You don’t need to read the federal indictments to spot the moment Russia began targeting the United States with its army of internet trolls.
Just chart the American flag emoji. Best estimates trace the founding of the Internet Research Agency to August 2013, and for eight months of its existence the Saint Petersburg troll farm was focused on influencing citizens in Russia and Russia’s near abroad. Then in April 2014 the Kremlin-linked organization launched “Project Lakhta,” according to federal prosecutors, aimed squarely at the United States.
The IRA’s first American flag emoji appeared on Twitter the next month, first a trickle, and then a torrent. Over the next four years the Saint Petersburg troll farm unfurled 70,372 American flags, reaching peak flagness on August 12, 2017 when one IRA sockpuppet crammed 43 flags into a single tweet with the hashtag “MAGA.”
Whether by luck or design, the IRA’s campaign against America worked, and is still working. An examination of Twitter’s new dump of Russian troll data this month shows that the IRA’s tactics worked far better in the U.S. than in Russia or the Eastern European nations where the troll farm cut its teeth. English-language tweets by the IRA’s sockpuppet accounts enjoyed nine times the engagement than tweets in Russian and other languages. And, remarkably, Americans fell for the Russian interference even harder after the 2016 presidential election than before.
“Poland, Estonia, Ukraine, Georgia, they know Russia’s coming after them,” said Ryan Fox, a former NSA official now serving as COO of the smear-fighting startup New Knowledge. “They’ve been culturally trained to recognize Russia as a threat, so the skepticism of people in that region is pretty high... The U.S. was a far better environment for it to succeed.”
This month Twitter released details of over nine million tweets associated with 3,841 different accounts that Twitter’s investigators linked to the IRA following Russia’s successful election interference operation. Broken down, Twitter’s dataset includes 5,708,020 original tweets and 3,333,168 retweets by the Russian operators, along with the urls included in each tweet.
The data show that over a period of four years, the IRA earned an average of 1.73 retweets, likes or replies for each Tweet written in Russian, or any language other than English. English language tweets, in contrast, received an average of 15.25 engagements over the same period.
The U.S. oriented tweets were crafted to appeal to all sides of America’s social and political divisions. Some accounts posed as Americans on the far right, others as left-leaning activists or members of the Black Lives Matter movement.
The alt-right accounts dominated, but in the wake of Trump’s surprise election victory the Russian accounts posing as African Americans enjoyed a surge in popularity that brought the overall engagement level even higher, from 7.56 engagements per tweet before November 8, 2016, up to 72.55 afterwards, despite the fact that the broad strokes of Russia’s U.S. meddling were well known.
“The conversation after the election was more about fake news, and not the content that was a form of weaponized truth, that picked at narratives that people want to believe and that have an element of truth to them,” said Fox.
To help drive that “weaponized truth,” the IRA often creates personal blogs, organization websites, and news sites with varying degrees of fakeness where the trolls can exercise greater control of their messaging. To get a sense of how successful the IRA was in driving traffic to their own properties, The Daily Beast examined the IRA-originated tweets that include links—some 2,045,698 different urls promoted in 2,689,679 tweets.
More than half of those urls were for link shorteners like Tinyurl or Bitly, and we were able to expand most, 1,181,593 of them, to determine the final destination.
The IRA’s own content is clustered at the top of the list. The website Livejournal, a blog platform popular in Russia, ranks number one by tweet volume—an unsurprising result. The IRA makes heavy use of fake Russian and Ukrainian bloggers in its Eastern Europe campaign. We counted 467,676 tweets directed there.
Below that is the main website for the IRA’s ersatz news agency FAN, with 467,676 tweets, followed by ReportSecret, an IRA English-language site. It was pushed by 138 different Russian accounts with headlines like, “CNN is Forced to Cover Trump’s Viral Video After Ignoring it For DAYS”, “BREAKING: The ‘Russian Lawyer’ Just DEBUNKED the Media’s HIT JOB on Don Jr.” and “Trump-Hater Alyssa Milano Got her ASS KICKED on Twitter and it was GLORIOUS!”
The ReportSecret tweets were part of the post-election push toward automation that’s also clear in the new data. When the Russian troll farm was forced to start replacing thousands of fake IRA accounts shut down by Twitter, it began augmenting its human sockpuppets with crude robo-tweeters programmed to tweet any new content that appeared on a particular website. Even genuine news outlets like CBS, the Washington Post, and the UK-based tabloid the Daily Mail were harvested for this effort.
The ReportSecret site itself followed this new strategy, with its content pulled automatically from the U.S.-based TruthFeed, a frequent purveyor of fake news operated by Trump supporters.
The IRA may have hoped to build a following for its robo-trolls without drawing attention from Twitter’s investigators, and then take the reins at a time of its choosing. But the Twitter data suggest the experiment in automation was a failure. The automated tweets enjoyed a dismal engagement level compared to the human-curated personas, suggesting that few Americans read or shared them. The 123,024 ReportSecret tweets, for example, garnered only 37,128 retweets from genuine users.
The IRA’s human-operated accounts performed much better. And, once again Americans proved a far more receptive audience than Eastern European targets. The top three performing websites, based on engagement, were all created as part of the IRA’s U.S. campaign.
The IRA’s Jenna Abrams persona, its most successful, was frequently quoted in the legitimate press and boasted 70,000 Twitter followers, despite arguing for a return to racial segregation and floating tweets like, “To those people, who hate the Confederate flag. Did you know that the flag and the war wasn’t about slavery, it was all about money.”
The account only linked to the IRA’s fake blog JennAbrams[.]com in 37 tweets, but those tweets drew an average of 30 organic retweets each, for a total of 1,126. The IRA’s civil rights sites blacktolive[.]org and blackmatters[.]us had a similarly generous retweet rate of 20-to-1 and 17-to-1 respectively.
A September 21, 2016 tweet promoting a story on BlackToLive enjoyed 400 non-IRA retweets, and 89 users clicked on the shortened link that took them to the IRA’s fake website, according to data extracted from Bitly’s API. That suggests Twitter performed well for driving traffic to the IRA’s own websites.
In the end the biggest contributor to the troll’s American success story was the divisiveness already present here, said Fox. “This content is still circulating because it resonates so well with people who are angry on both sides... It exploited an environment that we were and are living in.”
Even the IRA’s decision to slather its content with American flags might have been a deliberate ploy. A series of experiments in 2008 demonstrated that being in the presence of the flag doesn’t inspire patriotic love of country, as one would expect, but instead reliably summons feelings of nationalism, defined by the scientists as “an ideology of superiority of the ingroup over outgroups and implies the exclusion or even domination of others.” In its secret campaign to sow division in the U.S., Russia may well have tried to weaponize Old Glory.