Meta's ad system blocked promotion for a Virginia Woolf play in Spain

Automated content moderation at scale is one of the stranger challenges facing digital platforms. When a social network has billions of users posting constantly, human reviewers alone cannot possibly evaluate every piece of content against a company's policies. The result is a hybrid system: machine learning algorithms screen the majority of content flagged for review, and human moderators handle edge cases or appeals. These systems are trained on examples of policy violations (hate speech, graphic violence, sexual content, spam) so they can recognize similar patterns in new posts. The logic sounds straightforward in theory. In practice, algorithms often struggle with context, nuance, and the difference between describing something and promoting it. The specific mechanics of how these systems block advertising differ from how they moderate user posts. Ad systems have financial stakes on both sides. Advertisers want their promotions to reach audiences, but platforms also want to avoid placing ads next to content that reflects badly on brands, a phenomenon called "brand safety." To manage this, platforms use keyword filtering, image recognition, and behavioral targeting rules. An ad promoting a theatrical production might be flagged if it contains language the algorithm associates with sexual content, drug use, violence, or other restricted topics. Unlike a user's personal post, an ad is explicitly promotional material, so the bar for rejection can be higher. An algorithm might reject an ad thinking it violates policy even when a human reviewer would instantly understand the context and approve it. Literary works pose a particular problem for automated systems because classic literature often contains exactly the kind of language, themes, and descriptions that modern content policies restrict. Virginia Woolf's modernist novels include stream of consciousness passages, fragmented narratives, sexual references, and psychological disturbance. Shakespeare's plays feature murder, sexual violence, incest, and profanity. If an algorithm is trained to catch explicit sexual language or descriptions of harm, it will flag these texts without understanding they are artistic works or historical artifacts. The system cannot distinguish between someone posting a passage from "Lolita" as literature versus someone using similar language for harmful purposes. This problem reveals a fundamental tension in how platforms approach moderation. Rule based systems are fast and consistent, but they are also inflexible. They cannot read, interpret, or judge intention the way a person can. A human moderator can see that an advertisement for a play is legitimate cultural promotion, not an attempt to distribute prohibited material. An algorithm sees keywords or patterns and applies a rule. As platforms have scaled moderation to handle billions of pieces of content daily, they have generally chosen speed and automation over nuance. The cost is occasional a