Why Did 9,000 Porny Spambots Descend on This San Diego High Schooler?

A voyage into the strange underworld of spambots, shady marketing, and non-human intelligence.

Bots can be scary. This is not Olivia. (Shutterstock/mast3r)

It was around 5pm last Thursday when Olivia, a San Diego high school student, noticed that something interesting was going on with her Twitter account.

A swarm of 30 women with sexy profile pictures had just followed her on the social networking service. "guys wtf 30 PORNSTARS JUST FOLLOWED ME WHATS HAPPENING," she tweeted to her 600 or so followers.

Her friends started joking with her. One said: "I think they want you to join their profession." And it was a little funny. Weird, but funny. ("actually laughing so hard right now [emoji]," one friend tweeted at her.)

One minute she's doing homework and congratulating a friend on making varsity. The next, she's the center of the this swarm of porny weirdness. It was like the setup for a new Spielberg sci-fi movie.

Olivia posted screenshots from her account. She wrote in perhaps ironic all caps, "AM I PAYING FOR THIS" and "IS ANYONE ELSE AS CONCERNED AS I AM."

But... maybe being followed by pornstars would be the next trend at school? "why arent hundreds of porn stars following me," a friend tweeted.

Olivia started to notice some patterns, though. Each new follower was following precisely six people and almost everyone else the accounts were following were "verified" by Twitter. Many of the verified were legitimately famous people.

Seriously: what was going on?

Shantal Roddam (@Allieqtzm) was a typical example of one of her new followers. Shantal was a "Friendly beer fan" from Butte. She was following:

@ESPN, the world's leading sports brand;
@MarsPhoenix, a long-dead robot on Mars;
@ReutersScience, the news organization;
@KingJames, Lebron James, the NBA star;
@AlexisMadrigal, your faithful correspondent;
and Olivia, a high school student in San Diego.

By 8:25pm, Olivia could announce, "I have hit 3,000 everyone 3,000 porn stars."

At 9:05, she crossed 4,000. At 9:51, she hit 5,000. She changed her Twitter bio to, "5,000 pornstars follow me and idk what to do." (idk means "I don't know" for the acronymically uninitiated.)

A boy tweeted, to no one in particular, that Olivia was "officially famous as fuck wtf." Another said, "Let's be honest we all knew that Olivia was going to be twitter famous from the start." A third said, "New game: take a shot every time someone follows Olivia."

All the new followers had names like "Earlene Timperman" and "Valerie Wienandt" and their bios were like Mad Libs for lame social media wannabes: "certified food nerd," "Hardcore social media scholar Bacon ninja," "Typical tv trailblazer Hardcore introvert," "Bacon specialist Certified organizer," "Friend of animals everywhere Coffee enthusiast," "Coffee advocate Hipster-friendly analyst."

Hardcore social media scholar Bacon ninja.

They all hailed from seemingly random cities: Fairmont, Danville, Trenton. Never a state. Never a country. Never a joke.

Oh, and none of them had actually tweeted anything.

* * *

Perhaps you have guessed what happened by now. These "people" were not people at all, but automatically generated accounts created by somebody with a bit of programming knowledge.

The thousands of new followers that Olivia got were spambots emanating from the same source.

Now, if you are reading this story on the Internet, you have probably encountered spambots, or at least the spam that such bots generate.

Generally speaking, the bots tend to follow really popular accounts. And they tend not to come in swarms of thousands but one or two at a time, maybe a few dozen at most.

So the mystery remained: why was a San Diego high schooler suddenly a spambot magnet?

I began to search through Olivia's followers looking for patterns.

The first thing I noticed: Olivia wasn't part of every bot in the swarm's follow list, but she was predominant. No other account that I could find had been targeted so often, not even Lebron James.

There is an underground economy in fake-account creation, as Newt Gingrich discovered when his campaign was accused of buying Twitter followers. What people are buying, of course, is not real people, but robot-generated accounts created to make it look like people are more famous than they are.

This kind of bot normally just picks accounts from Twitter's suggested user list, the Lebron Jameses and ESPNs. But perhaps someone had tried to up its sophistication by including something some regular users. Or maybe there was some sort of bug in its "Who should I follow?" code.

Other kinds of bots decide to follow people based on what they tweet, but looking at just a few examples, it was clear that there was no content connection between Olivia and the other people these bots were following.

The second thing I noticed: the spambots were following a lot of golf caddies.  I couldn't explain that one immediately, but keep it in mind.

The third thing I noticed: Olivia wasn't the only San Diego high schooler. At least three other San Diego high schoolers, two of whom Olivia knows, were also targeted by the spambot. These kids, though, only got (at most) a few hundred spambot followers.

All this evidence led one of my followers, @001010110, to devise a wonderful new romantic comedy/Nico Muhly opera plot:

"Then there's the whole 'nerdy teenage boy creates botnet to impress a girl, follows others to cover his tracks' scenario. Poor kid who caddies at country club falls for rich girl whose family are members, pulls stunt to prove he's worthy..."

Which I, for one, loved as an explanation!

But then the bots started tweeting.

* * *

At first it was just one here or there, like the early kernels in a bag of popcorn.

Soon, most of them were tweeting. Not excessively, but a handful per bot. The messages weren't going out to the famous accounts, but to regular users across the world. They typically looked like this:

A typical bot-driven ad.

While the words in the tweet changed from person to person, each one tried to lead people to tweevip.com. (Don't go there.)

Tweevip's IP address suggests that the computer running the site is located in Roubaix, France, in the north, right near the Belgian border. The same Internet address also hosts these high-quality websites. Most seem set up to capture similar types of bot-generated traffic (for example vineluv, gramvip, which would seem to target Vine and Instagram).

Other spam sites you should avoid.

The point of Tweevip is to get you to enter your phone number and name. When you do, the site sends you a text message that says, "MOVIE EXTRAS WANTED! Make money in the movies. All looks, No Experience Required! To register call 1877-590-5505."

I called the number and confirmed with the representative on the phone that he was working for a site called Casting360.com. Meanwhile, he was aggressively trying to sell me their casting services.

What are they trying to get people who call the number to do? They want them to sign up for a $1.98 14-day trial, which gets auto-upgraded to a $34.90 monthly membership.

In exchange, you can use their casting services. How accurate or useful are they? I searched for my hometown, Ridgefield, Washington (pop. 5,260), 25 miles north of Portland, Oregon. In the local area, the site found 25 "entertainment professionals," almost all of whom are called "casting directors." The top three results were: a guy looking for a drummer, a freelance photographer, and a 16-year-old  with "four years experience behind a camera, and editing on both final cut pro, and iMovie." Hmm. The site sure doesn't seem like a way to break into the industry.

This is what Google suggests when you search for Casting 360:

Casting 360 is run by Igor Reiant, who previously ran Talent6, a similar casting service that was fined $45,000 by San Mateo County for fraud, as revealed by an anti-scam blogger.

Reiant, according to his Twitter feed, appears to like skiing in Squaw Valley. He has 16 followers. I contacted him to ask him about his company's marketing tactics, but he has not responded.

* * *

This is the current state of play in the Internet economy. At the top of the heap, there are the Twitters and the Apples, the name technology companies that are creating billionaires.

Then, down at the bottom, there are the bots, simple pieces of code that do one thing and one thing only, but do it relentlessly.

And in between, there's you and me and Olivia and Igor and LeBron James, and we're all connecting to each other and the Apple brand and Google and Twitter and the bots. These are strange relationships. The same bots that help scam some credulous person who wants to get into acting might make a San Diego high schooler feel famous.

We all want something from these networks of technologies. In a strange way, we all depend on one another. Igor needs the bots. The bots need Igor. I need Igor and the bots and Olivia. Twitter needs all of us, though they claim in regulatory filings that only five percent of their accounts are fake, based on an internal review. (It should be noted: the spambot problem definitely used to be worse.)

And yet, despite all of our connections and interdependencies, the logic of the bots remains mysterious to human beings. We know why Casting360 or whatever shady marketing company they hired sent out the bot swarm: to get "customers" and make money. But why did the bots decide to follow a San Diego girl and a bunch of golf caddies?

Trying to answer that question is like staring into the eyes of a snake. Or as Olivia summarized the existential state of social technology: "5,000 pornstars follow me and idk what to do."

But, I am willing to take a guess at the idea behind the metabot that created all the spambot accounts, with a huge assist from my colleague, Ian Bogost.

There are two things to keep in mind here.

One is that the tools of the metabot are the tools of the Twitter API, which is the interface that other software uses to interact with Twitter. That API allows coders to do some things that the standard user cannot, like return a list of all the people someone follows all at once. The metabot, that is to say, is not using only the tools you have available on your phone.

The second thing is that Twitter does have people working on bot detection, and regular users themselves are hip to the spambot game now, too. So, bots have to make some effort to disguise themselves. They have to "look" human. That's why they have those stupid bios; those are just snippets from other humans' real bios that have been chained together. That's also why they have those porny pictures. In some book of best practices for spam, it has probably been determined that you are X percent more likely to get someone to follow an account, if the avatar picture is sexy. And they want to be followed because that makes them more likely to be classified as "real" by Twitter's anti-spam bots.

Of course, the metabot does not want to create a single fake account that looks human. It wants to create tens of thousands of fake accounts that look sort of human. And it knows it should not overwhelm one user with 10,000 fake followers (though that's what happened). Rather, the bots should be spread out among many people, so that no one gets suspicious that thousands of porny spambots are following them.

So, the metabot has to find a way to swing from one set of people to follow to another set of people to follow.

We think the metabot first creates fake accounts that follow the most popular feeds. In the hundreds of fake accounts I've looked at, @ESPN and @JimmyFallon show up almost as often as Olivia, and that probably shouldn't surprise us because they are the first two suggestions Twitter gives, as you can see in the screenshot of the new user sign up to the right.

Remember, though, the metabot needs to spread the followers around. It needs more humans to follow than just Fallon and ESPN.

So, our guess is that the metabot requests the lists of people the really big accounts are following — "walking the network graph" — and then, when it creates a new account, it follows some of them. This sends spam followers to a wider and wider network as the metabot moves out from the most popular accounts.

That's why we see clustering in the types of accounts that any single spambot account is following, even though the bot swarm ranges across topics.

In my case, I was included with a bunch of science accounts (because Twitter actually recommends me for people who like science). But there are other spambots that are following mostly hockey accounts, or the aforementioned caddy cluster, or a fashiony group.

The metabot, therefore, is viral. You get followed because of who follows you.

This tendency explains the strange geographical cluster among San Diego high school students. Perhaps one of those kids was being followed by a really popular account (like @Interscope records, perhaps, which follows hundreds of thousands of people), and through that link, the bot stumbled into this little circle of San Diego teens.

All of this activity would have remained under the radar, of course, all part of the silent non-human web. Except something went awry. For some reason, Olivia got stuck in a weird loop, and the metabot kept spawning spambots that chose to follow her over and over, relentlessly.

Maybe once the metabot reached the San Diego kids, a bug kicked in. Instead of negative feedback keeping her (and everyone else) from being followed too often, we got runaway positive feedback. The bots followed her because other bots followed her. And on and on.

Which is, perhaps a kind of reasoning that we can understand:  It's the core logic of fame and celebrity itself. Attention flows to Snooki because attention flowed to Snooki. Attention flows to Olivia because attention flowed to Olivia.

Olivia and her friends weren't wrong when they thought she'd become suddenly famous. Her audience just wasn't human.

By Friday evening, Twitter's anti-bot team had deleted all of the accounts, and Olivia's follower count had returned to normal. She tweeted, "I'd rather have a couple of real followers who love me than thousands of fake ones."


Things were back to normal.