The Same Twitter Bots That Helped Trump Tried to Sink Macron, Researcher Says

Bots are a reality of political discourse. Research suggests the fortunes of Donald Trump in the lead-up to the 2016 election were buoyed in part by Twitter bots promoting him, and the "MacronLeaks" documents aimed at dashing centrist French President Emmanuel Macron's 2017 election chances were also spread by bots.

New research from Emilio Ferrara, the University of Southern California academic who exposed the role of bots in the 2016 US election, shows that many Trump bots went dark and later turned into MacronLeaks bots. This, Ferrara wrote in a new paper posted to the arXiv preprint server this week (which is currently being peer-reviewed), suggests that there may be a "black market" for right-wing political bots that can lay dormant for months before being activated to promote the next conservative demagogue.

"There are way too many coincidences here to keep us from thinking that there are venues where organizations with enough resources can access these botnets," Ferrara said over the phone.

Read More: Election 2016 Belongs to the Twitter Bots

The French election saw Macron, an elite of the financial class, face off against the far-right and xenophobic Marine Le Pen. The day before the election, a cache of documents including Macron staffers' emails emerged from 4chan to discredit the candidate. The documents were misinterpreted with the help of right-wing personalities and then boosted and spread around the internet by automated bot accounts. (Although, really, they didn't make much of a dent—Macron won.)

Ferrara collected 17 million tweets from roughly two weeks leading up to the French election and designed a custom machine learning algorithm (based on the Botometer, a public tool that looks for the defining marks of a robot controlling a given Twitter account), to parse the massive trove and pick out bot accounts. Of the nearly 100,000 users in the sample who participated in the MacronLeaks discussion on Twitter, 18,000 were bots, Ferrara said. According to the paper, some of the accounts that targeted Macron were actually created in the lead-up to the 2016 US election.

"These accounts were tweeting their support for Trump for about a week in the run-up to the 2016 election and then they went dark for a very long time," Ferrara said. "These same accounts picked up again and some even started tweeting in French—but the alt-right narrative was the same."

It's worth noting that some formerly pro-Trump Twitter bots switched to pushing Obama merchandise after the election, which suggests that at least some of the bots promoting Trump were for hire.

When reached for comment, a Twitter spokesperson directed Motherboard to the company's policy on bot accounts. Of a set of 15 bot accounts that Ferrara verified by hand, instead of with the algorithm alone, 13 had already been suspended by Twitter or deleted.

According to Ben Nimmo, an Atlantic Council senior fellow who spotted bots spreading the MacronLeaks message on election night, Ferrara's work confirms much of what he suspected when he initially analyzed the tweets. However, it's not clear, Nimmo said over the phone, if the bots are really part of a "black market" that is mercenary and commercial in nature, or if it's just one ideologically motivated person or group siccing their bots. But Ferrara's research shows they're connected.

"To see a set of bots which had been pro-Trump in November targeting Macron in May could indicate a black market and that the use of those bots has been hired out to political actors," Nimmo said. "It could also be that the same actor who supported Trump in November decided to start supporting Le Pen."

According to both Nimmo and Ferrara, more work is needed to investigate the true scope and scale of political bots supporting right-wing candidates around the globe. To that end, the algorithm developed by Ferrara to analyze millions of tweets and pick out a few bots could be a huge help to researchers.

"It's shown that you can do a credible and accurate analysis on a very large amount of traffic," Nimmo said. "That's a technique worth using many times over, because there are lots of possible botnets out there."

Motherboard staff is exploring the cultural, political, and social influence of the iPhone for the 10th anniversary of its release. Follow along .