“I was a bot”: the confessions of an agent dedicated to cheating on Twitter
A former agency employee explains how the sophisticated cheating machinery works on the social network
“I coordinated a trolling team made up of about 10 people. There are bigger teams in the sector, ”says a community manager who for much of the last decade worked at an international agency that sold artificial message amplification services on Twitter. “I have worked on projects in five countries, including Spain. There is demand for this type of contract that is not announced in Yellow Pages ”, he says.
With this experience, he now wants to openly reveal how this dark business capable of influencing public opinion works. For a few weeks now, he has started to tell details on a Twitter account ironically called @thebotruso. Third party agencies had verified his role in the agency, has exchanged dozens of messages with @thebotruso about how the famous bot and trolling campaigns for companies, political parties or sports clubs are prepared and work today. The former employee remains anonymous because a confidentiality contract prevents him from specifying the specific content or clients of his former work.
Twitter is not a replica of real life. But the debates or opinions that dominate the network often jump to the media. Or even to the rostrum of Congress: “They are filling the streets with bots,” Santiago Abascal, leader of Vox, told Pedro Sánchez on Wednesday. A part of the accounts that they discuss on Twitter are false. The problem is knowing how many or, more difficultly, what they are and what influence they have. “I liked to attract the attention of like-minded journalists. It greatly increases our ego that a journalist takes the contents of your trol account as a source in his articles, in addition to helping to gain visibility, ”he says.
1. Not everything is bots
By default, bot is used as a false account on Twitter. But a bot is an automated account and it doesn’t have to be fake. Years ago, bots were the main people in charge of amplifying messages: they made thousands of retweets or likes or were used to swell followers’ accounts. They continue to do so, but Twitter and the sophistication of users make them easier to discover and remove.
@Thebotruso’s agency has software to program bots. Give orders to retweet such accounts or launch pre-written tweets at set times. It is cheap but not very refined. “We are still working to amplify content, but we have to be careful, because it can pose a reputational problem to the client,” he says.
To avoid this, there are several resources that make it possible to humanize or disassociate these bots from a campaign: “The bots will not retweet everything that moves. If they carry out actions with, for example, a specific troll, you generate a pattern, with which a Competitive data analyst would be able to lift the hare, “he says.
The bots will also not follow the account of the paying customer in case they are ever exposed. “If we take the independence cause as an example, the bots will follow different parties, politicians and associations, but not the final client,” he explains. Accounts will also be monitored for real users interacting with any bot. In that case, an employee will come in to respond.
2. The key is two types of trolls
The usual definition of trol is hooligan or impertinent. At the agency they were his star accounts. Each employee could manage about 30 accounts, with their human behavior. The trolls were divided into alpha and beta.
Alpha accounts spread the message. They started with a follow me strategy and I follow you to gain weight, their initial tweets were inflated by bots and then they interacted with important accounts to get attention. Having nothing to do with it, in Spain the account of Miguel Lacambra became famous in a few weeks, achieving 20,000 followers with a similar strategy.
The beta trolls are the guerrillas. They are dedicated to tame criticism. These are accounts that respond to celebrity tweets with insults or threats. “Those affected by the beta attack see the responses to their tweets and often cut a little when tweeting about according to what topics. They feel uncomfortable and want to have a lower profile. The system is effective. That is why it continues to be hired and perfected. Users continue to strain it, ”he says.
3. The goal: trick Twitter
Part of the effort is devoted to avoiding detection of the traps and increasing the probability of success: “The trolls will not follow each other and the interaction between themselves will be limited to lead several conversation groups and because a habitual relationship between two or more trolls would put the operation at risk, ”he says. Maps of the accounts are made with a common tool among researchers called Gephi to see if the relationship between them is remarkable: “Those gephis would be able to expose the relationship between our accounts (trolls and bots), so every precaution is little, “he says.
There is also a permanent objective in all these operations: avoid patterns. “Each account that a troll manages must write differently: people have a tendency to use certain expressions and leave writing patterns, such as putting two exclamation marks or ending all sentences with an ellipsis,” he explains.
The tricks to trick Twitter are even more remarkable with automated bots: “The software carries all the bots, which are divided into groups. Each group uses an API [tool to use Twitter automatically]. And the IP address is randomly varied. We got to have 3,000 or 4,000 accounts in the same API and I know that more could be used. The problem is that if you have many accounts tweeting often under the same API, you can cause Twitter to block it, “he explains. Although these tools allow you to manage thousands of accounts, you have to personalize one by one with a photo, name and bio. It’s a long job for Twitter to load 2,000 at a time.
4. Every campaign has a plan
Each campaign is prepared with data analysis and daily goals. Bots and trolls are not created in a vacuum. Before a campaign, data scientists analyze the conversation on the subject that interests the client: a party that wants to get more deputies, a company that wants to beat a competitor or a club that has credibility problems.
“They see how many accounts are participating in a topic and it is established how many would be necessary to have influence,” he says. The sentiment and influencers of these issues are also analyzed: “Lists of favorable and contrary accounts are made and the weight they have is analyzed,” she says. This information is key to the operation of the campaign: “It helps the troll to know which users to interact with, to generate a cluster with them, which ones to respond to and harass with the beta and with which users it is not even worth wasting time . Starting an account in a crazy way is not the same as doing it with an immense amount of information. The alpha troll knows who to address, what tone, what to communicate. ”
A campaign can cost a million euros. The client expects concrete and demonstrable results. An average campaign may require between 1,500 and 2,000 bots and trolls.
While normal users on Twitter go to the network to see what happens, these operatives who manage hundreds of fake accounts have a daily plan. It is as if every day a mafia boss launches a group of commandos into a city with a delicately designed plan to do a set of specific missions and spread panic without anyone detecting them. The goal is to make citizens believe things that are not true: not only with false news, but with actions that show that there are more people who believe something than there really are. It would be something like inflating surveys. Of course, they often face teams that pretend just the opposite.
This does not remain in theory, its effects have consequences in the real world: “People tend to share their opinion when they feel supported by the community,” says this agent. “There was a time when many non-independence Catalans did not publish their ideas on networks because if they entered Twitter it gave the impression that half the world was independent.”
These actions in networks allow to open the way to extreme opinions. Suddenly, someone sees that open criticism of immigrants or women is allowed. Perhaps it is not done crudely, but with memes or code words, but there it is. It is not known if those accounts are controlled by a dozen employees in an office.
5. How to cheat a journalist
Journalists are often unaware of how easy it is to pass fake merchandise on to them. One of the reasons @thebotruso has created his account and wants to tell his experience is to try to warn of the dangers: “Those of you who work in the media don’t always know this ecosystem. In a way, it is easy to fool a journalist. You are always looking for information and, today, the Internet is a very large source. If a journalist finds your content, sees that he has support and matches what he wants (or has) to communicate, perhaps he will take it. The strategies for each case are different. For example: for a company accused of corruption, we created an ecosystem that defended jobs (workers concerned about the measures being asked to take against the company), ”he explains.
6. How to cheat a survey on Twitter
Bots are useful for winning a survey online. These are the steps that this agency did: “The survey is detected and a screenshot is made. A calculation is made to know how many votes are needed to rotate the results and the value of each 1%. The operation is launched to win the desired option. It is monitored to see that everything is working correctly. ”
7. The creation of a ‘trending topic’
The options for achieving a trending topic were smaller, but the strategy was clear. The agency worked for days to achieve its goal. “The first thing is to choose a day and an hour. We try not to coincide with events such as a soccer game or Big Brother ”, explains @thebotruso. “You have chosen a hashtag, it is important that it has not been used since they are more difficult to position. The team of writers who write thousands of tweets over several days is launched to be published by the bot network, “he adds.
Real users are also notified in case they are interested. “Communications are sent to like-minded people to warn of the action: on such a day at that time we will come out with such a hashtag to complain, we invite you to the action to see if we can be a trend and make them listen to us.”
The big day is here: “Analysts look at how many tweets it takes to sneak into trends. Tweets are uploaded to the bot platform. The client removes the first tweet. The botnet is launched quickly. It is crucial that there are many tweets in a short space of time. Alpha trolls come out with hit tweets. Data analysts monitor to find out if more tweets are needed, if the botnet needs to be stopped. Beta trolls support action, respond to critics, encourage other users with the same ideology. ”
After the mission, whether it is successful or not, it is time to hide the evidence: “The bot action stops and it is cleaned: the bot accounts tweet and retweet other topics so that, if someone enters to see those accounts, they will not see They only came in to tweet about trending and went to sleep. The trolls may continue to tweet for a while and then they will clean up, just like the bots, “he adds. And finally: “Finally, a report is prepared for the client.”
And all this, how does it affect who does it? “We were all aware of what we were playing at. And I like that word: game. I always recommended to everyone to take it as playing Risk because it is easy to take your projects personally and sometimes it is difficult to manage the difference between you and your trol account. Luckily, I never had anyone with depression, anxiety, or anything like that. ”