The Washington PostDemocracy Dies in Darkness

The unseen machine pushing Trump’s social media megaphone into overdrive

Researchers say the online feedback loop between Trump, high-profile influencers and rank-and-file followers is more dangerous than Russian disinformation.

(Glenn Harvey for The Washington Post)

This story was featured in Drop Me The Link, our one-story election newsletter. Sign up to get it in your inbox three times a week.

President Trump launched into a tweetstorm in April, banging out nine retweets of the Centers for Disease Control’s account on the dangers of misusing disinfectant and other topics — two days after he himself had suggested that people could inject themselves with bleach to cure covid-19.

But those tweets spread in an odd pattern: More than half the 3,000 accounts retweeting Trump did so in near-perfect synchronicity, so that the 945th tweet was the same number of seconds apart as the 946th, University of Colorado information science professor Leysia Palen found.

The unusual finding underscores some of the little-known ways in which Trump’s social media army — composed of devoted followers and likely assistance from software that artificially boosts his content — has helped him develop one of the world’s most powerful political megaphones, unlike any other in the English-speaking world.

That megaphone has become a frequent source of misinformation, some of it so toxic that Harvard researchers recently dubbed attacks on mail-in voting by Trump and right-leaning leaders “a highly effective disinformation campaign with potentially profound effects ... for the legitimacy of the 2020 election.”

Trump’s singular ability to spread his messages, often disseminating false or unsubstantiated information, comes from his prominence as president and the relentless clip of his tweeting to his 87 million followers. He is also aided by a vital feedback loop — often discussed but poorly understood — among the president, high-profile influencers and rank-and-file followers that both push messages in his direction and promote every online utterance.

His feedback loop, according to several new and forthcoming studies, has become a leading threat to the integrity of political debate in the United States, with an impact that to date appears far more damaging than the efforts of Russian operatives or other foreign adversaries.

A study released Thursday by the Election Integrity Partnership, a consortium of misinformation researchers, found that just 20 conservative, pro-Trump Twitter accounts — including the president’s own @realDonaldTrump — were the original source of one-fifth of retweets pushing misleading narratives about voting.

A recent Cornell University study, meanwhile, concluded that Trump was also the “largest driver” of misinformation in the public conversation about the coronavirus during the first half of 2020. The researchers found that nearly 40 percent of articles containing misinformation about the virus mentioned him, including articles about false cures and blaming China for the disease.

Since the start of the coronavirus outbreak, President Trump has repeatedly said that the virus will disappear. (Video: The Washington Post, Photo: Jabin Botsford/The Washington Post)

“Trump is hands down the most significant accelerant and amplifier for disinformation in the election,” said Graham Brookie, director and managing editor of the Atlantic Council’s Digital Forensic Research Lab, a branch of the prominent nonpartisan think tank that is a leading source of research on foreign and domestic disinformation. “The scale and scope of domestic disinformation is far greater than anything a foreign adversary could ever do to us."

Study shows Trump is a super-spreader — of coronavirus misinformation

The president, his reelection campaign and congressional Republicans have repeatedly said that Trump’s skill in deploying social media is key to delivering his message in the face of what they say is hostility from mainstream news sources and leading Silicon Valley companies. Trump and his defenders also contend that his comment about the potential medical value of ingesting bleach was intended as sarcasm, not a suggestion, and that the extensive news coverage of that and other elements of the White House’s handling of the coronavirus pandemic are signs of systemic bias that a potent Twitter following helps him overcome.

Trump campaign spokeswoman Samantha Zager said in a statement, “Neither Big Tech nor the mainstream media is the arbiter of truth or elections. ... President Trump is a staunch advocate for a free and fair election where every vote counts exactly once.”

Her statement did not directly address allegations that he and his social media army have become leading sources of disinformation in the election. Nor did it address claims by some researchers that he is a bigger source of disinformation than Russia or to evidence that he gets a boost from automated accounts.

While researchers believe there is a significant degree of automation powering Trump’s megaphone, they do not contend that he or his supporters routinely break Twitter’s rules. The company allows enthusiastic followers to use software to automatically retweet people, and though the company bans “bulk retweeting,” it does not disclose at what point frequent retweeting crosses over into prohibited behavior. Twitter suspended 100 out of 200 sample accounts brought to its attention by The Washington Post for spammy activity or other violations.

Tech companies have spent the past four years preparing to fight foreign disinformation campaigns, but now they are confronted with a situation where their political caution and deference to free speech, coupled with the power of algorithms and loosely enforced rules, have turned their platforms into hubs of domestic misinformation — with the misleading comments often led by the president himself.

Twitter, for example, did not have a policy banning any form of misinformation until 2020, and until just months ago, both Twitter and Facebook exempted politicians — including Trump — from their rules on the grounds that their comments were too newsworthy to censor.

Twitter labels Trump’s tweets with a fact check for the first time

Of the more than 22,000 falsehoods Trump has shared since the start of his presidency through this month, according to The Washington Post Fact Checker, more than 3,700 of them have been on social media. The number of lies is six times greater so far this year than during his first year in office, the Fact Checker has found. Since May, Twitter has put warning labels and restricted viewership on at least 15 of those comments, while Facebook has removed half a dozen.

“Protecting the integrity of the conversation on Twitter by stopping platform manipulation and addressing misinformation remains one of our top priorities,” Twitter spokeswoman Liz Kelley said in a statement, adding that since the 2016 election the platform has added protections to reduce manipulation and misleading information.

Facebook declined to comment.

But some of those actions may be too late to make a difference.

Examining the impact of Twitter’s enforcement against an Aug. 23 Trump tweet in which he called mail-in ballots a “security disaster,” Kate Starbird, a disinformation expert at the University of Washington, found that Twitter’s disabling of the retweet button after labeling the tweet effectively stopped the content from spreading. But the tweet had already gone viral.

“Twitter has grown quicker in taking action on President Trump’s tweets” since then, said Starbird. “But due to the size of Trump’s following — and perhaps other factors like coordinated tweeting, automation and an unusually attentive audience — his tweets can still spread quite far before the platform takes a corrective action.”

Trump’s social media megaphone

Trump had just 20 million Twitter followers on Inauguration Day in January 2017. Today, he averages more than 1,000 tweets a month, with nearly 17,000 retweets each, an unparalleled volume in the English-speaking world, according to researchers. That means Trump benefits from a powerful ecosystem that amplifies every post, and from Twitter’s own loose rules and algorithms that give an additional boost to messages based on the engagement they receive.

When Palen and her team examined Trump’s tweetstorm on the CDC, they discovered that 56 percent of each of the 3,000 retweets were shared in the same pattern over a period of 24 hours.

For example, an account called @tthseeker always retweeted Trump 42 minutes after he tweeted. Another account, @shauna33R, always tweeted 4½ seconds after that.

Part of the pattern involves a core group of more than 500 especially enthusiastic followers — with 832,000 followers among them — that retweeted Trump in a roughly identical order and length of time after Trump tweets from April to September, regardless of what the president tweeted.

Palen said the pattern was “remarkable” and could not have happened by accident. She said it strongly suggests there is some combination of automated accounts and human retweeting at work.

Then, comparing all Trump tweets in August to tweets from other high-profile accounts, such as New York Gov. Andrew M. Cuomo, British Prime Minister Boris Johnson and Canadian Prime Minister Justin Trudeau, as well as celebrity Kim Kardashian, Palen found that retweet patterns for the other accounts have about half the degree of similarity as Trump’s.

Twitter adds new warnings about misinformation in run up to election

The set of accounts identified by Palen swells with support for Trump and his political slogans, using hashtags that reference the idea that they are a social media army, such as #digitalsoldier, #maga and #fightback, and patriotic images such as flags and eagles. More than 60 reference the QAnon conspiracy theory. Many of the accounts, in their profiles, describe the owners as conservatives, devout Christians, veterans and gun rights enthusiasts. Retweets of Trump, and of one another, are far more common than original tweets for many accounts.

The Post attempted to contact more than 100 of the users through tweets and direct messages. Those who replied expressed enthusiasm for the president and an eagerness to amplify him at a time when, they contended, his voice was getting muddied by mainstream news sources and technology companies.

One person using the name PatriotMountainMan tweeted in reply to a Post query, “I retweet President Trump’s entire Twitter profile every day to BYPASS THE CORRUPT AND BIASED LIBERAL MEDIA who have lost their way and have failed to report the truth to the American people.”

Palen has also found that more than 100 of the original accounts that retweeted Trump in April have been suspended by Twitter. The company targeted the accounts for “bulk retweeting” and other violations. In addition, Twitter said it had already reduced the reach of 80 percent of the 200 accounts shared by The Post, meaning the accounts had less value in Twitter’s dissemination algorithms than those that behaved more authentically.

Palen, whose research is ongoing but has been reviewed by two outside experts on behalf of The Post, said she believes a substantial portion of retweets of Trump tweets are indeed authentic expressions of support from Americans who share his views. But she has ultimately concluded that programmed software, or automation, is also propelling Trump to go viral to a degree that has not previously been understood.

“Trump’s amplification machine is peerless,” Palen said.

A substantial portion of those retweeting Trump are authentic expressions of support from Americans who share his views, says Clemson University social media researcher Darren Linvill, who reviewed Palen’s work at The Post’s request.

“There’s a lot of passion for Donald Trump on social media, especially on Twitter. And a lot of that passion translates into blind retweeting that may appear automated but is not,” said Linvill, lead researcher for Clemson’s Media Forensics Hub.

Though Twitter bans bots, or fully automated accounts, the company allows a degree of automation in its service, by permitting people to use third-party software programs that automatically retweet or reply to tweets. In addition, Twitter’s algorithms are designed to reward engagement — or tweets that get retweeted or liked very quickly — which results in the company widely sharing that information beyond the person’s actual followers and in its trending feature. Twitter’s algorithms give lower weight to low-quality or spammy accounts.

Senate Republicans vote to authorize subpoena targeting Facebook, Twitter CEOs, citing their handling of New York Post story

Twitter’s Kelley said the company was still investigating the retweeting pattern. Of the 200 accounts identified by The Post, Kelley said they did not appear to be using third-party software but were likely real people engaging in behaviors and tactics that the company frowns upon.

“We often see people using Twitter in a way that can appear and sometimes is spammy, but many times is just how some people use the service,” Kelley said in an email. “Bulk and aggressive Retweeting is a violation of the Twitter Rules and may result in action, but that does not mean the behavior is inauthentic.”

She said the company is considering adding a warning notice to users when they are retweeting too much and about to get sanctioned.

The University of Washington’s Starbird found that on topics such as the coronavirus and the election, between 7 and 10 percent of the accounts that retweeted Trump were affiliated with QAnon, an online conspiratorial movement that Facebook and Twitter recently curtailed, suggesting that his support on these subjects came from large numbers of followers who have violated Twitter’s rules. Trump has in turn retweeted QAnon-affiliated accounts more than 250 times since taking office, according to the left-leaning watchdog group Media Matters.

Fighting manipulation is a “cat-and-mouse game, and the mouse has discovered a new trick,” said a person familiar with Twitter’s system who spoke on the condition of anonymity to comment on the new research.

How information gets to Trump

Researchers are also examining how disinformation makes its way to the president through campaigns that push distorted narratives in his direction.

At campaign rallies and during the presidential debate over the past month, for example, Trump has talked about how ballots with his name on them had been found “thrown in a river” — implying that politically motivated postal workers were sabotaging his chances of winning the election.

The tossed ballots narrative originated Sept. 23 with an article on the conservative website Gateway Pundit. It quoted a news report about a batch of mail being found in a ditch in Greenville, Wis., and claimed, without citing evidence, that the incident was part of a Democratic plot to steal the 2020 election because the lost mail included some absentee ballots. The Gateway Pundit story was tweeted one hour and one minute later by the president’s son Eric Trump, who, along with his brother Donald Trump Jr., are the most frequent tweeters about mail-in ballots among the 50 users Trump follows, according to researchers with the University of Washington.

The story was then tweeted by conservative action group Turning Point USA founder Charlie Kirk, Breitbart News, and the TrumpWarRoom, researchers with the Election Integrity Partnership, a consortium including Stanford University, University of Washington, the DFRLab, Graphika, and other leading disinformation researchers, found. Within 24 hours, the Gateway Pundit story had 60,000 retweets. Trump then mentioned it during the Sept. 29 presidential debate, though he changed the ditch to a river — an error the White House later conceded.

The Wisconsin Elections Commission later announced that no Wisconsin ballots had actually been found among the tossed mail. The matter is still under investigation by the U.S. Postal Service.

Gateway Pundit founder Jim Hoft told The Post that the article “was accurate.”

Trump’s re-shared tweets help shield him from Twitter’s bans

“The disinformation spread by Donald Trump doesn’t typically start with him,” said the Atlantic Council’s Brookie, who previously worked at the National Security Council during the Obama Administration as a civil servant. “It is the endpoint in a supply chain of disinformation — the tip of an iceberg with an amplification ecosystem of right-wing media, influencers and outright conspiracy theorists making up the bulk below.”

And the distorted narratives have made an impact even on the left. Sharon Broussard, an Oakland resident, said she decided to vote in person on the first day of early voting on Oct. 5 because she no longer trusted the Postal Service.

“Your ballot might be tossed in the trash or in the side of the road,” she warned in a recent Facebook post in a group for African Americans who support Democratic vice-presidential nominee Kamala D. Harris. “Please no distractions or voter suppression!”

Nate Persily, co-director of the Cyber Policy Center at Stanford University, said Trump’s feedback loop also involves users who aggressively reply to him to seek his attention and the president’s rewarding of followers through retweets. Both behaviors egg him on.

Florida yoga instructor William Painter said that before Twitter suspended his account during a sweep of QAnon this summer, he was on a campaign to encourage Trump to continue to promote the malaria drug hydroxychloroquine as a cure for covid-19. His Enki74 account was Trump’s second-most-frequent replier on the topic, messaging Trump 68 times over a 45 day period in the spring, said Avery Wendell, a misinformation researcher with Harvard’s Shorenstein Center.

“Trump is attempting to listen to the will of people,” Painter said.

Another supporter, Logan Cook, who goes by the username Carpedonktun, said he has been able to get Trump’s attention with his self-designed memes, which include a recent video Trump retweeted that was edited to make it look as if an Black toddler were running away from a White child (in reality, the two children hugged one another in the viral video).

Fake Twitter accounts posing as Black Trump supporters appear, reach thousands, then vanish

Cook said he didn’t know how exactly his memes, which he said had been retweeted by Trump 20 to 30 times before Twitter suspended his account in June for promoting “manipulated media” and copyright infringement in relation to the toddler video, made their way to the president. Trump hosted him twice at the White House last year, including a visit to the Oval Office. Cook described his work as political satire. “Memes are not misinformation,” he said.

Path to Trump’s megaphone

For most of its history, Twitter gave broad latitude to free speech. That changed this spring as it became the first social media company to sanction Trump, restricting access to and fact-checking a misleading tweet about mail-in ballots in late May.

Before this year, the company had no policies against misinformation, and the flagging of Trump’s tweet was also one of the first times the company had penalized anyone for false or misleading speech.

Executives thought the company should not be an arbiter of truth in domestic affairs, according to people familiar with Twitter’s efforts who spoke on the condition of anonymity to describe internal conversations, in part because they were aware that any action they took would be viewed as tilting the scales.

But that idea was challenged under the Trump presidency. The company faced criticism from politicians and employees that the company’s service was becoming a vector for abuse, misinformation and targeted harassment.

With Election Day looming, Twitter imposes new limits on U.S. politicians — and ordinary users, too

Trump’s account was protected by the company’s “public interest” clause that exempted politicians from its rules. In 2018, Twitter CEO Jack Dorsey told The Post that the company was considering whether powerful people like politicians should be better policed.

Twitter announced it would begin to label tweets from powerful people who broke its rules the following year, effectively ending the public interest exemption. The announcement followed a year’s worth of new policies promoting “healthy conversations” and a new focus on purging or restricting problematic accounts. Then, this year, the company banned misinformation about the coronavirus and the election, as well as manipulated video.

When Trump tweeted about mail-in ballots in May, the company had the policies in place to take action, but internally it took an act of courage to go against Trump, the people said.

“Much of the influence Trump has gained to shape the national conversation will far outlast his presidency,” Clemson’s Linvill added.