Platforms, Not Flaggers: Who Actually Removes Content in the EU

Public debate about online censorship keeps pointing at the same culprits: content moderators and the EU's certified “trusted flaggers.” The story goes that a network of NGOs and reviewers decides what gets taken down. We put that story to the test against the only authoritative evidence available — the platforms’ own transparency reports, mandated under the EU Digital Services Act (DSA). The reports contradict the narrative by roughly five orders of magnitude.

Key findings

• The platforms remove almost everything themselves. Across TikTok, YouTube, X, Facebook and Instagram, providers’ own automated detection accounts for 97.2%–99.9% of all moderation actions.
• Trusted flaggers are a rounding error. In their latest reports the five surfaces logged 1,031 trusted-flagger notices against about 377 million own-initiative actions — roughly one flagger notice for every 366,000 platform actions.
• Even ordinary user reports are minor. The ~2.7 million user notices are only ~0.7% of what the platforms action on their own, and only a fraction of those notices lead to a removal.
• Government orders are negligible. X received exactly one removal order from a public authority in its latest three-month period; the others are in the low hundreds to low thousands.
• Watch the labels. Platforms book almost all proactive removals as “terms & conditions” enforcement rather than “illegal content,” and their headline numbers are inflated by spam. Neither practice changes the conclusion.

What the DSA actually measures

Under Regulation (EU) 2022/2065, every Very Large Online Platform must report how its content moderation was triggered, broken down by source:

• Own initiative — the platform proactively detects and acts, almost always through automated systems (Art. 15).
• User notices — reports from ordinary users through the notice-and-action mechanism (Art. 16).
• Trusted flaggers — notices from entities formally certified by a national regulator, which platforms must prioritise (Art. 22).
• Authority orders — legally binding removal orders from member-state authorities (Art. 9).

The “moderators and flaggers run the show” claim predicts that the notice and trusted-flagger channels should carry a large share of the volume. Here is what the reports show instead.

Who removes content: the latest reports

The pattern is unanimous. The platforms’ own systems do essentially all of the work; trusted flaggers — the entities the censorship narrative points at — are statistically invisible.

The trend: flagger notices are growing from nothing to almost-nothing

Trusted-flagger usage is rising on some platforms, but from a negligible base and to volumes that remain trivially small next to automated removal.

Meanwhile the own-initiative channel is overwhelmingly automated: 94% of TikTok’s measures and 98.7% of X’s actions were taken by machines. The human moderator teams are small — roughly 1,350 at X, 3,674 at TikTok, over 8,600 at YouTube, and ~12,500 across Facebook and Instagram — and mostly review the automated queue and appeals rather than originate removals.

Where platforms get creative with labels

We looked specifically for categorisation choices that could distort the picture. We found four, and none of them rescues the “flaggers drive removals” claim:

1. “Terms & conditions,” not “illegal content.” Platforms book the vast majority of proactive removals under their own community guidelines rather than the DSA’s illegal-content basis. TikTok reports zero own-initiative actions on the illegal-content basis and routes all ~259 million through terms & conditions.
2. Spam inflates the totals. 98.7% of X’s own-initiative actions are automated spam-account suspensions, and spam is ~46% of Facebook’s. Yet even after stripping spam out, X’s remaining ~371,000 content actions still outnumber its 185 trusted-flagger notices by about 2,000 to 1.
3. Different things counted as one. “Actions” count individual pieces of content or accounts (a single YouTube channel termination cascades to ~9.7 million removals), while “notices” count inbound reports. The honest comparison is source-of-moderation, which still overwhelmingly favours the platform.
4. Shifting definitions and data bugs. Metric definitions change between editions, and because flagger numbers are so small, minor data-quality defects move them materially — Meta excluded 339 trusted-flagger reports over an attribution bug and admitted mislabeling flagger submissions to the EU database.

Bottom line

The claim that trusted flaggers or moderators are a major source of content removal is not supported by the platforms’ own DSA disclosures — it is the reverse of what they report. Automated detection by the platforms themselves accounts for essentially all moderation volume; user notices are a distant second; and trusted-flagger and government-order channels are, quantitatively, rounding errors. The finding holds even after correcting for the reports’ most significant labeling quirks.

Methodology & sources

We used the most recent DSA transparency report for each platform as of July 2026. TikTok publishes the standardised EU CSV template (read directly from the notice, own-initiative, automated-means and human-resources files). Google and Meta publish machine-generated PDFs with a real text layer, extracted table-by-table. X embeds its data as HTML tables with no summary totals, so per-country columns were summed programmatically. All figures are raw integers transcribed from the primary documents; trusted-flagger counts were cross-checked against each report’s narrative text. Reporting periods are not perfectly aligned (TikTok’s latest edition runs one half-year ahead of the others; X’s latest window is three months), so own-initiative trends should be read with that caveat. All conclusions concern the source of moderation, not its correctness.

References

1. European Parliament and Council. Regulation (EU) 2022/2065 (Digital Services Act), Arts. 9, 15, 16, 22, 24. OJ L 277, 27 Oct 2022.
2. TikTok Technology Limited. DSA Transparency Report — July to December 2025 (CSV dataset). tiktok.com/transparency (accessed 7 Jul 2026).
3. TikTok Technology Limited. DSA Transparency Report — January to June 2025 (PDF).
4. TikTok Technology Limited. DSA Transparency Report — July to December 2024 (PDF).
5. TikTok Technology Limited. DSA Transparency Report — January to June 2024 (PDF).
6. Google LLC. VLOSE/VLOP Transparency Report under the EU DSA — 1 January to 30 June 2025 (report-27, YouTube). Tables 3.1.1.h, 2.1.1, 2.2.1, 1.1.1, §3.3.
7. Google LLC. VLOSE/VLOP Transparency Report under the EU DSA — 1 July to 31 December 2024 (report-27, YouTube).
8. X Internet Unlimited Company. DSA Transparency Report — October 2025 (period 1 Apr–30 Jun 2025). transparency.x.com (accessed 7 Jul 2026).
9. X Internet Unlimited Company. DSA Transparency Report — April 2025 (period 1 Oct 2024–31 Mar 2025).
10. X Internet Unlimited Company. DSA Transparency Report — October 2024 (period 1 Apr–30 Sep 2024).
11. X Internet Unlimited Company. DSA Transparency Report — April 2024 (period 21 Oct 2023–31 Mar 2024).
12. Meta Platforms Ireland Ltd. Facebook — EU DSA Transparency Report (V5), 1 Jan–30 Jun 2025. Tables 15.1.c.(1), 15.1.a.(1), 15.1.b, 42.2.a.
13. Meta Platforms Ireland Ltd. Instagram — EU DSA Transparency Report (V5), 1 Jan–30 Jun 2025.
14. European Commission. DSA Transparency Database. transparency.dsa.ec.europa.eu.