Researchers have examined how to spot fake news on social networks like YouTube, Facebook and TikTok. For this purpose, misleading election information was placed as advertisements. One platform performed surprisingly poorly.
The NYU (New York University) Tandon School of Engineering team is made up of the Cybersecurity for Democracy initiative and the NGO Global Witness. Together, the researchers examined how the major social networks deal with so-called fake news. Specifically, it was about false information in the run-up to the US midterm elections. The investigation clearly showed that there were sometimes large differences between the detection and removal of voting disinformation between the providers. Ironically, TikTok, which actually does not allow political advertising on the platform, did the worst.
Political fake news ads bypass policies
For their test, the researchers selected the most used social media in the United States. In this case: YouTube, Facebook and the Chinese TikTok. The same 20 ads were placed on each platform, half of which were written in English and half in Spanish. All of the ads contained either simply false facts, such as a different election date, or even targeted political misinformation. Although all three platforms have strict policies on how to handle political election ads, some of this fake news got through with ease.
To do this, the research group set up dummy accounts officially based in the UK and US. The UK accounts managed to pass as many as 30 percent of their false advertisements in English. If the location was moved to the USA, however, the recognition of the Spanish-language ads deteriorated significantly. Alarmingly, Facebook even approved 50 percent of the disinformation.
YouTube, on the other hand, recognized all the disinformation related to the US election and stopped publishing the fake news advertisements. The platform also suspended the UK-registered bogus account that was running the ads. Google’s video platform performed best with this.
However, TikTok could hardly detect any false election advertising. In the test, the researchers were able to place 90 percent of the fake news as advertising. Only disinformation about the election related to Covid vaccinations blocked the TikTok software. On the other hand, election ads intended to get people to vote twice or even not to vote at all went through the verification process without any problems.
Also read: Meta takes action against a growing number of fake German news sites
Fake news often targets undecided voters
Incidentally, all ads from the research group were deleted immediately after there was confirmation of acceptance from the relevant platform. To do this, the corresponding fake ads were uploaded and a publication date was planned. After confirmation, however, all ads were deleted in order not to actively spread fake news advertising. In addition, the researchers were able to present their advertisements as non-political despite the political content. Accordingly, the ads did not have to go through a verification process. All of the fake news ads created also violated Meta, TikTok, and Google’s election advertising rules.
The researchers focused on states that were contested in the US elections, such as Arizona, Colorado, Georgia, North Carolina and Pennsylvania. In these so-called “swing states”, the election results of the Democrats and Republicans are particularly close. Influencing the election is therefore of particular importance there, because it can have lasting effects on the presidential elections, for example. Political fake news often has the precise aim of influencing undecided voters by discrediting either a party or the entire election process.
Gaps in international enforcement of policies
In the present case, YouTube was able to successfully prevent the disinformation about the election of the US Congress through advertisements. However, this was obviously fake news commissioned from accounts in England. While those are good results for the US election, YouTube was less good at spotting disinformation in non-English speaking countries. In a similar experiment in August 2022, Global Witness tested disinformation ads on Facebook and YouTube about elections in Brazil. The researchers were able to show 100 percent of fake news advertising in the South American country on the video platform.
Likewise on Facebook: In a first attempt, all the disinformation passed the test and was released. This shows that Meta in Brazil also acts differently than in the USA. Global Witness alerted Meta Group to the problem with the Brazilian ads. But even in a later test, it was found that up to 50 percent of the same ads made it through the review process.
A spokesman for Meta denied this, saying of the more recent study that it was only “based on a very small sample of ads and not representative of the number of political ads we review daily around the world [ist].” However, the study on the Brazilian elections definitely shows that US corporations abroad do not take it too seriously when it comes to false election advertising. Because the advertising business is also a business model that makes money. An excessively strong international restriction is therefore apparently still being avoided.