What Are the Politics of a Platform? What the Data Says About Content Moderation on X

2026-04-08 · Source: Tech Policy Press · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, long

Summary

An analysis of content moderation on X (formerly Twitter) under Elon Musk's ownership reveals significant policy shifts and conflicting data regarding "hate speech" prevalence. Following the acquisition, X loosened content moderation, reinstated banned accounts, and removed policies on COVID-19 misinformation, crisis misinformation, and election integrity. While X claims a "freedom of speech, not freedom of reach" approach reduced hateful impressions, external studies using post counts and keyword analysis generally indicate an increase in "hate speech" production, at least temporarily. The discrepancy likely stems from different measurement methodologies: external researchers track post volume, while X focuses on user impressions, a metric only X can fully access. X's internal data, though limited and lacking methodology, suggests a reduction in hateful impressions, with some external analysis by Sprinklr supporting this for early 2023. However, the lack of consistent, verifiable data from X makes definitive conclusions challenging.

Key takeaway

For research scientists and policy makers evaluating platform content moderation, you should prioritize access to impression data, not just post volume, to accurately assess the impact of "freedom of reach" policies. The European Commission's Digital Services Act (DSA) enforcement against X for data access obligations highlights the critical need for regulatory pressure to ensure transparency and enable independent verification of platform claims regarding content moderation effectiveness.

Key insights

Measuring "hate speech" impact requires distinguishing between content production and user impressions, with only platforms having full impression data.

Principles

• Content moderation policy changes significantly alter platform content dynamics.
• External and internal data often conflict due to differing measurement definitions.
• Algorithmic demotion aims to reduce content reach without outright removal.

Method

External analyses of "hate speech" volume on X typically involve counting posts using keyword lists or classifiers, while internal platform data focuses on impressions (views) of content.

In practice

• Consider both content production and consumption metrics when assessing platform moderation.
• Recognize that "hate speech" definitions vary widely across studies and platforms.
• Be wary of claims without transparent methodology or verifiable data access.

Topics

• Content Moderation Policy
• X Platform Governance
• Hate Speech Measurement
• Algorithmic Suppression
• Data Transparency

Best for: CTO, Executive, AI Scientist, AI Ethicist, Policy Maker, Research Scientist

Related on AIssential

• What the X Fine Reveals About Data Access Under Article 40 of the Digital Services Act — The EU's DSA Article 40.12 mandates broad, free, and independent researcher access to public platform data.
• Chinese activist in UK told by X that abusive deepfakes do not breach rules — X's content moderation failed to protect a Chinese activist from deepfake abuse, raising concerns about platform accountability and protection for dissidents.
• The DSA’s Sleeping Giant is Waking Up — Out-of-court dispute settlement under DSA Article 21 is gaining traction, revealing systemic content moderation failures despite implementation challenges.
• Platforms are functioning as invisible filters, slowing the propagation of messages that do not align with Western strategic interests or that document the humanitarian costs of the intervention. — During the 2026 Iran-Israel-US conflict, social media platforms LinkedIn and TikTok reportedly engaged in "algorithmic suppression" or "shadowbanning" of content, de-amplifying narratives that did not…
• Mark Zuckerberg and his Ray-Ban entourage have their day in court — Meta's CEO defended platform design choices in court by prioritizing free expression over unproven harm.
• A stacking of mechanisms that quietly convert speech into a permissioned activity—filtered by platform policy, priced by quasi-legal services, and chilled by corporate litigation strategies. — Converging private and automated systems are creating a permissioned speech environment, risking functional authoritarianism.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Tech Policy Press.