Twitter, Land of the Bots

A year ago, Barracuda Labs analyzed over 70,000 fake Twitter accounts and found that fake followers are cheap (only $18 per 1000) and plentiful (49k per average buyer). A more recent study by Andrea Stroppa and Carlo De Micheli examined the eight most popular seller services. They estimate there are 20 million fake followers in total, expertly designed by software: “It fills in every detail. Some fake accounts look even better than real accounts do.” That’s all the easier because Twitter allows multiple accounts per user and many legitimate users barely tweet, so simple identity or activity checks won’t work. In a follow-up report, Stroppa and De Micheli looked at popular Twitter accounts of various companies and politicians – and found numerous daily fluctuations by tens or hundreds of thousands (!) of followers, suggesting that a fake follower rental had begun or ended.

Indeed, real people are something of a rarity among the followers of popular Twitter accounts. Socialbaker estimates that about half are fake or inactive, while Status People puts the number closer to three quarters. Kevin Ashton, who created his own amusing fake account called “Santiago Swallow,” found similar numbers among greater and lesser celebrities. “On average, only 28% of people following the 20 most popular Twitter accounts are real. The remaining users are either fake or dormant.” This isn’t necessarily the fault of the account owners, by the way. Bots will follow random popular accounts so as to present a more natural-looking profile and avoid detection. You can use Status People’s Fake Follower Check to test your own account. (I was relieved to find that I have 85% good followers!)

This year’s Barracuda Labs study shows no improvement. The average price has dropped to $11 per 1000 fakes, the average buyer’s purchase has risen to 55k, and fake accounts are even harder to spot as they are now based on slightly altered real profiles. Christopher Null suggests that Twitter should hide the number of followers to destroy the fake follower industry. He knows Twitter will never do this, but a recent study on social influence bias (Science paywall) suggests that it would be insufficient anyway: Twitter would also have to hide all feedback on every tweet. The researchers manipulated initial comment ratings on a discussion website, and found that reader impressions were significantly influenced by fake upvotes:

The first person reading the comment was 32 percent more likely to give it an up vote if it had been already given a fake positive score. There was no change in the likelihood of subsequent negative votes. Over time, the comments with the artificial initial up vote ended with scores 25 percent higher than those in the control group.

Twitter users hardly seem to resist this effect. On the contrary, I’ve seen people request that their feeds should be filtered to show only favorited or reshared tweets. Bot sellers are unlikely to run out of business in such an environment! Indeed, users may even come to rely on socialbots to make their online experience more pleasant:

Christian Rudder [of OkCupid] said that when his dating site recently bought and redesigned a smaller site, they witnessed not just a sharp decline in bots, but also a sudden 15 percent drop in use of the new site by real people. This decrease in traffic occurred, he maintains, because the flirtatious messages and automated “likes” that bots had been posting to members’ pages had imbued the former site with a false sense of intimacy and activity. “Love was in the air,” Mr. Rudder said. “Robot love.” […]

But the bots are likely to venture into [our realm], said Tim Hwang, chief scientist at the Pacific Social Architecting Corporation, which creates bots and technologies that can shape social behavior. “Our vision is that in the near future automatons will eventually be able to rally crowds, open up bank accounts, write letters,” he said, “all through human surrogates.”

Well, that sounds encouraging. I expect that Twitter in 2020 will consist exclusively of bots following other bots. Then they’ll rally crowds of bots, collect lobbying funds, and write letters to politicians in order to end discrimination against bot accounts.

2013-08-14: Brian Krebs just published a very detailed report on Buying Battles in the War on Twitter Spam. As with other variants of spam and malware, there’s an arms race between researchers attempting to identify fake accounts and sellers trying to evade them. That race isn’t likely to end any time soon.