Fake accounts are sowing division in Western democracies. The US State Department says (pdf) Russia spends at least $500 million per year on media infrastructure to wage disinformation and counter-narratives campaigns about the European Union and the United States. Security researchers agree that coordinated social media attacks, with Russian backing, have targeted the French and US presidential elections over the last year, while domestic hate groups in the US and Europe are harassing journalists, activists, and others.

Stemming this tide of misinformation and violence means, in part, purging Twitter and Facebook platforms of trolls, bots, and bullies. Tools exist to do this far better than the platforms have done in the past, say researchers such as Emilio Ferrara, a computer science research at University of Southern California.  The companies could deploy algorithms to weed out millions of fake accounts, and hire humans to restore real accounts snared by the bot dragnet (Facebook recently said it would hire 1,000 contractors to better police ads and malicious targeting).

Removing fake accounts might decrease total users, at least temporarily, but the effect would likely be offset by greater trust, giving advertisers and analysts better understanding of exactly who is on social media platforms.

But there’s limited financial incentive for companies to do this. Executives fixated on user growth (or lack thereof) are focused on gaining and retaining new users, without much emphasis on deterring the fake ones. “I can make an argument that for all these companies, the user metric is probably the single most important data point people focus on,” said Scott Kessler, the director of equity research at CFRA in an interview.

That number was crucial in dragging down Twitter’s valuation as new user growth slowed after its 2003 IPO, said Michael Pachter, an analyst at Wedbush Securities. ”The depressed Twitter share price is solely because of lack of user growth,” he said in an interview.

The scale of the bot army

No one knows how many fake accounts exist, but tens of millions on just Twitter and Facebook is likely conservative. Neither company permits outside researchers to verify their figures and the public companies divulge little in their regulatory filings. In 2014, Facebook estimated as much as 11.2% of its monthly active users (MAU) were fake, but claimed in filings (pdf) last year that the number had fallen to just 1% of its 2.01 billion MAU. It announced some public efforts to fix the problem—Facebook eliminated 30,000 fake accounts while imposing measures to “reduce the financial incentives for spammers” just before the French election this spring – although the nature of the accounts and measures was not disclosed.

Twitter’s problem is messier. After years of anemic growth, the company has yet to shed its reputation as a haven for fake accounts, despite claims that it’s stepping up its efforts. In September, Twitter wrote its automated systems were catching more than 3.2 million suspicious accounts per week, more than double last year’s pace. Overall, it estimated false or spam accounts represented less than 5% of its MAU.

But Ferrara at the University of Southern California says that number drastically underestimates the problem. ”It’s probably not even the smallest tip of the iceberg,” he said in an interview. Ferrara and his colleagues analyzed 14 million accounts (pdf) via Twitter’s API estimating that between 9% and 15% of all Twitter accounts are fake. Even that number is low, they warn, because sophisticated AI-driven accounts can evade detection.

Twitter counters that third-party studies “systematically under-represent our enforcement actions” because the studies don’t detect the company’s actions taken via the API or through product features that suppress low-quality content. In particular, it is turning to machine learning to shut down spammers and bots before they set up accounts, and eliminate them if they do (Twitter’s latest transparency report claims it suspended about 300,000 accounts linked to terrorism in the first half of this year; 95% were surfaced by automated machine learning tools  compared to 74% six months prior).

Yet even the absolute number of bots obscure the full scope of the problem. “Bots are much more active than humans,” says Ferrara. “That’s a much more important statistic that Twitter does not release.”  Automated accounts can spew out hundreds or thousands of tweets, shares, and posts per day exaggerating their influence far beyond what numbers alone would suggest. Ferrara estimates during the most recent US presidential election accounts bots generated around 20% of Twitter activity on the topic.

How to fix it

It’s possible we’re on the brink of seeing the social media platform crack down on millions of fake accounts. That may not hurt Twitter or Facebook in the long run since some of the damage from fake accounts is baked into the stock price today. “Everyone knows some portion of their followers are fake,” said Pachter. So long as Twitter and Facebook demonstrate user growth over time, and earn more revenue per user by increasing engagement, then their stock prices should emerge unscathed. ”Nobody would care,” he said.

Ironically, it may even increase the value of the stock. If growth on the platform is primarily from real people, not bots, that will reassure advertisers they can reach an actual audience. “One reason advertisers are skeptical of Twitter is they’re not sure who is advertising to,” said Pachter. Ridding the platform of bots “would be better off for everyone.”

At the moment, there is limited transparency and trust in the numbers the companies disclose, especially at Twitter. Facebook and Snapchat reveal their daily active users (DAU) to investors each quarter, but Twitter has not done so, providing only percent growth of DAU. Analysts are also skeptical that Twitter has excluded all their fake accounts from user estimates, despite its insistence on this in regulatory filings (pdf) . “I’m not sure the company has communicated with the analyst community how they account for them,” Kessler at CFRA said. Pachter agreed. “Ask Twitter a question [during investor calls]…and you get this ridiculous evasion,” he said. Twitter says it’s now embracing greater transparency in its recent disclosures.

Getting rid of fake accounts won’t solve everything. Facebook’s decision to focus on surfacing content that maximizes engagement means it’s likely to serve up incendiary and false information over more pedestrian (and factual) fare from mainstream publications. In the final months of the 2016 campaign, Buzzfeed found fake election stories outperformed factual coverage. Ridding the producers of content is one thing. Managing the content is another.

But the proliferation of misinformation is putting the the world’s social media platforms on defense. Some in the US now see them as contributing to “a national security threat.” European regulators have labeled them publishers with “more responsibility to ensure the right content.” Executives from Facebook, Google, and Twitter are being summoned to testify publicly before the Senate.

Until now, Silicon Valley companies have sought to present themselves as tools for people “to share and make the world more open and connected,” as Facebook once defined its mission. In this quest to remain “neutral,” and perk up figures that please Wall Street, they may only set themselves up for a bigger fall.


Correction: A previous version of this post misstated the spelling of Emilio Ferrara.

📬 Sign up for the Daily Brief

Our free, fast, and fun briefing on the global economy, delivered every weekday morning.