Audit of takedown delays across social media reveals failure to reduce exposure to illegal content

TL;DR

This study examines the relationship with a systematic audit of takedown delays using data from the EU Digital Services Act Transparency Database, covering five major platforms over a one-year period, and points to the benefits of faster content removal to effectively curb the spread of illegal content.

Abstract

Illegal content on social media poses significant societal harm and necessitates timely removal. However, the impact of the speed of content removal on prevalence, reach, and exposure to illegal content remains underexplored. This study examines the relationship with a systematic audit of takedown delays using data from the EU Digital Services Act Transparency Database, covering five major platforms over a one-year period. We find substantial variation in takedown delay, with some content remaining online for weeks or even months. To evaluate how these delays affect the prevalence and reach of illegal content and exposure to it, we develop an agent-based model and calibrate it to empirical data. We simulate illegal content diffusion, revealing that rapid takedown (within hours) significantly reduces prevalence, reach, and exposure to illegal content, while the longer delays measured by the audit fail to reduce its spread. Though the link between delay and spread is intuitive, our simulations quantify exactly how takedown speed shapes exposure to illegal content. Building on these results, we point to the benefits of faster content removal to effectively curb the spread of illegal content, while also considering the limitations of strict enforcement policies.

Figures (9)

• Figure 1: Pipeline for estimating the effectiveness of illegal content removal. (a) Empirical data calibration: We estimate a plausible range for the probability of illegal content $p$ from policy reports and infer platform-specific takedown delays from Statements of Reasons (SoRs) in the DSA Transparency Database (DSA-TDB) for five major platforms: Facebook, Instagram, YouTube, TikTok and Snapchat. These delays are summarized by the expected content takedown delay $\tau$ and used to parameterize content removal. Simulations are run on an empirical follower network derived from Twitter data. (b) Information diffusion and removal: An agent-based model simulates the spread of content over the network. At each timestep, illegal content survives with constant probability $p_s$, calibrated from the empirical takedown delay $\tau$; once removed, content is deleted from all user feeds. (c) Moderation effectiveness estimation: Simulations are performed across a broad range of takedown delays to quantify the impact of removal speed. The effect of content removal is measured as the relative reduction in the prevalence and exposure of illegal content, compared with a baseline without removal.
• Figure 2: Complementary cumulative distribution functions (CCDFs) of time delay in illegal content takedown on five platforms. Each plot reports empirical data from the DSA-TDB (green dots) along with an exponential fit (purple line).
• Figure 3: Impact of takedown delay on illegal content. The reduction in the prevalence (pink), reach (green), and impressions (purple) of illegal content is plotted as a function of the expected takedown delay $\tau$. Shading represents the 95% confidence intervals estimated using non-parametric bootstrapping. The vertical lines indicate the expected takedown delays obtained by fitting the DSA-TDB data for different platforms.
• Figure S1: Robustness of results with respect to different distributions of $p$ for $\tau=2.89$ (median takedown delay of two days, top) and $\tau=11.54$ (median takedown delay of eight days, bottom) based on 10 simulations. The illegal content reduction does not vary significantly ($P \ge 0.13$ across pairs of simulations with different distributions, using Mann–Whitney U tests with Bonferroni correction).
• Figure S2: Robustness of results with respect to different high-risk group sizes $s_H$ for $\tau=2.89$ (median takedown delay of two days, top) and $\tau=11.54$ (median takedown delay of eight days, bottom) based on 20 simulations. The illegal content reduction does not vary significantly ($P \ge 0.72$ across pairs of simulations with different $s_H$ values, using Mann–Whitney U tests with Bonferroni correction).