So What If It Was AI Generated? It Conveys a Message": How X Community Notes Debates Norms Around Synthetic Media

Abstract

As AI-generated images and videos circulate at scale on social platforms, crowdsourced moderation systems face a question that is as normative as it is technical: should synthetic media be considered negatively? We examine this question through X’s Community Notes, analyzing 2,527 conflict notes on 1,219 tweets where contributors simultaneously argued content was misleading and that no label was needed. Using semantic clustering of LLM-extracted open codes, we identify 8 forensic correction strategies and 7 normative defense strategies, and three recurring normative tensions: audit-redundancy, expressive-forensic, and reverse-deepfake disputes. Although defense notes are almost never published, their presence is associated with a statistically significant drop in correction note publication rates. Our findings suggest that crowd moderation of AI-generated content involves not only a detection problem, but also norm negotiation, with important implications for the governance of synthetic media at scale.