Meta has revealed the primary of its quarterly integrity reports since Mark Zuckerberg walked again the corporate’s hate speech insurance policies and adjusted its method to content material moderation earlier this yr. In line with the stories, Fb noticed an uptick in violent content material, bullying and harassment regardless of an total lower within the quantity of content material taken down by Meta.
The stories are the primary time Meta has shared information about how Zuckerberg’s resolution to upend Meta’s insurance policies have performed out on the platform utilized by billions of individuals. Notably, the corporate is spinning the modifications as a victory, saying that it decreased its errors by half whereas the general prevalence of content material breaking its guidelines “largely remained unchanged for many downside areas.”
There are two notable exceptions, nevertheless. Violent and graphic content material elevated from 0.06%-0.07% on the finish of 2024 to .09% within the first quarter of 2025. Meta attributed the uptick to “a rise in sharing of violating content material” in addition to its personal makes an attempt to “scale back enforcement errors.” Meta additionally noticed a famous improve within the prevalence of bullying and harassment on Fb, which elevated from 0.06-0.07% on the finish of 2024 to 0.07-0.08% initially of 2025. Meta says this was attributable to an unspecified “spike” in violations in March. (Notably, this can be a separate class from the corporate’s hate speech insurance policies, which were re-written to permit posts concentrating on immigrants and LGBTQ individuals.)
These could sound like comparatively tiny percentages, however even small will increase could be noticeable for a platform like Fb that sees billions of posts on daily basis. (Meta describes its prevalence metric as an estimate of how typically rule-breaking content material seems on its platform.)
The report additionally underscores simply how a lot much less content material Meta is taking down total because it moved away from proactive enforcement of all however its most critical insurance policies like baby exploitation and terrorist content material. Meta’s report exhibits a big lower within the quantity of Fb posts eliminated for hateful content material, for instance, with simply 3.4 million items of content material “actioned” below the coverage, the corporate’s lowest determine since 2018. Spam removals additionally dropped precipitously from 730 million on the finish of 2024 to only 366 million initially of 2025. The variety of faux accounts eliminated additionally declined notably on Fb from 1.4 billion to 1 billion (Meta does not present stats round faux account removals on Instagram.)
On the similar time, Meta claims it is making far fewer content material moderation errors, which was one among Zuckerberg’s main justifications for his resolution to finish proactive moderation.”We noticed a roughly 50% discount in enforcement errors on our platforms in the USA from This fall 2024 to Q1 2025,” the corporate wrote in an replace to its January put up announcing its coverage modifications. Meta did not clarify the way it calculated that determine, however mentioned future stories would “embrace metrics on our errors so that individuals can observe our progress.”
Meta is acknowledging, nevertheless, that there’s a minimum of one group the place some proactive moderation continues to be vital: teenagers. “On the similar time, we stay dedicated to making sure teenagers on our platforms are having the most secure expertise attainable,” the corporate wrote. “That’s why, for teenagers, we’ll additionally proceed to proactively cover different kinds of dangerous content material, like bullying.” Meta has been rolling out “teen accounts” for the final a number of months, which ought to make it simpler to filter content material particularly for youthful customers.
The corporate additionally provided an replace on the way it’s utilizing giant language fashions to help in its content material moderation efforts. “Upon additional testing, we’re starting to see LLMs working past that of human efficiency for choose coverage areas,” Meta writes. “We’re additionally utilizing LLMs to take away content material from overview queues in sure circumstances after we’re extremely assured it doesn’t violate our insurance policies.”
The opposite main part to Zuckerberg’s coverage modifications was an finish of Meta’s fact-checking partnerships in the USA. The corporate started rolling out its own version of Neighborhood Notes to Fb, Instagram and Threads earlier this yr, and has since expanded the trouble to Reels and Threads replies. Meta did not supply any perception into how efficient its new crowd-sourced method to fact-checking is perhaps or how typically notes are showing on its platform, although it promised updates within the coming months.
