Disinformation, 'fake news'and influence campaigns on Twitter

4 Oct 2018

Using tools and mapping methods from Graphika, a social media intelligence firm, we study more than 10 million tweets from 700,000 Twitter accounts that linked to more than 600 fake and conspiracy news outlets. Crucially, we study fake and conspiracy news both before and after the election, allowing us to measure how the fake news ecosystem has evolved since November 2016.

Consistent with other research, we find more than 6.6 million tweets linking to fake and conspiracy news publishers in the month before the 2016 election. Yet disinformation continues to be a substantial problem postelection, with 4.0 million tweets linking to fake and conspiracy news publishers found in a 30-day period from mid-March to mid-April 2017. Contrary to claims that fake news is a game of “whack-a-mole,” more than 80 percent of the disinformation accounts in our election maps are still active as this report goes to press. These accounts continue to publish more than a million tweets in a typical day.

Sixty-five percent of fake and conspiracy news links during the election period went to just the 10 largest sites, a statistic unchanged six months later. The top 50 fake news sites received 89 percent of links during the election and (coincidentally) 89 percent in the 30-day period five months later. Critically—and contrary to some previous reports—these top fake and conspiracy news outlets on Twitter are largely stable. Nine of the top 10 fake news sites during the month before the election were still in or near the top 10 six months later.

Our study finds much more fake news activity than several recent studies, largely because it examines a larger corpus of fake and conspiracy news sites. Fake and conspiracy news sites received about 13 percent as many Twitter links as a comparison set of national news outlets did, and 37 percent as many as a set of regional newspapers.

Machine learning models estimate that 33 percent of the 100 most-followed accounts in our postelection map—and 63 percent of a random sample of all accounts— are “bots,” or automated accounts. Because roughly 15 percent of accounts in the postelection map have since been suspended, the true proportion of automated accounts may have exceeded 70 percent.

In both the election-eve and postelection maps, our methods identify an ultra-dense core of heavily followed accounts that repeatedly link to fake or conspiracy news sites. Sites in the core are typically not the highest-volume tweeters of fake news. However, the popularity of these accounts, and heavy co-followership among top accounts, means that fake news stories that reach the core (or start there) are likely to spread widely. The pre-election fake news network is one of the densest Graphika has ever analyzed, necessitating unusual map drawing procedures.

While a large majority of fake news came from supposedly pro-Republican and pro-Donald Trump accounts in the month before the election, smaller but still substantial amounts of fake news were passed on by liberal or Democratic-identified accounts. After the election period, though, left-leaning fake news decreased much more than right-leaning fake news.

Publication Details
License Type: 
Published year only: 
Subject Areas
Geographic Coverage