What Are the Politics of a Platform? What the Data Says About Content Moderation on X
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
Editorial summary, takeaway, and curation by AIssential. Original article published by Tech Policy Press.