By Leonid Bershidsky
JUDGING from reported growth in user numbers, one could conclude that Facebook is beating Twitter in the battle of social networks. Problem is, it’s hard to know how real all those users are.
Twitter reported this week that it had 328 million monthly active users at the end of June, the same as at the end of March. Facebook reported 2.01 billion monthly active users, up from 1.94 billion in March. Another measure that the companies now like to stress—the average number of users per day—also showed Facebook ahead. Twitter said its daily active users were up 12 percent in the second quarter from a year earlier, without naming an absolute number (Re/Code calculates it’s some 157 million). Facebook said it was up 17 percent to 1.32 billion.
On the face of it, Twitter—the purveyor of news about what’s going on in the world right now—appears to have lost out to Facebook’s concept of community-building. Forcing people to be concise and letting them see posts in chronological order proved less attractive than Facebook’s embrace of all kinds of content and insistence on using a non-transparent algorithm to form the news feed.
This could prompt me—as someone who has quit Facebook and stuck with Twitter—to complain that people don’t want control over their information consumption, to mock them for being apathetic in the face of the intrusively irrelevant self-launching videos that have made Facebook such an uncomfortable place for me, to ask how they tolerate the proliferation of ads, mindless clickbait and fakes.
But I’m not going to do that. Why? Because I don’t trust either company’s numbers. And unlike in the old days of newspapers, when circulations were audited, there’s no way for an outsider to verify them. To understand my skepticism, consider a new report from the Atlantic Council’s Digital Forensic Research Lab on the Twitter commentary surrounding a New York Times column headlined “Trump Is His Own Worst Enemy.” It features news bots, pro-Trump bots, anti-Trump bots, a commercial botnet and an online service that allowed users to automate their posts with an “if…then” routine—for example, retweeting posts that contain a certain word or phrase. If any real people were involved in the discussion, they appear to have been heavily out-tweeted.
A lot of the discussion on Twitter is exactly as described in the Digital Forensic Lab’s report. It’s close to impossible to determine with any accuracy how much traffic comes from fake accounts. One recent study, using a behavior-based model, estimated that nine percent to 15 percent of active accounts were bots (a lot more than Twitter has reported). That’s still a big range: Based on the company’s latest estimate of monthly active users, it adds up to anywhere between 29.5 million and 49.2 million bots.
I assume good faith on Twitter’s part; it just can’t know what’s really going on. Botnet owners are smart. They’re constantly perfecting their techniques for imitating human behavior, staying a step ahead of efforts to detect them. So if, say, Twitter is seeing more growth in average daily users than in end-of month users, it could simply mean that the bots are being used more intensively and efficiently. The trend doesn’t necessarily reflect any change in human usage.
The same is true of Facebook. For all the academic interest in catching fake activity, for all the company’s own efforts to remove fake accounts, comprehensive policing is extremely difficult. “Our results indicate that many fake users are classified as real suggesting clearly that fake accounts are mimicking real user behavior to evade detection mechanisms,” Aditi Gupta and Rishabh Kaushal wrote in a recent report. One medium-sized botnet would be enough to account for all the 70,000 monthly active users that the company added in the last quarter.
Nonetheless, advertisers and investors remain obsessed with user data. Facebook’s shares jumped and Twitter’s tanked after the companies announced their numbers. Billions of dollars in market value were gained and lost.
It’s time to establish mechanisms for verifying social networks’ user data. As with TV ratings or circulation audit systems, external researchers should be able to check the companies’ claims and have a crack at working out the number of fake and duplicate accounts. They may never be perfect, but with a consistent and transparent methodology they could at least be comparable. Although investors and advertisers could be in for some surprises, they’d get a better sense of what’s really going on.