Wikimedia research methodologies: a bibliometric history of peer production and AI (2005–2025)

Wikipedia is currently facing multiple crises. LLM-enabled chatbots have normalized the answer-based web in ways that decrease human readership of the encyclopedia while also increasing the burden of the community to adapt to increasing amounts of AI-generated content. The typical answer to emergent issues on Wikipedia has been to use big data analysis and semantic extraction to understand these controversies. But big data analysis is ill-equipped to deal with these crises on their own as it struggles to address “context,” a phenomenon that is better suited for humanities-oriented research. However, the fragmented nature of Wikimedia research limits such cross-disciplinary communication. To understand the nature of this fragmentation (and to assist in its repair), this article conducts a bibliometric and reflexive thematic analysis of two decades of the most popularly cited research about Wikimedia projects. In doing so, this article surfaces seven matters of concern and numerous methods that animate this “fractionated trading zone.” Furthermore, this science mapping illustrates how industry-funded research popularized the idea that Wikipedia existed as a semantic resource to be mined. In turn, this legitimized using community content as a training ground for corporate experiments in artificial intelligence. By paying attention to popularly cited research, this article comes to two conclusions. The first is that there are a set of concerns surrounding social media platforms, authority, and reliability that can serve to mediate disconnected research across the zone. The second conclusion is that Wikipedia’s relationship with researchers is a recursive one, since disciplinary choices have changed the cultural meaning and the sociotechnical function of this encyclopedic community.