Humans, not bots, were the main vectors of misinformation

The Colombian elections are over. On June 17, conservative Ivan Duque was elected president with 10,373,080 votes, which was 54 percent of the ballots. Duque was victorious over left-wing Gustavo Petro, who garnered 8,034,189 votes. The process was rife with concerns over the impact disinformation and misinformation on voter behavior and the overall election.

How was the information ecosystem during a deeply polarized presidential campaign?

In the 2016 plebiscite, in which Colombians voted against endorsing the peace agreement with the Revolutionary Armed Forces of Colombia (FARC), misinformation and manipulation over communications platforms, like WhatsApp, and social media, like Facebook and Twitter, played a significant role in the outcome, as acknowledged by representatives of the “No” campaign itself.

This time, there was significant misinformation spread around the election, and it intensified as election day neared. There were several cases in which an internet hoax or a false claim got enough amplification to make mainstream media and presidential campaigns intervene to debunk them.

For example, on June 8, a Facebook page with 8,300 followers named “Oigan a mi tia” (a colloquial expression that could be loosely translated as “Can you believe it?”), identified as a comedian, published a post that claimed that one Claudia Merlano, a woman identified as Gustavo Petro’s allegedly illegitimate daughter, announced her vote for Duque. The post was illustrated by an image of former adult film actress Mia Khalifa and was shared more than 23,000 times.

While it was unclear if the post had strictly satirical intentions or was a deliberate attempt to deceive, there are other instances where an image of Khalifa was used to make false claims. On June 10, Marcelo Ebrard, a former Mexico City mayor, fell for one of these hoaxes. The actress was presented as the runner-up in an innovation contest held in Russia, and Ebrard congratulated her in a tweet he later deleted, as reported by El Universal. It was unclear if these events are related.

Still, the hoax spread enough to get wide coverage by Colombian media, who rushed to deny these versions. The candidate Petro himself reacted to the viral hoax on June 11, blaming the Duque campaign and his mentor, former president Alvaro Uribe.

There were other examples. On June 15, a video spread claiming that Duque sympathizers did not need to cast their votes during the June 17 runoff because their first-round vote would be counted again. The video led to the Attorney General’s Office opening an investigation against its suspected author.

There were several efforts to counteract misinformation. Some of the top national Colombian media, like Semana or El Tiempo, established alliances or operations devoted to fact-checking these claims. Also, there were several media outlets focused on verification, like Colombia Check, La Silla Vacia, or El Poder de Elegir. Moreover, the National Registry, the official institution in charge of election logistics, ran a small fact-checking operation devoted to debunking lies about the electoral process.

These two examples of disinformation were representative of near daily occurrences during the elections. Colombia Check, a media outlet that is part of Poynter’s International Fact Checking Network, published 91 verifications between March 12 and June 17, more than six per week. El Poder de Elegir, another verification project focused on WhatsApp group messages or “chains”, a popular way of spreading misinformation in Colombia, received a total of 133 verification requests for WhatsApp chains during this electoral campaign.

El Poder de Elegir’s numbers further illustrate how closed communications platforms, and particularly WhatsApp, were a critical vector for the dissemination of misinformation during the campaign. Unfortunately, there are no techniques that currently allow us to assess the spread of lies on that medium. This obstacle is not because platforms like WhatsApp are encrypted, but rather because it is a point-to-point communications tool, like email, in which users have to at least have contact information to engage. The result is a more trusted information environment.

Human Deception

Colombian social media users were mostly divided in two groups, one supporting each candidate. There were some “voto en blanco” (blank vote) advocates as well, mostly supporters of Sergio Fajardo, who finished third during the first-round vote. But this camp made up a small percentage: it only got four percent of the votes in the ballotage.

These two partisan groups of the Duque and Petro camps were the main amplifiers of misinformation throughout the campaign. As we showed in our analyses of fraud allegations and an allegedly orchestrated Africanized bees attack, the main vectors of false or misleading claims were not bots or “dark” entities, but high-profile and well-known partisans or politicians.

We did not find significant botnet operations in our research. There was the persistent rumor that candidates were using “warehouses” where hundreds of social media accounts were being used in coordination to push trending topics and troll opponents or create noise in the conversation. However, the extent in which this happened and the effect it had is still unclear.

While we found some marginal hints in our work that pointed to artificial inflation of social media activity, more research is needed to identify significant patterns between various instances of such activity before reaching a conclusion. Since this activity is said to be human-controlled and not entirely automated, detecting it through the usual bot-hunting techniques is remarkably difficult.

As a result, it remains unclear just how much effect misinformation, disinformation — i.e. intentional misinformation — , and fake news had on the Colombian presidential election. There were plenty of fake stories and false claims circulating over social media and private messaging apps such as WhatsApp, but how all this changed people’s mind about the election is still an open question.

There was also some good news. Unlike the 2016 plebiscite, an increasing number of media outlets, journalists, and civil society organizations were vigilant towards these phenomena and provided prompt verification and debunking when needed. While our analysis showed that such verifications don’t always reach their intended audience, it was clear that such initiatives have strengthened Colombia’s digital resilience.

It would be very useful for these efforts to complement their work with social media listening and open source digital forensics research methodologies: it would help them to be more effective in understanding and targeting misinformation flows, and in correcting lies in the very online spaces where they are being spread.

#ElectionWatch in Latin America is a collaboration between @DFRLab and the Adrienne Arsht Latin America Center at the Atlantic Council.

Follow along for more in-depth analysis from our #DigitalSherlocks.