Algorithms and democracy

Summary

The Finnish Innovation Fund Sitra commissioned The Behavioural Insights Team (BIT) and  Bondata to examine the political content encountered by 18–24-year-olds on social media and assess its potential risks to civic discourse. Such risks are among the systemic risks identified in the European Union’s Digital Services Act.

BIT conducted a systematic audit using 24 avatars on Instagram, TikTok and X in Finland, France and Romania. Avatars simulated different levels of engagement in political content, including a ‘Tilted trajectory’ phase where they signalled exclusive interest in either left-ofcentre or right-of-centre content. In total, 1,719 political posts were manually coded for political leaning and problematic content, including misinformation, conspiracy theories and hate speech.

Bondata complemented the audit with an online survey of 3,063 young adults in the same  countries, examining social media use and exposure to problematic political content. 

Right-wing content dominated in the BIT platform audit, accounting for 58 per cent of all politically classified posts, compared with 26 per cent left-wing and 16 per cent centrist content. This dominance often persisted even during the ‘Tilted trajectory’ phase when avatars signalled interest in left-wing politics, suggesting a disproportionate amplification of rightwing perspectives regardless of user preferences. Romanian feeds were an exception: they were largely dominated by centrist content, particularly government communications. Bondata’s survey complemented these findings and found that 44 per cent of users in Finland who strongly identify with the left felt that the content they received corresponded very poorly to their views, compared with only 5 per cent of those strongly identifying with the right, with similar results in France and Romania.

The audit revealed algorithmic unpredictability and a lack of user control. Engagement  signals had no consistent or reliable impact on the content delivered, and feeds could shift  suddenly and substantially without any clear trigger. 

The majority of problematic content observed did not violate community guidelines. Outright misinformation, conspiracy theories, and hate speech were relatively rare in the BIT platform audit. Instead, feeds were dominated by unverifiable, opinion-based content (67 per cent of all posts), often expressing extremist views.

The audit identified AI-generated content as an emerging trend. This included deepfakes of politicians and synthetic avatars (such as cartoon animals) used to disseminate offensive humour, strong political messaging, or hostile commentary. Overall, 5 per cent of political posts were clearly AI-generated.

Findings demonstrated the ongoing deterioration of social media quality, sometimes referred to as ‘enshittification’, as platforms shift from prioritising users’ experience to  maximising engagement and monetisation. Bondata’s survey indicates that more than one third of respondents encountered problematic content regularly or repeatedly. Bondata’s  survey also found that half of young respondents reported feelings of disappointment, fear, anger, or sadness when encountering political and social discussions on social media.