Yesterday, a US jury held Meta and YouTube liable for addictive platform features, showing that the legal battle over social media is increasingly shifting from content to design. The EU, too, has begun to confront this issue, while proposals to ban social media proliferate. The European Commission’s preliminary finding on TikTok’s addictive design from last month might be a game-changer for protecting users’ digital well-being under EU law.

The Commission’s focus moves beyond illegal content on the platform to the design of the platform itself, in particular, the features such as infinite scroll, autoplay, push notifications, and personalised recommender systems. These addictive design features can push users towards compulsive or prolonged use of social media and bring risks to users’ mental and physical well-being. In the Commission’s view, TikTok breaches Art. 34 of the Digital Services Act (DSA) because it did not adequately assess and mitigate the systemic risks that arise from these features.

This blog post begins with the Commission’s findings to advance the first steps toward a theory of digital well-being within the EU platform regulation framework.

The (legal) battle for attention in social media design

The starting point is simple: social media platforms do not merely host content; they compete for attention. In the attention economy, user attention is the key resource from which platforms derive value (UN, 2023). The longer users stay, click, scroll, and react, the more data is collected, the more ads are sold, and the more effective the platform’s recommender system becomes. Attention is therefore not a by-product of social media design. It is the business model.

This helps explain why certain design features (e.g., infinite scroll, autoplay, etc.) are central: they are techniques for capturing and retaining attention, often by reducing friction, increasing immediacy, and making disengagement harder.

But when does optimisation for engagement turn into the engineering of dependency, loss of control, or other foreseeable harms? We argue that the threshold is crossed when it creates harm to “digital well-being”, which we propose to operationalise in simple components and requirements.

“… including minors and vulnerable adults”

The Commission’s preliminary findings refer to risks for users’ physical and mental well-being, “including minors and vulnerable adults”. That emphasis is understandable: some users are plainly more exposed than others to the harms generated by attention-capturing design.

Under the DSA, vulnerability is not developed as a clear, self-standing category with defined criteria. Rather, it appears only in a fragmentary way (e.g., minor-specific protections in Art. 28), which makes it tempting to fall back on broad labels instead of a more precise analysis of how vulnerability is produced in platform environments.

A better approach is to treat vulnerability as contextual and relational, more in line with the wording in recitals 94-95 of the DSA, which refer to users “in vulnerable positions” rather than labelling some users as “vulnerable adults”. Users may be placed in more vulnerable positions because of power-imbalanced relationships, personal characteristics, and socio-political determinants (Malgieri 2023; Rebrean & Malgieri, 2025). The point is not that certain groups are inherently vulnerable. It is these specific “addictive” design features that may interact with existing asymmetries of power or capacity in ways that make disengagement harder and harm more likely (Helberger et al., 2022). Vulnerability here is not fixed – depending on the situation, anyone can find themselves in a vulnerable position online.

If the DSA is to take digital well-being seriously, it should not look only for abstract “vulnerable users” but for situations in which platform design foreseeably intensifies exposure and dependency, reduces control, and exacerbates social disadvantage. Minors deserve special attention, but they are not the only users for whom the architecture of attention can become an architecture of vulnerability.

Digital well-being at risk

Article 34 (1)(d) DSA explicitly lists serious negative consequences to the person’s physical and mental well-being as a systemic risk that VLOPs should consider. These risks can stem from online interface design that stimulates behavioural addictions among recipients (Recital 83 DSA). In addition, the Commission’s Guidelines on DSA for minors identify digital risks as including negative impacts for physical and mental health and well-being, e.g. addiction, depression, anxiety disorders, deregulated sleep patterns, and social isolation.

However, the DSA does not provide a clear definition of well-being, which creates challenges for compliance and enforcement of Art. 34 DSA. Other EU laws are also not helpful. Although Article 3(1) TEU identifies the promotion of well-being as a general EU objective, references to well-being in secondary EU laws are very slim (see Recital 2 GDPR; Recitals 6, 27, 48 AI Act).

In practice, well-being is a multidimensional concept whose meaning depends on the analytical lens and objectives (e.g., psychological, social, economic). Some definitions draw on psychological understanding of well-being, often described as “functioning well and feeling good” (Ryan & Deci, 2001), others take a more socio-economic approach and are used to measure people’s living conditions, focusing on material conditions, quality of life, and social relationships (OECD). In general, competing approaches to well-being can be grouped into four categories (Adler & Fleurbaey, 2016): objective goods theories, subjective well-being approaches, preference-based approaches, and capabilities approaches(Nussbaum & Sen, 1993). This diversity, subjective elements, and measurement challenges makes the concept difficult to operationalise for regulatory frameworks such as the DSA and Art. 34.

The digital well-being formula

If the Commission is serious about treating addictive design as a systemic risk under Art. 34 DSA, enforcement needs a practical way to translate a broad concept (“physical and mental well-being”) into something that can be assessed, evidenced, and mitigated.

What we propose is a digital well-being formula based on two complementary components: attention abuse plus fundamental rights cascade impact. Attention abuse is measured through time spent online and exposure (content intensity and emotional intensity). Cascade impact affects at least one of the following 3 values: autonomy, emotional/mental health, or social participation.

The cascade impacts on fundamental rights

Features such as infinite scroll, autoplay, streaks, and persistent notifications are engineered to reduce friction and keep users in a loop. Users stay longer or return more often than intended and find it harder to stop, even when they want to. In regulatory terms, this looks like manipulation through design – an infringement of autonomy, which can be understood as an “individual’s capacity to make meaningfully independent decisions” (Susser et al., 2019). The harm to autonomy is not only that users spend more time online, but that these features can lead to radicalisation or to commercial or electoral decisions that users would not have otherwise taken (Susser et al., 2019; Yesilada & Lewandowsky, 2022; Whittaker et al., 2021).

The advantage of an autonomy lens is that it does not require a clinical diagnosis of “addiction”. A platform can cause foreseeable loss of autonomy at scale, even if only a subset of users develops severe symptoms. The DSA’s logic is systemic. If the design predictably undermines user control for a meaningful proportion of users, especially those in vulnerable positions, then risk assessment and mitigation should address it.

Another value affected by cascade impact is health. Health harms are the most cited, but also the easiest to mishandle, either by overstating the evidence (“social media causes X”) or by demanding an unrealistic standard of proof (“show clinical addiction in each case”). A more credible approach is to focus on problematic or compulsive use patterns and their foreseeable negative consequences, without collapsing into single labels.

On the substance, the health risks most consistently discussed around attention-capturing design include anxiety and depressive symptoms, dysregulated sleep, reduced sustained attention, and reinforcement loops driven by intermittent rewards. “Dopamine” shorthand is common, but it is usually better to avoid neuro-deterministic claims and keep the language cautious. A design that systematically lengthens sessions, accelerates consumption, and encourages constant checking can plausibly raise health risks even when the content is not illegal.

Addictive design can also impair social participation. Recital 83 explicitly mentions social isolation as a form of DSA systemic risk to well-being. At a basic level, prolonged and compulsive use can displace offline activities, including sleep, exercise, schoolwork, in-person relationships, and civic engagement. Over time, that displacement can contribute to social isolation and loneliness. The social participation frame also aligns with EU law’s broader recognition of social participation as a fundamental right and value. Participation is a recurring value in areas such as disability and accessibility laws, where the problem is often defined as the interaction between individual conditions and external barriers that limit “full and effective participation in society”. That same relational logic fits platform design.

“Attention abuse” components

The risks to autonomy, health, or social participation often increase as platforms capture (and abuse) users’ attention. However, in the social media context, attention is difficult to grasp and measure directly. We propose analysing attention from the perspectives of time (both time spent online and loss of time control), content exposure, and individual responses to it.

Time can play a role as a revealer of attention abuse and as an “amplifier” of the risk associated with a platform’s addictive design. Different empirical studies examined the role of time across 3 dimensions.

The first dimension is total time spent on social media. Heavy use of social media has been linked to a higher risk of low mood or depression (Frielingsdorf et al., 2025; Liu et al., 2022). Some research also suggests that intense digital media use may come before these health problems (Tang et al., 2021).

Second, certain design choices on social media platforms, such as short video formats or infinite scroll, can reduce users’ ability to correctly assess how much time they spend online and therefore weaken their time control (Jiang et al., 2025; Tobin & Grondin, 2009; Ruiz et al., 2024) as well as create gaps in their memory (Baughan et al., 2022; Ruiz et al., 2024).

Third, the moment of social media use is also relevant. People who check social media before going to bed are more likely to experience sleep disturbances (Levenson et al., 2017).

Time spent on social media is only part of the picture. What matters is also exposure to content (content intensity) and users’ responses to it (emotional intensity), which together can increase risks to autonomy, health, and social participation. This is particularly disturbing when platforms use highly personalised recommender systems (Metzler & Garcia, 2023). Even if users limit total time on social media, they can still be exposed to aggressive advertising or emotionally engaging content that might affect their behaviour and decisions (Opoku et al., 2025; Savani, 2025). Responses to exposure may depend on individual factors, including personality traits or motivations for using social media (Lin et al., 2017).

A conclusive proposal for the Digital Fairness Act

The Commission’s preliminary findings on TikTok matter because they treat platform design – not just content – as a site of systemic risk. For the DSA to make this enforceable, “digital well-being” needs to be operationalised in a way that regulators and platforms can actually measure and mitigate. In our suggested simple structure, attention capture is the upstream driver; harms to autonomy, health, and social participation are the downstream effects. Those harms are often amplified by two practical proxies for attention, namely time (including loss of time control and timing of use) and exposure (including emotional responses and the effects of personalised recommendations).

At the same time, the DSA cannot carry the entire burden. Article 34 applies only to VLOPs, while attention-capturing and dependency-optimising design is widespread, including in smaller services such as dating apps. This is where the Digital Fairness Act could usefully complement the DSA by spelling out a clearer set of problematic practices. The Digital Fairness Act is an EU legislative initiative (still under development) aimed at addressing problematic practices, such as dark patterns, influencer marketing, addictive design features, and online profiling, particularly when they exploit consumer vulnerabilities for commercial gain. One option to regulate these practices is a tiered approach, similar to the Unfair Commercial Practices model: a “grey list” that shifts the burden onto platforms to show effective risk mitigation when they deploy features like autoplay, infinite scroll, push notifications, or personalised recommendations, and a “black list” trigger where multiple such practices are combined. This would help ensure that regulation captures not only individual practices but also their cumulative effects on users’ time and attention, as well as the risks to autonomy, health, and social participation.

The post Towards a Legal Concept of Digital Well-Being  appeared first on Verfassungsblog.