Money Can't Buy Friends

But it can buy Twitter followers, and that’s good enough, right?

A friend of mine was wondering recently how many people on Twitter get thousands of followers, despite apparently never tweeting anything profound. One answer to this is that Twitter doesn’t favour profundity, but another is that people simply buy them.

I’ve done a fair amount of sampling from Twitter, using its APIs as well as services such as DataSift and Gnip. There are some estimates that estimate that most people on there are bots, interacting with other bots in a big human-like graph of plausible deniability. On various projects we’ve glimpsed these, or have got caught in their subgraphs when traversing, but largely they remain hidden from normal users.

For some people, this is a problem: they want to use Twitter to measure some kind of human impact or reach: Twitter routinely detects and bans certain types of bot (such as those responsible for spam or manipulation of metrics), and there are 3rd-party services to do this too.

Mo is not one of those people. He heard this tale and imagined the fame and fortune that comes from having a small army of devoted followers hanging on his every word.

So, we went on fiverr.com and bought him three thousand new friends, for the princely sum of £3.84.

We weren’t quite sure what, if anything, would happen, but the seller has good feedback and we know this is technically possible. Delivery time: 3 days.

For the first 24 hours nothing happened, then our contact counted Mo’s followers and reported back, presumably to prove the count. Near the end of the delivery period, followers started flooding in:

Over the next 12 hours or so, Mo’s follower count ballooned to just over 4,000 before sinking slowly back into the high 3000s. We wondered if this atrophy would continue, perhaps as Twitter caught the bots or the seller removed them to seem plausible, before one last wave of bots added to the count again.

At the time of writing, Mo has 3,905 followers. About 3500 of these are not real.

There are a few things we found interesting about this process.

The first is that the bots themselves are not all obviously bot-like. Many have real names and real tweets, and seem to either be taken from real people, or else very convincingly constructed. Some of them even have Twitter’s blue “verified” badge ().

This leads to some interesting questions, which presumably someone more familiar with the shady world of SNS marketing will be able to answer: are these accounts fabricated using NLP, or are they stolen by compromising client machines and retrieved using a botnet? Do the controllers of the Twitter botnet simply rely on people not noticing they’re following new people?

Secondly, the manner of Mo’s transition from zero to hero was nonlinear and gradual. It seems that this is a mechanism for avoiding Twitter’s detection algorithms: the ‘second wave’ of followers was connected to the first, in a sort of friend-of-a-friend manner that apes real interactions:

We still don’t know why followers disappeared: if Twitter detected the accounts one would imagine it would also detect most of the fake accounts on the botnet. Perhaps this is another ploy to seem more organic.

Clicking on Mo’s fake followers now allows us to find other clients of the service we used. This is intriguing: it’s a list of smallish companies and organisations who clearly wish to be seen as popular, and care more about that appearance than the reality of actually sending a message to many people. It seems these services are used as ‘seed followers’ to convince people of their legitimacy.

However, the way we’re able to list these companies as fellow clients buys them the opposite. Someone we know who studies trust networks for sentiment analysis pointed out that this is a great way to smear rivals (particularly in his field): when we bought his followers, Mo didn’t need to do anything — in fact, I bought followers for him, and simply provided a link to his profile. It would be fairly easy to stuff someone else’s follower list full of fakes.

For now, Mo is very happy with his bot army. We remain curious as to what will happen as Twitter improve their bot detection, or what difference his newly-engorged follower count will actually make to his web presence: will people index or rank him higher? Will bots we actually care about (like those that aggregate users in various fields like NLP) start caring more when he tweets? Will his new-found popularity allow him to realise the dream of being followed by Barack Obama?

As ever, the line between seeming something and actually being it is blurry.