The usage of Large Language Models to edit Wikipedia is a constant point of discussion and leave uncertainties for new and seasoned editors alike. In their workshop on 30 September 2025 in the Universitätsbibliothek Zürich, Wikimedia CH Innovation Manager Shana Marinitsch and Wikipedian Diego Hättenschwiler shared insights into the inner workings and reliabilities of GenAI-models, the current state of discussion especially in the German- and English Wikipedia and the Wikimedia Foundation’s strategy.
A central part of the workshop was to play with LLM-models as a proofreader and suggestion-maker, to test out how those models can hold up not in generating content but aiding in the assessment of articles and to check inactive links and references. The result? While especially the leading models can sometimes help prompting editors towards incomplete statements, the participants also detected a lot of circular logic and referencing, given the prominent placement of Wikipedia and its sister projects in the training of these models.
Read more
