Content Moderation on Social Media Platforms: An Empirical Analysis Through the Lens of the Digital Services Act

Antoni Mut-Piña (Universitat de les Illes Balears) & Rosa Barceló-Compte (Universitat de Barcelona)

Forthcoming, European Society for Empirical Legal Studies Annual Conference, Ljubljana, June 2026

Research question: Does the Digital Services Act (DSA) create the right incentives to ensure content moderation that effectively balances users’ freedom and online security?

Abstract: This paper analyses how the DSA structures private enforcement in the field of content moderation, with a particular focus on dating applications. It combines doctrinal analysis with empirical evidence derived from the Statements of Reasons (SoR) issued under Article 17 DSA in this sector. The findings show that moderation decisions on dating platforms are grounded predominantly in the enforcement of contractual terms and conditions rather than in assessments of illegality, positioning Article 14 as the key regulatory provision. The study concludes that significant vagueness persists in both platforms’ terms and conditions and in the explanations provided for moderation decisions, incentivising broad and low-cost moderation practices. As a consequence, brief and standardised SoR remain the prevailing practice, and no substantive shift towards enhanced transparency or accountability has yet been observed, despite the enforcement and sanctioning regime established under the DSA.

Data: The analysis draws on the DSA Transparency Database (DG CONNECT, 2023), which contains metadata on content moderation actions enacted by major online platforms operating within the EU. Due to the database’s size (>30 billion observations), a random sample was constructed from dating app platforms, ensuring proportional representation. The final sample consists of 60,000 observations per platform, totalling 4,202,630 observations.

Method: Logistic regression models are employed to identify the key variables explaining differences in the vagueness of content moderation across platforms. Vagueness is defined as a binary variable: a SoR is vague (1) if the user cannot infer why their content was removed based on the available information, and not vague (0) otherwise.