Failed comprehensiveness, successful minimalism: Wikipedia’s 3-year struggle to govern AI-generated content (2022–2025)

Abstract This article examines how the English Wikipedia community governed the emergence of large language model (LLM)-generated content between 2022 and 2025, tracing a protracted policy process that moved from early recognition of risk to repeated institutional failure and, ultimately, to narrow but enforceable rule formation. Drawing on process-tracing of publicly accessible revision histories, talk page deliberations, and Requests for Comment (RfCs), this study reconstructs how two comprehensive policy proposals failed to achieve consensus despite widespread agreement that AI-generated content posed a serious threat to knowledge integrity. In their place, the community eventually adopted two minimal interventions: a speedy deletion criterion targeting un-reviewed AI-generated pages and a one-sentence guideline prohibiting the generation of new articles from scratch. This article argues that this trajectory reveals a structural limitation in consensus-based commons governance, which I conceptualize as the comprehensiveness trap : the tendency for attempts at anticipatory, all-encompassing regulation to generate too many points of disagreement to secure adoption. In contrast, governance succeeds when it proceeds through deliberately incomplete, narrowly scoped rules that address observable harms while deferring unresolved normative questions. This shift is theorized as a form of minimalist constitutionalism emerging within a volunteer-driven institutional environment. Beyond its empirical contribution—the first full process-level account of Wikipedia’s AI governance—the article situates these findings within broader debates in commons and platform governance. It shows how generative AI disrupts the temporal logic of Wikipedia’s policy formation, which typically codifies stable practice rather than anticipating technological change, and destabilizes the boundary between infrastructure and editorial control that underpins the platform’s governance model. The analysis further highlights an unresolved tension between restrictive AI governance aimed at protecting epistemic integrity and more enabling approaches that could expand participation in historically marginalized knowledge domains.