Marcos Echevarria-Eirea is a PhD Researcher at the Department of Political and Social Sciences at the European University Institute. His research focuses on the impact of social media on democratic systems. You can contact him on LinkedIn: Marcos Echevarría Eirea.
Ranko Vranić is a Research Fellow at the Center for International Relations and Sustainable Development (CIRSD) and the Managing Editor of Horizons.
TikTok is easy to dismiss as a carousel of dance trends, cat videos, and conspiracy theories. But for younger citizens across the democratic world, it has quietly become a serious source of political news and information—and a stage on which politicians and parties increasingly compete for attention. Despite this, the platform has remained far less studied than Facebook or X, leaving a real gap in our understanding of how political elites operate there and who, exactly, comes out ahead.
Marcos Echevarria-Eirea’s upcoming paper, “The Influencer I’ll Vote For: Mapping Politicians and Parties on TikTok,” sets out to close that gap. At its core is the MPPTikTok dataset (Mapping Politicians and Parties on TikTok)—the most extensive collection of its kind to date, covering members of parliament and parties across 17 Western democracies, with more than 2,000 verified accounts. Building it was no small feat: because TikTok handles cannot be reliably scraped or pulled from existing knowledge bases, Echevarria-Eirea collected and authenticated each MP and party by hand.
The payoff is a uniquely broad, cross-national picture of political life on the platform. Analyzing both the supply of content (how much politicians post) and the demand for it (how audiences view, like, comment, and share), the paper reaches a striking conclusion. Radical party families do dominate TikTok—but contrary to the well-documented “advantage of the right” found on other platforms, it is the Radical Left that most consistently leads, in both how much it posts and the engagement it attracts. The paper also probes how much of that success rests on “virality from nowhere”—individual videos propelled by the algorithm rather than by a loyal following—and what TikTok’s ban on paid political advertising means for democratic competition.
Crucially, the MPPTikTok dataset is being made freely available to the research community and will be updated annually with new countries and elections. The paper and dataset will be available to be accessed in the coming weeks via the author’s website: https://marcoseche00.github.io/academic_website/.
In the conversation that follows, we discuss what makes TikTok distinct, why it has been so hard to study, how he built the dataset, and what his findings mean for the future of elections and democratic debate.
Welcome, Marcos. TikTok is widely seen as a place for entertainment in many ways—short cat videos, conspiracy theories about aliens and when they landed on the planet, and then serious politics. Yet your research treats it as a significant arena for political communication. What is the evidence that it matters in this political sphere? And furthermore, what sets it apart from platforms like Facebook or X that researchers have studied for years?
First of all, thank you, Ranko, for having me—it’s a pleasure. What sets TikTok apart from other platforms is a massive emphasis on audiovisual communication that hasn’t traditionally been present on X and Facebook. X especially has been mostly textual over the years, although it has become more audiovisual with the introduction of newer platforms like TikTok and Instagram, as they influence each other.
And importantly—this is crucial—TikTok has a heavily algorithmically curated form of content delivery. As a user, when you enter the platform, you are immediately placed into this algorithmically driven feed of videos, where you have very little control over what you will watch next. In that sense, the platform emphasizes interaction with content far more than interaction with your friends or the people you follow. This matters a lot, because it increases the probability that you will encounter something you were not expecting, and therefore reduces self-selection into certain content—including political content.
Let’s say you are a voter on the center-right. Because the radical right has been garnering a higher level of support on the platform, you start getting these viral videos from the radical right as small shocks to your own opinions. That might affect your views—though of course it depends on how likely you are to change your opinion—in a way that might not happen on other platforms, where you self-select into content you already like. As a platform, it has a greater probability of exposing you to out-group or out-partisan content, and therefore potentially changing your opinions.
Given that—and I think you make a good case for this separation between Facebook, X, and what TikTok can offer—why would you say TikTok has been studied less than the other two platforms that I mentioned? What exactly is missing from our understanding of TikTok, and why has it proven so difficult to study?
The first reason is that Facebook and X have been around a lot longer, and that’s given scholars far more time to analyze their content, including political content—and as you know, academia takes a long time to process things. Because they’ve been around longer, more politicians have joined too, they are seen as more politically established platforms, and they carry a lot more political and academic interest.
TikTok, on the other hand, only emerged internationally in 2017, and it was not prominent—according to the data I have—until early 2021, when parties and MPs began to join in greater numbers. There simply has not been much time to study it.
The second issue is access and data. A lot of these platforms are reticent about giving academics access, because the data can reveal patterns that might not benefit them—as we have seen in recent court cases against Facebook and X, especially Facebook, regarding teenagers, mental health, and misinformation. For X we had a very good API, and Facebook as well—useful APIs that gave researchers a lot of data. That changed in recent years, especially after Elon Musk’s acquisition of X. On TikTok these things moved more slowly. A few years ago, we finally got researcher access to data, but it’s still heavily limited. For instance—and this would be super useful—you cannot access the full list of followers of an MP or party, which would give you a sense of the ideological position of a lot of users. This would allow a researcher to match the content they consume to the MPs they follow. Unfortunately, that’s not possible right now.
The short answer is time constraints—how long these platforms have been around—and data constraints.
Moving into how you have contributed to solving some of these issues with Tik Tok: you created a dataset you call MPPTikTok. Does MPP stand for anything? In concrete terms, what does the dataset cover, and by having such a dataset, what does it allow researchers to do that they simply could not do before?
MPP stands for Mapping Politicians and Parties. In terms of what the data holds, the main thing—and it’s very useful—is the handles of these MPs and parties. I went through manually and checked, for every single MP and party across a range of democratic parliaments, whether they have an active TikTok account, and then added it to the dataset.
That gives you access to everything TikTok offers about what a user does on the platform: whether they are verified, how many followers they have, how many videos they have posted, and all the metadata for those videos—likes, shares, and so on. It gives you a real sense of how popular each MP or party is on the platform, which matters a lot for the literature on political communication, especially online.
What this mostly contributes to is the idea that the radical right dominates on social media in general. That’s been true for platforms like X and Facebook, where most of the research has been built. However, TikTok’s own affordances—the way it operates, the way users use it—and its generally younger audiences might create a space where the radical right is not the most popular, but instead shares that popularity with, say, the radical left. That’s what this paper brings to light.
And that’s important, because if we think social media popularity can have tangible electoral benefits—especially here, reaching younger voters who are voting for the first time and are very amenable to new opinions—then being popular on the platform is crucial. This opens up a broader picture of who is more popular, because previous research had only looked at individual countries, individual campaigns, or individual types of accounts, such as parties. Parties matter, sure—but what about MPs? Individual countries matter—but what about many more countries that together stand for Western democracies? In that sense it gives a much broader perspective, and it shows that it’s actually the radical left on TikTok that dominates in a number of respects. And in situations when it does not dominate, it shares that dominance with the radical right. The radical right does not stand alone, as it does on other platforms.
Before we get to what the data shows, I want to ask how you built it. You compiled many of the handles by hand, rather than by computer. How did you go about it? What was the time investment? And is there a way to automate the process, so that you are not doing it manually every election cycle, in every country that you have analyzed
In terms of the process—it’s a bit tedious, but once you get the hang of it, it goes much faster. The first thing is to go to the national parliament’s website and get the full list of MPs and parties. I normally start with parties, because they tend to be on TikTok and social media more often, since they have the party’s centralized resources. I also check their official websites, in case it’s listed there, and collect those.
Then I move on to the MPs. Here I look them up on Google—the MP’s name and party plus “TikTok.” If they do not appear, I follow the same steps directly on TikTok, in case I missed them on Google. Of course, I cannot just add any account without checking its validity, otherwise the data would not be useful. To check validity, the simplest thing is whether the account is verified—that automatically qualifies it. The second is whether it’s followed by other MPs and parties that I have already verified, and whether it follows them back, because that lends the account a lot of legitimacy. If the leader of your party follows you, this is very likely your real account. When that was not the case—which was not very frequent—I used my own discretion, based on the overall quality of the profile: the profile picture, the bio, whether it looks professional, whether they have a substantial number of followers. I also went through the followers to see if they looked like bots. And finally, and most importantly, the quality of the videos. This did not happen often, but when it did, those are the things I looked at.
In terms of hours, it took a lot—in the hundreds, though not 500. Maybe around 200 hours to do these 17 countries. I also had to write the scripts to collect their posts afterwards. But for just the usernames, I would say about 150 to 200 hours.
Is there a way to automate it? I have tried, but it’s difficult, for two reasons. First, TikTok is relatively novel and is not in datasets that were well maintained in the past, like Wikidata, which unfortunately is no longer being updated—most politicians, unless they joined very early, are not in them. Second, most politicians and parties do not list the platform on their official websites, because of its Chinese provenance—in fact, some countries have banned it from official devices, thus they have to keep these accounts on their personal devices. It still means they have them, but they do not list them anywhere that I can grab automatically using a programmed scraper. That includes parliament websites: in the contact information for an MP, they will usually have the email, the Facebook username, the Instagram username—but rarely the TikTok handle. That is the unfortunate reason it has to be done manually, and why the literature has focused mainly on individual countries, or just parties—because they only need to go through, say 500 to 1,000 parties, whereas I have gone through around a combined total of 6,000 MPs including their parties.
From the data you did collect, your figures showed that only about a third of individual MPs were on TikTok, but nearly four in five parties were. Do you have any explanation for the gap? And is the group holding back simply slower to arrive on TikTok—as you mentioned, it’s novel compared to X or Facebook—or are they staying away for reasons that may persist into the future?
In terms of the gap, the main answer is resources—and that parties, or party brands, serve as the centralizing point a voter goes to. Parties also tend to have the centralized resources—potentially a team dedicated just to the party brand, rather than to every single MP. It is therefore expected that parties will be on the platform more than individual MPs.
As for why not all MPs are on the platform, and whether that will increase: it depends mostly on the resources of the individual MP. Leaders are much more often on the platform, and on more platforms—the literature on the digitalization of politics finds this in general. TikTok may have a lower rate, even though it’s increasing rapidly, because it’s a lot more work. Because of its audiovisual emphasis, you can no longer just write a quick message, post it, and forget about it. You have to think about a video, plan it, look good, smile, speak—and you might get it wrong. It takes a lot longer and some MPs think it might not be worth it, especially those who do not want to appeal to younger voters.
Will it stay this way? I do not think so. I think we will see an equalization thesis confirmed, where mainstream MPs not only join the platform more often but also get better at engaging users on it. That’s because of TikTok’s increasing importance in many contexts—take Romania, for instance, where a relatively unknown candidate reached the second round of the elections—but also the growing prominence of younger, first-time voters, whom you want to influence, especially since, as several theories in political science hold, they will keep voting for you based on that initial experience. Short answer: I think it will increase over time.
Let us come back to your headline finding—that the radical left leads. As I was reading your piece, I made a similar connection to what has been said in the news media and in some earlier research that I have read about the favorability of the radical right in the social media sphere. Your conclusion stands, in some ways, in opposition to that standard. How confident are you in that result? And what is it about TikTok specifically that might explain such a reversal from other social media sites, or from the previous literature?
I will start with the second question to not lose my train of thought. It varies in two important ways—and this is what the revised version of this paper will try to explain in more detail.
The first is affordances—again, how a platform operates. TikTok emphasizes audiovisual content much more, and with audiovisual content it’s a bit harder to be negative. This is a very primitive hypothesis, but it may be the case that it is harder to be negative because you are showing your own face. Messages of hate, which might earn the radical right more popularity elsewhere, may not be as popular on TikTok. And that is in fact what I find in my master’s thesis: moral-emotional language on TikTok, contrary to expectations from other platforms like X, is not beneficial for an MP or a party. It does not lead to more likes—at times it leads to less engagement. If the radical left uses emotional language but not such negative moral language, that might be one of the reasons. But again, this is a very primitive hypothesis.
The more established hypothesis is that TikTok’s population is simply different from other platforms—in particular, a lot younger, and tending to be left-leaning, at least in the U.S. On a platform where younger people, who in general tend to be a bit more left-leaning—although that is changing across the Western world—make up the audience, these parties might just receive greater support on that basis.
Regarding your first question—how confident am I in these results? Very confident, in the sense that it’s the largest analysis of the platform to date and it accounts for potential country specificities. I also run two-way fixed effects, accounting for country-year specificities, and find the same result: the radical right and the radical left are both more popular than the mainstream.
But you are right that there are limitations, if that is what you are implying. The first is that the results are conditional on adoption—being present and active on the platform is itself conditional on adopting it. The results may reflect a more strategically selected set of adopters: for the left, say, the ones going into this are those with the biggest teams, who are best at communicating, and maybe the better-looking ones—again, because of that audiovisual aspect. There is some endogeneity, or unobservable variables, that might be driving these results. And it’s descriptive, I do not make any claims about the potential effects or benefits. Although the big-picture analysis pushes back against that a little, I still need to look at the radical-right-versus-radical-left comparison by country, in addition to the two-way fixed effects—I am waiting to see what the Journal reviewers for this paper say. And finally, there are API errors that can creep in. I ran it with two different sets of code, one in R and one in Python, to minimize that, but there is little one can do beyond acknowledging this limitation.
Let me expand a little into your theory. You divide the analysis into the supply of content—how much politicians post—and the demand for it—how much audiences watch, like, and share. The radical left appears to lead in both, as we have mentioned already. Could you explain what each side, supply and demand, captures exactly? And which do you think is the better measure of real political reach—or is it a combination of both?
In terms of the two dimensions—these were proposed by my supervisor, Kevin Munger, and his colleague—they deal with the two main dynamics that take place on any social media platform: some people post things for others to react to, and vice versa.
On the supply side, the features that shape supply determine how much people post. On YouTube, for instance, this is slower, because videos tend to be longer and take more effort, whereas on TikTok, because they can be very short, you might upload hundreds of videos a year. The first thing the supply side assumes is that an account needs to be open—without an account you cannot post at all. And the idea is that the more you supply the platform with your own content, the more likely something is to catch on, or that someone encounters your profile. It’s important to be out there; the more you are out there, the more likely you are to be found.
But if those videos are not popular—if people are not engaging with them, if they do not like the actual content—then it is not just about quantity, it is also about quality, and the video will not get very far, and neither will your profile. You also need the audience to engage: to view, first, because they could just skip the video, and TikTok would record that as a very short view and penalize it. They first view the whole thing, or as much as possible, then like it, comment, and share it—because that distributes the video across the platform and hopefully reaches a wider audience that starts watching your videos more often. This is the demand side.
One concept on the demand side that can stimulate demand for radical parties is “virality from nowhere”—that a video can take off because the algorithm itself pushes it, not because a politician has built a loyal following that keeps giving them the same attention. You find that radical parties rely on this “virality from nowhere” more than others. How did you measure that reliance? And does it qualify the picture of their success?
Thank you for that question. Quickly—the reliance is a bit of a double-edged sword. It also means their content is being amplified to greater extents on the platform; it’s just that not every post is being amplified. It goes against the theory that the radical right is always the most amplified—here we see it’s the radical left. But on the negative side, it means you might be more reliant on the algorithm, and if you do not get it right, your videos will not go very far.
In terms of how I measure it, again I follow my supervisor and his colleagues: I create a measure of how equally distributed the engagement is across an account’s set of videos. Say I’m an MP with one million views across a bunch of videos, but only one video has the million views—that means I went viral once. I have a much higher average than a lot of people because of that one video, but all my other videos did very poorly. That’s what it’s capturing. And what we see is that most parties and MPs operate a bit like that: they have a small set of videos that went very viral, but most stay a bit under the radar. That’s how social media works in general—not every video can go viral, because of time constraints, people’s attention constraints, and so on. Can you repeat the final part of your question?
Does it qualify the picture of their success? Does it undermine it, limit it—or is it potentially even a positive, in the sense that they know how to game the algorithm?
That’s a great question, because it is something that we will start seeing more often. It has a positive side, as I said: it shows that these parties are just better at knowing what the algorithm wants, what people want, what’s viral and interesting. For instance, it could be a trend—like in the French parliament, where showing your office became very popular at one point. These people are just better at knowing that kind of thing. In that sense they are better adapted, and it could be positive, because as long as they keep up with the algorithm, they should stay more popular.
The negative side is that it can be fleeting. If you get cancelled—cancel culture on social media—or if for whatever reason you are no longer hitting the right notes with your videos, you might run out of a lot of those views. But most importantly, it might mean your audience is not very attached to you as a person or as an MP; they simply come across your videos occasionally, because they go viral, they like them, and that’s the end of it. They do not think of you as someone whose videos they want to see all the time. It could be a more fleeting audience.
And the final point: it also means mainstream parties could potentially catch up. In fact, in the appendix of this paper, there are figures showing that over time the trend has been waning a little and even reversing on some metrics. This could be because of what the literature has studied on other platforms—an equalization trend over time, where mainstream parties get better at these platforms, invest more resources in them, and eventually catch up to these radical or outsider/new parties.
Expanding on that into the broader conversation about TikTok’s role in our democratic discussions. TikTok is now a major news source for younger citizens and unlike other platforms, it bans paid political advertising leading to a scenario where reach cannot be bought. If certain actors are simply more effective under these conditions, what does that mean more broadly for our elections and for democratic debate in general?
In one sense it’s positive, because the richest party can no longer simply flood the platform with their content. But there is an asterisk: they may no longer be able to do so formally, but they can still do it informally, through influencers. TikTok wanted to reduce that kind of advertising, but it forgot to account for the fact that a politician can simply pay an influencer who, with their own organic audience, spreads their message. The literature has shown this is indeed happening, unfortunately—these influencers might go against, say, the academic consensus on something like climate change, which I would argue is negative for democracy. It could also spread misinformation, which I think we all agree is negative for democracy.
Hence, something that looks quite positive—that TikTok might level things between richer and poorer parties, or up-and-coming smaller parties, and therefore offer the electorate more options—might not work, because of this informal payment system emerging through influencers who have already built an organic audience of people who will listen to them as almost the absolute and final truth on any matter.
You are planning to release this dataset freely, and to expand it each year to new countries and elections. Looking ahead, what questions do you most hope it will help answer, and what ideas do you already have for expanding the research further?
The reason I make this commitment is that I think it’s crucial to see how things evolve over time—especially on social media, because things change very quickly on it. TikTok is massive now, and it could remain massive over the next few years. But what’s especially important is that the algorithm itself might change over those years, and it’s important to capture the before and after. For instance, when the Trump administration made a deal to now own the algorithm in the U.S.—that’s important to record before and after, because it can tell us a lot about what happens with algorithms and how they matter for political opinion in a natural setting, rather than an experimental one.
I think an important missing connection in the literature on social media in general—but especially the newer platforms like TikTok and Instagram—is the link to the actual benefits of being popular on social media. There’s a lot of correlational work, but it’s really hard to get at the causal side, because there are many unobservable variables and there is a lot of endogeneity. Having as fine-grained a dataset as possible gives researchers a better chance of uncovering those causal mechanisms—or at the very least, of producing new, better-motivated questions for causal experiments that are then better informed.
And this moves on a little to my own future research and my PhD thesis, which is what I’m trying to do there: get a very fine-grained dataset and try to uncover the actual impact of being on social media on the probability of a party faring better in elections, public opinion, traditional media, and so on.
Do you include in this future research any position on what role AI is taking in all of this? One pejorative for AI videos on TikTok and YouTube is “AI slop”—these AI-generated videos that maybe are not good, not bad, but some people love them. Do you see that having an impact on political messaging?
Absolutely, I think it has an impact on both the messages and the data collection. On the messages front, we have now seen a lot of radical-right parties—this is being studied by some researchers, for instance at UNC-Chapel Hill—posting videos that are completely AI-generated, often to send xenophobic messages against certain minorities. These videos, in the past, took a lot longer—they had to be produced, maybe even with an actor—whereas now an intern can create one on their own computer in a few hours. That’s the main impact on the messages these parties can put out. But also in terms of misinformation: AI is, I think we all agree, going to increase that risk a lot more—not only for younger citizens, who are more native to this type of environment, but especially for older adults, who might not be able to discern what’s real and what’s fake.
In terms of data collection and the research itself, I hope it increases the rate at which I can collect these accounts—but unfortunately, we are not there yet. For instance, TikTok blocks LLMs from accessing its website, so you cannot really ask ChatGPT to collect all the MPs for you and decide whether a username seems real or not. But when we get there—and I think this is one of the better uses of AI—my dataset will stand as a gold standard to check whether AI is actually able to do this. In that sense, I think it will remain relevant as a dataset, hopefully for many years to come. That’s my answer.
