Meta has released its first quarterly integrity report since Mark Zuckerberg abandoned the company’s hate speech policy and changed its approach to content moderation earlier this year. According to the report, Facebook has seen an increase in violent content, bullying, and harassment, despite an overall decrease in the amount of content Meta removes.
The report marks the first time Meta has shared data on how Zuckerberg’s decision to roll back Meta’s policies has affected the platform used by billions of people. Notably, the company is presenting the changes as a victory, claiming that it has reduced the number of errors by half, while the overall prevalence of content that violates its rules “has remained largely unchanged for most problem areas.”
There are two notable exceptions, however. Violent and graphic content increased from 0.06%-0.07% in late 2024 to 0.09% in the first quarter of 2025. Meta attributes this increase to “an increase in the distribution of content that violates the rules” as well as its own efforts to “reduce enforcement errors.” Meta also saw a notable increase in the prevalence of bullying and harassment on Facebook, which rose from 0.06-0.07% in late 2024 to 0.07-0.08% in early 2025. Meta attributes this to an unspecified “surge” of violations in March. (Notably, this is a separate category from the company’s hate speech policy, which was rewritten to allow posts targeting immigrants and the LGBTQ community.)
That may seem like a relatively small percentage, but even a small increase can be significant for a platform like Facebook, which sees billions of posts every day. (Meta describes its prevalence rate as an estimate of how often content that violates its policies appears on its platform.)
The report also highlights how much less content Meta is removing overall, as it has moved away from proactively enforcing all but the most serious of its policies, such as child exploitation and terrorist content. Meta’s report shows a significant decline in the number of posts on Facebook removed for hateful content, with just 3.4 million pieces of content “reviewed” under the policy, for example, the company’s lowest number since 2018. Spam removals also fell sharply, from 730 million at the end of 2024 to just 366 million at the beginning of 2025. The number of fake accounts removed also fell significantly on Facebook, from 1.4 billion to 1 billion (Meta does not provide statistics on fake account removals on Instagram).
At the same time, Meta claims to be making far fewer errors in moderating content, which was one of the main reasons Zuckerberg decided to end proactive moderation. “We saw an approximately 50% reduction in enforcement errors on our platforms in the United States from Q4 2024 to Q1 2025,” the company wrote in an update to its January post announcing the policy changes. Meta did not explain how it calculated that figure, but said future reports “will include metrics on our errors so people can track our progress.”
However, Meta acknowledges that there is at least one group where some proactive moderation is still needed: teens. “At the same time, we remain committed to providing the safest possible experience for teens on our platforms,” the company wrote. “That’s why we will also continue to proactively hide other types of harmful content, such as bullying, for teens.” Meta has been rolling out “teen accounts” over the past few months, which should make it easier to filter content specifically for younger users.
The company also offered an update on how it uses big data models to help moderate content. “After further testing, we are starting to see LLMs performing beyond human performance for certain policy areas,” Meta wrote. “We also use LLMs to remove content from review queues in certain circumstances where we are confident it does not violate our policies.”
Another major component of Zuckerberg’s policy changes was the end of Meta’s fact-checking partnership in the United States. The company began rolling out its own version of Community Notes to Facebook, Instagram, and Threads earlier this year, and has since expanded that effort to include Reels and Threads replies. Meta didn’t provide any information on how effective its new crowdsourcing approach to fact-checking might be or how often the notes would appear on its platform, though it promised an update in the coming months.
