Understanding political content with PoliTikTok
Can TikTok engagement predict political success? NRK analysed more than 10,000 videos to find out. Through the PoliTikTok project, they explored how politicians use the platform and how audiences engage with political content. By combining AI with editorial analysis, they moved beyond assumptions to understand what really happens on TikTok during the elections.
A data-driven approach to political content
Rather than relying on individual observations, NRK built a large-scale dataset of political TikTok activity. Around 3,000 political candidates were initially considered, of which around 700 active TikTok profiles were identified. Their content was systematically collected in 2025, resulting in a dataset of more than 10,000 videos and 267,000 comments before the elections in September. NRK shared the database with Faktisk.no, who where planning a similar project, and collaborated on the analysis.
Using a combination of APIs and AI tools, NRK gathered metadata such as captions, comments, likes and shares, transcribed spoken content through speech-to-text, and extracted text from visuals. The dataset was then analysed using large language models and queries in the database to identify patterns and themes at scale.
What actually happens on political TikTok?
Rather than confirming existing theories, the data revealed a more complex and sometimes counterintuitive reality. The political ecosystem on TikTok turned out to be highly diverse, with a wide range of political opinions. While some politicians managed to generate strong engagement, this visibility did not necessarily translate into political influence.
NRK also observed clear differences in audience reactions. More radical opposition voices and protest-driven content received the most positive engagement, in some cases up to 75% positive comments. In contrast, politicians from governing or established parties faced more negative responses, with up to 80% negative comments. The most discussed topics remained relatively traditional, focusing on themes such as the economy, taxes, and the job market.
Most notably, there was no correlation between TikTok popularity and election results. One of the most engaging politicians, Gyda Oddekalv, got lots of support on the platform, but did not secure a seat in parliament. The eventual winner, the Labour Party, had a more limited presence on TikTok. This shows that the often more provocative content that performs well on the platform does not necessarily translate into votes. The analysis also found no evidence of bias in the algorithm towards specific ideologies, nor any signs of foreign interference influencing the election outcome.
From experimentation to editorial insight
AI played a big role in enabling the analysis of such a large dataset. It allowed NRK to classify topics, analyse sentiment and identify engagement patterns across thousands of videos.
At the same time, the project clearly showed the limits of automation. NRK wanted to experiment with automated fact checking as a part of the workflow as well, beeing in the initial stage of a long term collaboration with Factiverse. However, they soon learned that it would be way too time consuming to review 10.000 automated fact checks, so this part was eventually left out of the project.
Editorial judgement remained essential throughout the process. AI proved to be a valuable tool to filter and structure large amounts of data, for example, by making spoken content searchable and enabling filtering of comments from positive to hate speech, which allowed journalists to focus on interpretation and verification.
Key takeaways for media organisations
What does this tell us about how to analyse and understand new platforms like TikTok?
- AI makes it possible to analyse platform ecosystems at scale, but only when combined with strong editorial framing and clear research questions.
- High engagement on platforms like TikTok doesn’t equal real-world impact, as seen in the gap between viral political content and actual election results.
- Audience behaviour differs depending on political positioning, with opposition and more provocative voices receiving significantly more positive engagement than governing parties.
- AI is most valuable as a filtering and structuring tool (e.g. transcription, sentiment analysis), while interpretation and verification remain human tasks.
- Large-scale analysis can challenge assumptions and provide evidence-based insights, replacing theories with data-driven understanding of platform dynamics
By combining large-scale data analysis with editorial expertise, NRK’s PoliTikTok project shows how media organisations can better understand the role of platforms like TikTok in elections, without losing sight of what truly matters in journalism.
