Wikipedia is not only a reliable source of information for many people in everyday life; it also forms one of the key foundations for training Large Language Models (LLMs). Its content is often systematically given greater weight in order to provide models with broad knowledge. In a sense, Wikipedia therefore trains AI systems – and is now being overtaken by them: while visits to the online encyclopaedia are declining, ChatGPT now records significantly more monthly traffic.
For Wikipedia, this trend creates a double challenge. On the one hand, falling visitor numbers could be accompanied by a decline in content quality. On the other, AI-generated text is creating new problems for the platform. Articles produced with AI often show significantly lower quality and less diversity in their sources. To counter this development, the WikiProject “AI Cleanup” was created to identify and remove AI-generated content from Wikipedia.
The BR article therefore raises a broader question: how will knowledge and the way we access it change if the quality of Wikipedia content begins to erode – and with it the training data used by the very AI systems that, in turn, draw on this material?
By Ida Kandler
Source: BR24, 5 October 2025, by Gregor Schmalzried and Christian Schiffer
