Content moderation

Instagram’s February 2019 announcement described allowing content showing contemplation or admission of self-harm to help people find support, while considering the effects of those images on others7. Mosseri explained that responding to harassment involves not only taking down rule-breaking posts but also deciding which posts and comments appear prominently and using privacy settings to reduce young people’s exposure8. Age-based content settings provide another form of restriction, and Meta’s Newsroom said Instagram’s updated teen settings drew inspiration from public movie-rating guidelines and parent feedback910. The Motion Picture Association does not rate Instagram content or endorse those settings, and Instagram’s moderation systems are not a movie ratings board910. Copyright enforcement has its own tools and procedures: Meta’s Newsroom describes Rights Manager, specialized reporting tools, a team handling copyright complaints and a policy for repeat infringement on Facebook and Instagram11.

Instagram’s moderation process uses classifiers—automated systems that identify likely problems such as nudity or hate speech—which Mosseri said human reviewers then assess for most types of violations2. Mosseri described technology as useful for handling large volumes and people as better at understanding nuance, while acknowledging that even a small error rate across millions of daily reports leaves many mistakes25. Meta’s Transparency Center says Instagram’s comment system also applies filters to remove comments that may break its Community Guidelines12. Meta’s Newsroom reported in August 2022 that warning screens had prevented some potentially rule-breaking hate speech or bullying content from being posted across its platforms13. In February 2024, Meta’s Newsroom described Meta’s tests of large language models—AI systems trained on its Community Standards to assess possible violations—and reported early results suggesting better performance than its existing models14. The announcement said Meta’s use of generative AI for enforcement remained limited14.

Meta’s patent on testing content-review policies describes a proposed way to evaluate moderation rules before applying them broadly15. Meta’s patented approach compares people’s interactions with feeds containing content flagged under a rule against interactions with feeds containing only approved content15. In Meta’s patent, the comparison helps the online system decide whether to adjust that rule15.

Instagram’s Account Status identifies removed photos or videos and explains why content cannot appear in recommendations such as Explore; Instagram’s explainer and Mosseri describe options to remove the identified content or appeal, and Mosseri also describes editing it1316. For an unlabeled AI-generated profile excluded from recommendations, Instagram for Creators describes two paths back to eligibility: adding the AI-generated profile label or appealing the decision6. For aggregator accounts that have become ineligible for recommendations, Instagram for Creators says eligibility can return when most of their recently posted photos, carousels and reels are considered original in a rolling 30-day period17. That originality restriction does not affect showing posts to existing followers and is separate from intellectual property rules17.

Instagram enforcement appears in Meta’s Community Standards Enforcement Reports through measures such as content removals and the percentage of content acted on that Meta found before someone reported it18. Meta’s Newsroom reported in August 2021 that this proactive detection rate exceeded 90% in nine of the 11 Instagram policy areas covered, and that hate-speech removals across Facebook and Instagram had increased more than fifteenfold since reporting began18. Meta’s Newsroom has also acknowledged enforcement mistakes and reported metrics about appeals1819.