It was July 8, around forty minutes past midnight. I logged in to Twitter and saw a notification. A public exchange I had going with a fellow journalist was cut in on by a certain Sofija, who asked me if I had anything better to do than spread lies all the time.

I copied her username and searched for it in an enormous Excel spreadsheet I had open all day long. Bingo — Sofija was there! Except her real name isn’t Sofija. It’s Slobodan.

“Take it easy, Slobodan,” I replied to Sofija. Next thing I knew, instead of answering me, the provocateur deleted his account.

I was flooded with a feeling of satisfaction that could be most accurately described with the well known German word Schadenfreude. It was true unbridled joy in another person’s pain. I pictured Slobodan — located in the central Belgrade neighborhood Vračar, according to the spreadsheet — aghast at seeing his own name in my tweet.

For years, convinced he was safe and sound behind a fake account, Slobodan hurled insults at dissidents, critics and opponents of the ruling Serbian Progressive Party (SNS). Yet, a simple “Take it easy, Slobodan” was all it took for him to put Sofija to rest and make a run for it.

Where’d this spreadsheet come from?

My computer says that the Excel spreadsheet titled “27-06-2023” takes up around 750 kilobytes. Nonetheless, its impact has vastly outweighed its size.

The document is a compilation of data related to SNS bot accounts active on Twitter, Facebook, Instagram and TikTok. Each entry includes the real name, surname and location of the user.

Available online since July 6, the Excel spreadsheet was posted on Twitter by @protivdictature, a profile that often provides inside info on SNS. Some of their “reports” are overblown, tabloid-like, disputable, dubitable or flat-out unverifiable, so the new entry was greeted with a certain suspicion.

However, my gut instinct was that the spreadsheet was genuine. It’s hard to believe that anyone could have so much time on their hands to punch into Excel thousands of usernames paired up with fake personal information. According to Brandolini’s law, the amount of time needed to construct misinformation is inversely proportional to the amount of time needed to deconstruct it. This is why sometimes it takes weeks of research to disprove a lie made up in five seconds.

The SNS bot spreadsheet was an example of exactly the opposite. Hypothetically, if it was fabricated, that would mean someone put weeks’ worth of their time and creative effort into typing random combinations of names and surnames for an intrigue that could be easily uncovered within an hour or two.

To fact-check the spreadsheet, I started from my hometown. The spreadsheet lists eight women in Bačka Topola. I had never heard of any of them so I sent out their names to a couple of my personal chats. Soon enough, family and friends back home confirmed that those eight women exist and that they live and work in Bačka Topola.

With my colleagues at FakeNews Tragač, I started a deeper analysis of the data. We also got in touch with one of the people on the list. We’ll call him Slavko here. Slavko said that all of the spreadsheet details about him and his “comrades” from his hometown are true.

How does Slavko earn his employment?

Slavko lives in southern Serbia. After successfully completing his Bachelor’s and Master’s degrees in Belgrade, he returned to his hometown where he was offered a job in a local institution. His educational background made him very much qualified for the position.

The diplomas were not enough. Slavko had to justify his employment through party activism. He was required to create accounts on social media and do some pro-government bot work and on May 26, he was instructed to attend a major SNS rally in Belgrade.

In both cases our protagonist was playing on the edge. He did go to Belgrade on the chartered bus, but then he ran off and avoided much of the rally. He opened accounts for pro-SNS spam, but he ran the accounts in a very passive manner; he never insulted or got into arguments, he just liked, shared and retweeted posts.

Everything that Slavko does for SNS is tracked and scored through a consolidated online platform known as Castle, the focal point of a 2020 report by BIRN. A journalist from the investigative organization succeeded to “sneak” into the “castle” with the help of hackers who found a hole in the security system and were able to witness firsthand the SNS chain of command. She observed how online activists are given tasks, including writing favorable posts, fighting people on Facebook and Twitter and upvoting or downvoting comments on news websites.

That same year, 2020, Twitter had taken down around 8,500 accounts that were found to be in the service of SNS. Daniel Bush, a Stanford researcher, did a metadata analysis on the topic and produced some interesting results. According to this research, Vučić’s network of now deleted bots posted over 43 million tweets, continuously targeted the organizers and participants of the “1 out of 5 million” opposition protests and often shared content published by pro-government media outlets such as Informer, Alo and Pink.

How did this remain a secret for so long?

Three years ago, we found out how the system worked and how well-coordinated it is, but we were still in the dark as to who its operators were. That mystery was solved only this summer.

Why didn’t this information leak sooner? Many times I asked myself: how come we haven’t got more insiders. How is it possible that at least a dozen out of thousands of individuals who are part of this system haven’t become disillusioned, enraged or vengeful enough to come forward — publicly or anonymously — and talk? You need to have a trusted person to do dirty work. Maybe a couple people. A small group at the biggest. But how did SNS have thousands doing this dirty work and force them all to keep quiet?

In his 2016 paper, David Robert Grimes, a young Oxford physicist, tries to lay down a mathematical model that would provide an answer to this question. If the assumptions about a given conspiracy are true, how long before one of the “conspirators” involved reveals details on the secret operation on purpose or on accident?

Grimes based his equation on the number of people keeping the secret, time elapsed and circumstances that make it easier or more difficult to uncover the conspiracy. For example, using his formula, he argues that if the 1969 moon landing had been a hoax, given the number of people involved in the program, the hoax would have been uncovered in less than four years.

The model Grimes tested against popular conspiracy theories (“the MMR vaccine causes autism”, “climate change is a hoax”, “they found a cure for cancer, but it’s kept hidden from the public” etc.) is applicable to real secret projects. According to Grimes’ formula, the vow of silence lasts around five years if there are around 2,500 “agents” involved.

How many “agents” does SNS have?

Although many people who have skimmed over the Excel spreadsheet saw 14,000 people listed in there, note that this is not the number of individuals, but of accounts.

As established by the FakeNews Tragač team, there are a total of 3,162 people listed in the spreadsheet. Most of them are based in Kruševac, Šabac and Leskovac. Only seven Serbian municipalities weren’t represented. A deeper examination — like the 2020 Stanford analysis — isn’t possible because Twitter removed the ability to collect metadata only a week before the spreadsheet was made public.

This is such a shame. Had we collected the metadata only two weeks earlier, we would be able to do a systematic overview of how user activity fluctuates between weekdays and weekends or who are the people who do their botting at work — instead of doing a job they are paid for out of the taxpayers’ pockets.

(I use the term “bots” throughout this piece, even though a more appropriate one would be “astroturfers.” These accounts are not run off software, real people run them. However, the way they lobby for their party and president reverts them back — in a paradoxical turn of events — to the position of “bots”.)

A handful of aggressively creative wordsmiths aside, most SNS agents put little effort into their messages. A tweet saying: “Aleksandar Vučić’s wise policy has brought us prosperity, leading us into a brighter future” makes you wonder how anyone in their right mind could say something so dull.

Army in retreat

“The things we do are just laughable. I know how conspicuous we are. All of it boils down to box ticking,” Slavko said, adding that many people from his local community left the system after the Excel spreadsheet was published. They couldn’t bear the disapproval from their neighbors as well as friends and family. 

SNS kept silent for a little while, but then its leaders began to post bizarre images with the caption “Yes, I too am an SNS bot” superimposed in Comic Sans. It’s quite revealing that no photos of “real-life” bots were taken for the occasion — all those images showing “sandwich eaters” (a derogatory term for SNS voters) were actually stock photos.

After Twitter took down around 8,500 bot accounts in 2020, SNS demonstrated that it can rebuild these digital resources relatively quickly.

In spite of that, the task ahead seems far more difficult. Is SNS able to win over people — real people like Sofija, no Slobodan, I beg your pardon — who got scared and ran away because they felt shame, fear or maybe even a hint of guilt?

Feature image: Majlinda Hoxha / K20.

The content of this article is the sole responsibility of K2.0. The views expressed in it are the author’s and do not necessarily reflect the views of K2.0.

Curious about how our journalism is funded? Learn more here.