From Wikipedia to AI Answers: Gender Gaps in the Knowledge Pipeline
AI search assistants are increasingly used as entry points to public knowledge and frequently rely on Wikipedia as a foundational information source. This 5+5 session introduces a study examining how gender disparities may propagate across the emerging Wikipedia–AI knowledge pipeline, and invites Wikimania attendees to engage with and reflect on the downstream consequences of Wikipedia content.
The study analyses English Wikipedia *Articles for Deletion* (AfD) discussions from January–March 2025 (2,317 biographies) to examine gender differences in article retention and referencing practices. These patterns are compared with outputs from two search-enabled AI systems (Google Gemini and Perplexity Sonar). Across both Wikipedia and AI outputs, the analysis evaluates coverage gaps and the composition and reliability of references using Wikipedia’s perennial sources framework.
The results reveal contrasting dynamics across the pipeline. Within AfD discussions in our sample, biographies about women are more likely to be retained than those about men, but they also exhibit stronger referencing practices, suggesting a higher evidentiary threshold for inclusion. In downstream AI systems, references tend to be of lower reliability overall, and outputs about women show weaker referencing practices than those about men. These findings suggest that asymmetries in upstream knowledge infrastructures can propagate through AI-mediated systems that increasingly shape public access to knowledge.
After a short talk, participants will take part in a brief activity querying an AI search tool and reflecting on response quality, apparent reliance on Wikipedia, and the trade-offs between reliance on Wikipedia and information voids where strong coverage is absent.
