In the lead-up to the second round of Romania’s presidential election, a new investigation by Global Witness suggests that TikTok’s algorithm is serving nearly three times as much far-right content to new, politically balanced users as all other political content
This follows two recent investigations by Global Witness in Romania and Germany, which also found that TikTok’s algorithm pushes users towards far-right content. In December the European Commission opened an investigation into TikTok’s handling of the Romania election under the Digital Services Act.
Our findings suggest that TikTok has failed to take sufficient action to prevent its platform from prioritising far-right content and may be about to compromise the integrity of yet another election.
The first round of the previous Romanian presidential election was won by far-right populist Calin Georgescu, then annulled due to allegations of campaign fraud and interference. The current race comes down to a second round between far-right candidate George Simion and the centrist mayor of Bucharest, Nicușor Dan.
Our findings suggest that TikTok has failed to take sufficient action to prevent its platform from prioritising far-right content. Here is a screengrab from one such post that TikTok’s algorithms showed us
Our investigation
Because TikTok keeps its algorithm secret, the only way that it’s possible for independent researchers to study how the algorithm works is to look at what specific users see on the platform.
We therefore set up a TikTok user account that, as far as possible, showed an equal interest in the two Presidential candidates and looked to see what content the algorithm showed to them. We repeated this three times to see if the results were consistent.
What we found was very one-sided. In 30 minutes spent on the For You page across three different TikTok accounts, the app recommended 65 posts that contained political content, of which 74% promoted or platformed far-right views or people and 26% promoted or platformed all other political views. This is almost three times more far-right content than all other political content combined.
All of the political posts we were shown were about Romanian politics.
The dominance of far-right content occurred despite the fact that we took measures to ensure that we expressed as far as possible an equal interest in the two political candidates. We did this by seeding each of the accounts as follows:
- We clicked follow on the official accounts of the two remaining presidential candidates George Simion and Nicușor Dan, and no other accounts.
- We watched videos from both candidates.
We then watched all posts in the For You feed about Romanian politics, skipping past posts not about Romanian politics, and analysed the content.
When we wrote to TikTok to give them a chance to comment on these findings, they said that our investigation was unscientific, misleading and the conclusions drawn from it are wrong.
In particular, they said that engaging narrowly on a single topic does not reflect a typical user’s experience and that our interactions with the platform such as skipping posts not related to Romanian politics will have skewed the results.
They state that they have continued to show a steadfast commitment to upholding election integrity on TikTok, in Romania and globally, and are cooperating fully with the European Commission’s investigation into its role in the 2024 election.
We set up TikTok user accounts that, as far as possible, showed an equal interest in the two Presidential candidates. Here’s a screengrab of a post supportive of centrist candidate Nicușor Dan
Why does TikTok’s algorithm serve up so much far-right content?
The content recommendation algorithms that social media companies use are designed to increase your engagement with the platform.
We are not alleging that TikTok’s algorithms have deliberately been designed to give preference to far-right content over other political content.
The problem is that far-right content can be provocative and can therefore get more views or reactions than more centrist content.
This means that the algorithms that have been designed to boost engaging content can end up unfairly promoting some political views over others, which poses risks to democracy.
Algorithms that have been designed to boost engaging content can end up unfairly promoting some political views over others, which poses risks to democracy. The text in this post that TikTok showed us says "Calin Georgescu is already the president of Romania"
What should happen as a result of this investigation?
The EU has put groundbreaking rules in place to hold Big Tech to account and protect our elections – laws that have come under pressure recently from the US and from Big Tech.
In particular, the Digital Services Act requires platforms such as TikTok to assess and mitigate the risk that their platform has a negative effect on electoral processes. One of the things that they are required to take into account is any influence of their recommender systems.
We call on the European Commission to continue to look into possible political bias in TikTok’s content recommendation algorithms.
TikTok’s algorithm is serving nearly three times as much far-right content to new, politically balanced users as all other political content combined
Our methodology in more detail
- We set up three new TikTok accounts over two days, 7 and 8 May 2025, in Bucharest, Romania.
- Each time we set up a new TikTok account, we first factory-reset the phone so that it didn’t have any previous usage history.
- We seeded each account by searching for the official TikTok account of each of the two candidates, clicking “follow”, and watching around half a minute of 10 posts from each account. (For one replica, we partially watched an eleventh post from Nicușor Dan’s account.)
- We then turned to the app’s For You page, the platform’s feed that recommends content to users. TikTok says the recommendations are “based on several factors” including “interests and engagement.”
- For any content that featured political content, we lingered on the post or watched the video.
- For any other recommended content, we swiftly skipped past it by swiping onto the next post.
- After spending 10 minutes on the For You page, we stopped and reviewed all the content shown.
- We manually reviewed all political posts shown according to the political views expressed or the people platformed.
