and politicians”. And on 17 of the 18 attitudes and behaviours they measured, the authors could rule out any relationship bigger than 0.2 standard deviations. An earlier study by Christopher Bail and colleagues, on 1,239 Twitter users in late 2017, found “no evidence that interaction with IRA accounts substantially impacted 6 distinctive measures of political attitudes and behaviors”.
1. 1% of users
“only 1% of users accounted for 70% of exposures.”
A panel of 1,496 Americans matched to what they actually saw on Twitter in 2016.
2. Drowned out
“exposure to the Russian influence campaign was eclipsed by content from domestic news media and politicians.”
3. What it changed
“we find no evidence of a meaningful relationship between exposure to the Russian foreign influence campaign and changes in attitudes, polarization, or voting behavior.”
From Eady et al., Nature Communications, 2023. Chart mine; the clippings are cut from the paper, which is published under CC BY 4.0.
Both studies are about Twitter, both cover a few weeks or months, and neither can say anything about other platforms. But the authors of the first one offer a guess about where the real damage went. The campaign, they write, “may have had its largest effects by convincing Americans that its campaign was successful.”
That is the liar’s dividend from the other side. The flood does not need you to believe it. It needs you to believe everyone else does, or to stop believing anything.
There is one counter-signal, and it is recent. A preprint by Trevor Incerti, Jonathan Elkobi and Daniel Mattingly, posted in January 2026 and not yet peer reviewed, showed about 8,600 Americans short TikTok videos about China. Content from pro-China influencers raised favourability toward China. Content from Chinese state media did not; in one version of the test it slightly lowered it. Neither moved people on a petition about Tiananmen. So the flood can persuade, a little, when it arrives in someone else’s voice. That is post one’s argument again.
The Multiplier
States are not the only ones choosing what you see. The platforms rank it.
In 2022 Ferenc Huszár and colleagues at Twitter found that “in six out of seven countries studied, the mainstream political right enjoys higher algorithmic amplification than the mainstream political left”. They did not find a bonus for the far right over moderates, and they say they don’t know the mechanism. In February 2026 a randomised experiment in Nature, by Germain Gauthier, Roland Hodler, Philine Widmer and Ekaterina Zhuravskaya, went further. They randomly assigned American X users to the chronological feed or the algorithmic one for about seven weeks in 2023, and 4,965 saw the study through. The algorithmic feed carried 60% more posts from conservative accounts. Users moved from the chronological feed onto it became 4.7 percentage points more likely to prioritise issues associated with Republicans, 7.4 points less likely to view Zelensky positively, and measurably more pro-Kremlin, by 0.12 standard deviations. Moving users the other way, off the algorithm, had no comparable effect.
The effect is real, and it comes from a company’s own ranking system; no state was involved in the experiment. The larger studies of Facebook in 2020 found small or no measurable effects on polarisation from changing the feed. Read together, the effects are small, contested and different on every platform. What the algorithm decides is how much of the flood reaches you. Nobody has shown it was built for the purpose.
The Recruiter in the Chat
The US government floods too. It just calls it marketing.
In March 2019 the Army’s assistant secretary for manpower and reserve affairs, Casey Wardynski, told a conference of the Association of the US Army why recruiting had to start earlier. “We have to confront this question of, will we wait until they’re 17, or will be start talking to them at age 12, 13, 14, 15, when they form the set of things they are thinking about doing with their life?” he said, according to Army Times. “If we wait until they’re 17 or 18, we will not be the first impression.”
The Army had already built a first impression. America’s Army, a free online shooter launched in 2002 as a recruiting tool, had cost $32.8 million by 2009, according to figures obtained through a freedom of information request. By 2020 it had an esports team streaming on Twitch, and the Army said it had spent $1.5 million on the team so far.
In July 2020, The Nation reported that pop-ups in the team’s chat offered a chance to win an Xbox Elite Series 2 controller, worth “upward of $200”. The link led to a recruiting form, “with no additional mention of a contest, odds, total number of winners, or when a drawing will occur.” The team’s sign-up form linked from its Twitter account let people “as young as 12” submit it, though a notice afterwards said recruiters could not contact anyone under 16. A Twitch spokesperson told Kotaku: “This promotion did not comply with our Terms, and we have required them to remove it.” Army Recruiting Command said each giveaway had its own tracking code connecting entrants to it. Army Times reported both.
1. The prize
“upward of $200”
Pop-ups in the Army esports team's Twitch chat offered a chance to win an Xbox Elite Series 2 controller.
2. Tracking codes
Army Recruiting Command said each giveaway had its own tracking code connecting entrants to it.
3. The form
“with no additional mention of a contest, odds, total number of winners, or when a drawing will occur.”
The giveaway link led to a recruiting form. The team's sign-up form linked from its Twitter account let people "as young as 12" submit it.
4. Twitch
“This promotion did not comply with our Terms, and we have required them to remove it.”
A Twitch spokesperson to Kotaku, as reported by Army Times.
5. The photo
A soldier at an Army esports competition in Ansbach. US Army photo, public domain.
A comic-book ad, drawn by me. The soldier is at an Army esports competition in Ansbach (US Army photo, public domain). The words are from The Nation and Twitch’s statement to Kotaku; the clippings are from the GAO’s report on military recruiting.
The same month, viewers started asking the stream about war crimes, and the team started banning them. Among them was Jordan Uhl, the progressive activist who wrote The Nation’s piece. The Army says the comments were “meant to troll and harass the team”. An ACLU lawyer told Vice: “It looks like what happened was a violation of the First Amendment.” After a pause of about five weeks the Army came back on 14 August and unbanned about 300 accounts. “We absolutely believe that we banned the users because of their behavior and not because of their viewpoints,” said Col. Megan Stallings, who commanded the Army’s marketing brigade.
Congress had a vote on it. On 30 July 2020, Alexandria Ocasio-Cortez proposed an amendment to stop the military using recruitment money on streaming and esports. It failed, 126 to 292. Every Republican who voted was against it, and so were 103 Democrats.
The Counter-Flood
Volume isn’t only a state tool. In 2022 a crowd of accounts with cartoon Shiba Inu avatars, calling themselves NAFO, the North Atlantic Fella Organization, started answering Russian officials online, mostly with mockery. One of them, the diplomat Mikhail Ulyanov, replied to a Fella that “You pronounced this nonsense, not me,” and then went quiet on Twitter for about a week, Vice reported.
The Fellas also raise money. Ukraine’s own fundraising platform, UNITED24, says they “raised as much as $276,429” over two months in late 2022, from more than 3,000 contributions, for one naval drone called Raccoon’s Revenge. Larger totals circulate, but they are self-reported, and this is the only one confirmed by the people who received it.
The Fellas are volunteers, so the comparison only goes so far. It does show the method works for whoever picks it up.
The Ledger
Every post in this series has had a price. Here is this one’s, with a label on every line, because the figures are not equally solid.
| What | Figure | How solid |
|---|---|---|
| Project Lakhta, Russia’s troll operation, actual spend 2016 | about 720m roubles (about $12m) | US prosecutors’ allegation |
| Project Lakhta, actual spend 2017 | about 733m roubles (about $12.2m) | US prosecutors’ allegation |
| Project Lakhta, proposed budget Jan 2016 to Jun 2018 | over 2bn roubles (over $35m) | US prosecutors’ allegation; proposed, not spent |
| Tenet Media, the US company in post one | nearly $10m | US prosecutors’ allegation |
| Russian federal budget for state media, 2026 | about 146bn roubles | Draft budget, as reported; mostly domestic |
| RT’s line in it | 32.08bn roubles | Reported |
| China, propaganda in traditional media, per year | over $1.5bn | Atlantic Council claim, no method given |
| China, “external propaganda”, per year | about $10bn | Unnamed analysts’ estimate, cited by Shambaugh, who says “Nobody knows for sure” |
| China Daily’s US spending, second half of 2019 | over $5m | Its own disclosures, via Freedom House |
| US military advertising and recruiting, FY2023 | $1.853bn, of which the Army $909m | Official (GAO); covers far more than esports |
| The Army’s esports team, to May 2020 | $1.5m | The Army’s own figure |
| NAFO, one naval drone | as much as $276,429 | Reported by the recipient |
Sources: Khusyaynova complaint, Tenet indictment, The Moscow Times, UNITED24 Media, Atlantic Council, Shambaugh, Freedom House, GAO, Military.com, UNITED24.
I read three things in it, and the reading is mine.
The covert flood is cheap. Lakhta’s budget, as prosecutors describe it, came to about 720 million roubles in 2016. The open one, Russia’s state media, is drafted at about 146 billion for 2026. Different years and different exchange rates, but the gap is more than a hundredfold.
The cheapest-looking line is the hardest to price. No source gives a cost per post, per comment or per view, for anyone.
And the largest figures are the least documented. China’s estimates run from $1.5 billion to $10 billion a year, and none of them shows its working.
Lakhta’s budgets, according to the complaint, covered “office rent, utility payments, and garbage disposal”, alongside “[a]dvertisement on Facebook”. And when Tenet’s backer, the investor who didn’t exist, once mistimed an email, the indictment says, he searched for “time in Paris” and wrote back: “Sorry, wrong hour. Didn’t sync the calendar.” It was 4:58 p.m. in Moscow.
1. Garbage disposal
“In addition to administrative expenses, such as office rent, utility payments, and garbage disposal, the budget identified IT expenses”
From the budget submitted to Concord in January 2017, which also accounted for 2016's actual spending, as US prosecutors describe it. An allegation.
2. The ad lines
“"[a]dvertisement on Facebook," "[a]dvertisement on VKontakte," "[a]dvertisement on Instagram,"”
The complaint lists these as social media marketing expenses. It gives no amount for each.
3. The total
“a backward-looking accounting of actual expenses for calendar year 2016, which totaled approximately 720 million Russian rubles (approximately $12 million U.S. dollars).”
4. As alleged
A criminal complaint sworn in the Eastern District of Virginia in September 2018 and unsealed on 19 October 2018, case 1:18-MJ-464. These are prosecutors' allegations, not findings of a court.
The receipt is mine; every line on it is the criminal complaint’s wording, an allegation: the items are from a budget submitted in January 2017, which also accounted for 2016’s actual spending. The complaint gives no price for each item, so neither does the receipt. The original paragraph is below.
From the criminal complaint in United States v. Khusyaynova, 1:18-MJ-464 (E.D. Va.), page 10. Boxes mine.
RT’s budget is set to fall. The enacted programme cuts its main line from 32.08 billion roubles in 2026 to 22.86 billion in 2027 and 2028, according to State Media Monitor, the only source I found for the cut.
Three Questions
Think of the last thing that reached you today and made you angry: the clip, the post, the headline, the reply under it. Hold it still for a moment.
Whose name is on it? You trusted it because of the name: a paper you read, a charity you’ve heard of, a commentator you’ve watched for years. In the first post the name was the costume. A masthead copied down to the font. A professor listed at the UN a year after he died. An investor the indictment calls a fictional persona. Do you know who paid for the face? Have you ever checked?
Who had notes on it? The film you watched at the weekend, the series everyone’s talking about. Before it reached you, did someone in uniform read the script? Did a ministry choose the theme from a list? Did the studio guess what a censor would want and cut it first, so that nobody ever had to ask? You will almost never see the note. You only see what survived it.
Why is there so much of it? It’s in your feed, in the group chat, in the replies, in the stream your twelve-year-old has open in the other room. Who gains from the room being this full? Somebody is on a shift. Somebody is filling a quota. Somebody has a budget with your attention on it, filed somewhere between the office rent and the garbage disposal. How much of what you think this week did you choose to look at?
When did you last ask any of them?
RAND’s advice, in 2016, was to get there first. “Forewarning is perhaps more effective than retractions or refutation of propaganda that has already been received,” the paper says, and it suggests putting “raincoats on those at whom the firehose of falsehood is being directed.” Ten years on, the firehose has a payroll. The raincoat is still those three questions, and nobody is paid to hand them out.
There is also a law. Article 20 of the International Covenant on Civil and Political Rights, adopted in 1966, says in its first paragraph:
Any propaganda for war shall be prohibited by law.
The UN Human Rights Committee says states must pass “the necessary legislative measures” and provide “an appropriate sanction in case of violation”. Russia ratified the Covenant in 1973, as the USSR, and the UN treaty register shows no reservation on Article 20. China signed the Covenant in 1998 and has not ratified it. The United States ratified it in 1992 with a reservation saying Article 20 does not authorise or require legislation restricting free speech. The United Kingdom, Denmark and Finland also entered reservations or declarations against Article 20, on the same register.
In 2025 Russia’s culture ministry and the state Cinema Fund backed 12 films “dedicated to the special military operation”, Interfax reported.
Sources and Further Reading
- Christopher Paul and Miriam Matthews, RAND. “The Russian ‘Firehose of Falsehood’ Propaganda Model: Why It Might Work and Options to Counter It.” 2016.
- Gary King, Jennifer Pan and Margaret E. Roberts. “How the Chinese Government Fabricates Social Media Posts for Strategic Distraction, Not Engaged Argument.” American Political Science Review 111(3), 2017.
- Gregory Eady et al. “Exposure to the Russian Internet Research Agency foreign influence campaign on Twitter in the 2016 US election and its relationship to attitudes and voting behavior.” Nature Communications 14:62, 9 Jan 2023.
- Christopher A. Bail et al. “Assessing the Russian Internet Research Agency’s impact on the political attitudes and behaviors of American Twitter users in late 2017.” PNAS 117(1), 2020.
- Trevor Incerti, Jonathan Elkobi and Daniel Mattingly. “Foreign influencer operations: How TikTok shapes American perceptions of China.” arXiv 2601.14118. Preprint, not peer reviewed. 20 Jan 2026.
- Ferenc Huszár et al. “Algorithmic amplification of politics on Twitter.” PNAS 119(1), 2022.
- Germain Gauthier, Roland Hodler, Philine Widmer and Ekaterina Zhuravskaya. “The political effects of X’s feed algorithm.” Nature, 18 Feb 2026.
- Andrew M. Guess et al. “How do social media feed algorithms affect attitudes and behavior in an election campaign?” Science 381, 2023.
- Brendan Nyhan et al. “Like-minded sources on Facebook are prevalent but not polarizing.” Nature 620, 2023.
- Meghann Myers, Army Times. “Most American youth first meet a recruiter at 17, but the Army wants to start younger.” 27 Mar 2019.
- Jim Edwards, CBS News. “‘America’s Army’ Recruitment Video Game Cost Taxpayers $33M.” 11 Dec 2009.
- Hope Hodge Seck, Military.com. “As Military Recruiters Embrace Esports, Marine Corps Says it Won’t Turn War into a Game.” 12 May 2020.
- Jordan Uhl, The Nation. “The US Military Is Using Online Gaming to Recruit Teens.” 15 Jul 2020.
- Harm Venhuizen, Army Times. “Army esports team denies accusations of violating First Amendment, offering fake giveaways.” Jul 2020.
- Matthew Cox and Joanna Campione, Military.com. “Army Unbans Hundreds of Gamers as It Resumes Streaming on Twitch.” 14 Aug 2020.
- US House of Representatives, Office of the Clerk. Roll call 173. 30 Jul 2020.
- Matthew Gault, Vice. “Shitposting Shiba Inu Accounts Chased a Russian Diplomat Offline.” 12 Jul 2022.
- UNITED24. “NAFO fellas.” 30 Dec 2022.
- US Department of Justice. Criminal complaint, United States v. Khusyaynova, 1:18-MJ-464. Sworn Sep 2018, unsealed 19 Oct 2018.
- US Department of Justice. Indictment, United States v. Kalashnikov and Afanasyeva, 24 Cr. 519 (SDNY). Sep 2024.
- The Moscow Times. “More Taxes for More War: Unpacking Russia’s 2026 Budget.” 7 Oct 2025.
- Ivan Khomenko, UNITED24 Media. “Putin Doubles Down on Propaganda With Record Spending in 2026 Budget.” Oct 2025.
- State Media Monitor. “RT.” Updated Sep 2026.
- Kenton Thibaut, Atlantic Council. “Chinese discourse power: capabilities and impact.” 2 Aug 2023.
- David Shambaugh, Foreign Affairs. “China’s Soft-Power Push.” Jul/Aug 2015.
- Freedom House. “Beijing’s Global Megaphone.” 2020.
- US Government Accountability Office. GAO-25-106719, military advertising and recruiting. 20 Nov 2024.
- UN Treaty Collection. International Covenant on Civil and Political Rights, status. Checked 1 Oct 2026.
- International Covenant on Civil and Political Rights, 1966 (University of Minnesota Human Rights Library copy).
- UN Human Rights Committee. General Comment No. 11, 1983 (University of Minnesota Human Rights Library copy).
- Interfax. “Minkultury v 2025 godu vydelilo finansirovanie na proizvodstvo 172 filmov” (The Culture Ministry funded 172 films in 2025). 29 Jan 2026.
- This blog. “Present: Mr. Louis Shon.” and “Remove the Material.” Oct 2026.
- This blog. “Understanding 20 years of Propaganda.” and “Cry Deepfake.”
- US Department of Justice. Indictment, United States v. Internet Research Agency et al., 1:18-cr-32 (D.D.C.). Feb 2018.
.jpg){kind=link}
