Author: Paula Gori, Secretary-General and Coordinator of EDMO, based at the Florence School of Transnational Governance at the European University Institute 

Very large online platforms and search engines (VLOPEs) have become public spaces where information is shared and accessed and conversations among citizens take place. This has completely re-shaped public discourse, the actors and the format of information content as well as access habits and preferences. It has also opened an animated debate on such a space being hosted by actors who run for profit and more broadly on the EU digital sovereignty. 

Freedom to hold opinions and to receive and impart information and ideas is a key pillar of democracy. As such, public civic spaces must be based on transparency and accountability and guarantee information integrity. However, this is not what is currently happening on the main social media platforms. On one side, users (citizens) face opacity and uncertainty over how algorithms work and why given content is prioritised over others. On the other hand, malign actors profit from the design and functioning of such social media platforms, search engines and AI Chatbots to implement their agendas (mainly political and/or economic). Meantime, to little is done by those players to ensure information integrity and an even level playing field.  Outrageous content gets generally more viral and is hence more monetary profitable. 

The EU Digital Services Act (DSA) was introduced to contribute to solving this issue at a structural level. In the words of the European Commission, the DSA aims at creating a space which respects the rights of citizens and is based on a risk-based approach. Risks which stem from “the design or functioning of their service and its related systems, including algorithmic systems, or from the use made of their services” (Article 34). The DSA Regulation has been directly applicable in EU Member States since February 2024. However, its full enforcement turned out to be a rather longer process (e.g. delays in the nomination of some Digital Services Coordinators and in the adoption of some of the delegated acts). As it often happens with legal tools, the interpretation of the different articles and definitions (and/or lack of) animates the debate around the enforcement of the DSA.  

Under the DSA, VLOPSEs must run, at least yearly, risk assessment exercises to identify whether their design or functioning (as above mentioned) can pose actual or foreseeable negative risks to several circumstances listed in Article 34, including to civic discourse and electoral processes. These assessment exercises must be followed by the identification and adoption of relevant risk mitigation measures. This whole activity goes through an independent auditing done by a third-party organisation and is sent to the European Commission and the relevant Digital Services Coordinator. 

Now, what emerges from the first reports is that: i) The different VLOPSEs adopt different risk assessment methodologies and different ways of labelling the gravity of the risks; ii) The methodologies themselves are not explained in detail; iii) The reports abound in listing a series of activities carried out by the VLOPSEs as ex-post mitigation measures, but little information is provided on the results of the risk assessment exercise (besides very broad categories and wording). In other words, little is said if they indeed found evidence of systemic risks stemming their design and functioning (including algorithmic systems) and consequently how they addressed it structurally; iv) the focus is more on the malign actors and on Coordinated Inauthentic Behaviors (CIBs) to which those risks are attributed. This reasoning is however not aligned with the DSA, which aims at addressing the structural issues (thus design and functioning of the VLOPSEs). In other words, the possibility of such malign actors operating on the VLOPSEs and implementing malign tactics, techniques and procedures lies also in the way those platforms are designed and function. As such, changes in the design and functioning may consequently obstruct malign actors from operating so easily, as they would lack the infrastructure to operate on. 

Access to the data of the VLOPSEs for research purposes (foreseen in Article 40 of the DSA) is key not only to monitor and assess compliance with the DSA, but also to contribute to the detection, identification and understanding of systemic risks. Processes for data access (both public and private to vetted researchers) are still not successful. This is delaying the opportunity to get a clearer picture and understanding of what is happening on the VLOPSEs and on the actual risks listed in the DSA.  

The study on Algorithms and democracy: How social media shapes young Europeans’ worldviews, commissioned by Sitra and run by BIT is inserted into this picture and  provides very valuable and key results, which can be included in the broader exercise of risk assessments under the Digital Services Act. The study looks at political content (i.e. potential systemic risk to civic discourse) provided to young people. In particular, it analyses if such content is recommended to users and if it is possible to identify patterns and/or reasons why that given content appeared. The focus is on young people aged 18-24, in three EU countries (Finland, Romania and France) on three social media platforms (Instagram, TikTok and X). Avatars, whose behaviors changed over the different sessions, were used. In particular, they simulated different levels of engagement in political content, including a ‘Tilted trajectory’ phase where they signaled exclusive interest in either left or right centred content. Very importantly, throughout the exercise, of the 1719 manually coded political posts, none was political advertising. 

It is worth underlining that political and opinionated content are at the core of the civic discourse. What is problematic is the opaque algorithmic amplification which does not allow us to know why a given content is seen in that very moment and consequently leads to the loss of agency on our information diet.  

Political content shared through VLOPSEs is the subject of many research and policy debates. Many politicians and political groups legitimately have their accounts on those platforms through which they share their political opinions and positions. In addition, political content can also be promoted as political advertising. In this regard, the Regulation on Targeting and Transparency of political advertising has recently entered into force in the EU. As a reaction, META and Google decided to no longer provide political advertising as a service in the EU market, claiming that compliance with this new regulatory tool would be too costly. TikTok as well does not allow for political advertising on its platform. However, research shows that such content is actually present on those VLOPSEs. 

The study run by BIT and commissioned by SITRA is extremely interesting for several reasons. 

First, it confirms previous research results finding that radical right-wing political content is disproportionately amplified on social media. As above mentioned, the political content encountered by the avatars did not include political advertising, hence the legitimate question of why content from one side of the political spectrum is more recommended than the other (despite the behaviors of the avatars). The results of this study may be aligned with findings confirming that negatively emotionally charged and outrageous content gets more viral and is consequently more profitable. Such content is statistically more likely to be shared by radical right-wing parties. The virality of such type of content should not be neglected also in light of the decision of some platforms not to allow political advertising. In particular, if politicians cannot pay anymore to increase the reach and visibility of their content, they may opt for an outrageous and misleading communication style to make content organically more viral. 

Another key finding of the study is the unpredictability of how algorithms behave. The engagement signals (online behaviours) of the avatars had no consistent impact on the content they were recommended. In other words, users cannot predict nor impact which content the algorithm will decide to provide them with as recommendation. This is even more alarming if we consider that according to the study, there were cases of sudden shifts leading to the exposure of extreme content, without any clear trigger by the avatar. Transparency and accountability are key features of democracy and information integrity. A public civic space which does not respect these principles is not a safe place for citizens. 

The direct consequence of this (also mentioned in the study) is that citizens’ agency is undermined. If the agency of citizens in how they want to get informed is undermined, then information integrity and the democratic process are undermined too. Citizens are provided with content they have not asked for and are not interested in, to the detriment of their agency and to plurality of information. 

Studies like this are important for both research and policy purposes. The more empirical evidence on mechanisms happening on VLOPSEs, the more we can advance in the overall understanding of what happens in the public space and elaborate evidence-based approaches to ensure it is safe and even for all citizens. It would be extremely valuable and helpful if the same exercise could be run also in the remaining EU Member States. This study highlights common trends and distinctions among the three countries involved, and it would represent a significant opportunity to have an overall EU picture. A similar study could also be run by focusing on influencers, if and how their content is labeled by platforms when they share political content and how this relates to political advertising. 

To conclude, this study is a strong reminder of the importance of proper enforcement of the DSA and the key role played by Article 34, 35 and 40. It also is an important nudge to advance conversations around incentives to building European platforms based on fundamental values  alternative ways of content moderation/curation, such as the middleware solution. According to the latter idea, VLOPSEs would keep their infrastructure level but competition would be opened at the level of the curators. Users would then be able to choose the certified curator they prefer, which would then curate the content by accessing the platform via an API. In other words, a transparent algorithmic competition model. Such a solution would also be in line with a commitment taken by the European Commission in the EU Democracy Shield, namely to prioritise enforcement actions promoting an independent and diverse media landscape, including by increasing transparency of online advertising, under the Digital Markets Act (DMA). 

When we participate in public life and democratic debates, we are not users, we are citizens. We must have agency over the content we access and we must ensure our debates happen in a safe, fair, transparent, accountable space based on democratic principles.