An analysis of 1.58 billion self-reported content moderation actions has found that Europe’s largest social media platforms showed no meaningful signs of adjusting their moderation practices around the 2024 European Parliament elections, even as they publicly claimed to have deployed election-specific safeguards. The study, published in EPJ Data Science, drew on the European Union’s Digital Services Act Transparency Database, a centralized repository launched in September 2023 that obliges very large online platforms to report every moderation action taken within the EU.
The researchers, from a team including Benedetta Tessa, Gautam Kishore Shahi, Amaury Trujillo and Stefano Cresci, examined an eight-month window spanning roughly 14 weeks before and 20 weeks after the June 2024 elections, held across all EU member states from 6 to 9 June. They analyzed statements of reasons, the standardized records platforms must submit for each moderation decision, from the eight largest social media platforms in Europe: Facebook, Instagram, LinkedIn, Pinterest, Snapchat, TikTok, X and YouTube. TikTok dominated the dataset with 646.1 million records, followed by Instagram with 300.8 million and Facebook with 260.2 million, while LinkedIn contributed just 293,000.
The central question was whether platforms changed the volume or timeliness of their moderation in response to heightened integrity risks. Using time series analysis, decomposition into trend and seasonal components, anomaly detection and change point detection, the team found that moderation activity remained largely stable before, during and after the elections. Trend slopes shifted only modestly across the pre-electoral, electoral and post-electoral phases, and the changes could not be traced to election-related content. The researchers excluded the election days themselves and the 18 July presidential election day to avoid spurious fluctuations.
The analysis of moderation delays proved revealing. Several platforms, including YouTube, Pinterest, Snapchat, Facebook and Instagram, showed noticeable increases in delay in the post-electoral period, particularly from late August to early September 2024. This raised a hypothesis that platforms may have deferred moderating election-period content until after the polls closed. However, a detailed heatmap analysis comparing content publication dates with moderation dates found no distinct pattern of mass moderation of election-period content after the fact, and the researchers rejected that hypothesis.
Change point detection did identify dates when moderation patterns shifted, but these aligned with other world events rather than the elections. A recurring change point on 29 June, detected across five platforms, coincided with the EU adopting new sanctions against Belarus. Another on 19 July, detected for Facebook, Instagram and LinkedIn, coincided with the global IT outage caused by a faulty CrowdStrike update. Shifts in moderation delays clustered around the European Parliament’s vote on migration and asylum reforms in April and periods of intensified conflict in Gaza and the West Bank. The one striking election-linked exception was LinkedIn, which on 11 September 2024 showed a marked spike in moderating comments for election-related misinformation, explicitly labeled under the database’s category for negative effects on civic discourse or elections. That content had been posted in mid-July but was not moderated until roughly 45 days later, within the inter-electoral phase.
Most other moderation anomalies could not be tied to the elections at all. A YouTube peak on 30 September stemmed from fraudulent advertising abusing the ad network; a TikTok surge on 10 May involved mass account suspensions described in generic terms; a LinkedIn peak on 8 May targeted illegitimate job offers; and a Facebook spike on 29 September involved account actions under Meta’s terms of service with explanations too vague to interpret. The researchers attributed this persistent uncertainty to shallow reporting: mandatory fields filled with generic values and optional fields that could provide crucial context left severely underused.
The second part of the study reassessed the database’s reliability nearly a year after launch, comparing the first 100 days of submissions with the final 100 days of the observation period. Data quality problems persisted. TikTok, Facebook, Snapchat and LinkedIn submitted records indicating content had been moderated before it was even published, an impossibility that points to inadequate quality checks; TikTok alone submitted more than 12,000 such erroneous records on a single day, about 0.1 percent of its total for that date. Duplicate records with identical identifiers also appeared in Facebook and Pinterest submissions.
While all platforms consistently filled the required fields, ensuring formal compliance, optional fields were almost entirely ignored. TikTok, Instagram, Facebook and LinkedIn never populated any optional field, and the other platforms used only a few in a small fraction of records. Reliance on the generic category scope_of_platform_service, which covers a wide range of unrelated restrictions, barely changed, slipping from 42.68 percent of records in the initial period to 41.04 percent in the latest. X remained the most problematic reporter: it continued to report zero moderation delay while claiming all moderation is manual, and 99 percent of its recent records still reported purely manual moderation despite deepfakes, its primary moderation focus, being difficult to detect manually at scale. The differences between periods were statistically non-significant.
The authors are careful about what the null result means. The absence of observable changes in the database does not prove platforms took no action; they may have implemented preventive or internal measures that do not generate additional statements of reasons, and the database’s coarse predefined categories may mask real adaptations. Yet the findings sit uneasily beside the platforms’ own DSA risk assessment reports, which claimed election-specific measures such as increased monitoring and dedicated teams, and beside external analyses documenting persistent disinformation about electoral fraud and foreign interference during the campaign. The European Commission has separately opened formal proceedings against Facebook, Instagram and X over shortcomings in mitigating threats to civic discourse and electoral integrity related to the 2024 elections, shortcomings that were not visible in the database.
The study’s conclusion is that the Transparency Database currently functions as a mechanism of formal transparency rather than actionable transparency: platforms publish the required data, but regulators and researchers cannot effectively use it to scrutinize how moderation is actually carried out. The authors call for clearer reporting standards, stricter validation rules, a core set of high-value fields that must be completed for actions touching systemic risks such as electoral integrity, and more active regulatory use of the database to demand clarifications when reported trends contradict declared mitigation efforts. Ongoing European Commission work on researcher data access under Article 40 of the DSA, which would allow cross-checking self-reported records against non-public platform data, could address one of the study’s main limitations, the inability to link statements of reasons to the actual moderated content. Until then, the world’s most ambitious transparency experiment in content moderation reveals stability where adaptation was promised, and opacity where accountability was intended.
Subject of Research: Auditing platform content moderation and DSA Transparency Database data quality during the 2024 European Parliament elections
Article Title: When transparency falls short: auditing platform moderation during a high-stakes election
Article References: Tessa, B., Shahi, G. K., Trujillo, A., & Cresci, S. (2026). When transparency falls short: auditing platform moderation during a high-stakes election. EPJ Data Science, 15(1), Article 79. https://doi.org/10.1140/epjds/s13688-026-00704-8
Image Credits: AI Generated
DOI: 10.1140/epjds/s13688-026-00704-8
Keywords: content moderation, Digital Services Act, Transparency Database, 2024 European Parliament elections, social media platforms, platform governance, statements of reasons, data quality, regulation, disinformation, co-regulation, European Union
News Source: Denise Maddox. (October 29, 2026). Platforms showed little moderation change during EU elections, database audit finds. Scienmag.
