What is it about?
This study compares nearly 18,000 articles from Wikipedia and Grokipedia, an AI-generated encyclopedia launched by xAI as an alternative to Wikipedia. We examined how similar the two platforms are in terms of writing style, structure, sourcing, and political orientation. We found that while many Grokipedia articles closely resemble their Wikipedia counterparts, a substantial number differ considerably. In particular, Grokipedia articles tend to be longer, more complex, and contain fewer references per word. The largest differences appeared in politically and culturally sensitive topics such as religion and history, where Grokipedia showed shifts in the political orientation of frequently cited media sources. The findings suggest that AI-generated encyclopedias may not simply reproduce existing knowledge systems, but selectively rewrite them, raising broader questions about transparency, bias, and accountability in AI-generated knowledge.
Featured Image
Photo by Declan Sun on Unsplash
Why is it important?
AI-generated knowledge systems are rapidly becoming part of how people access information online, yet we still know very little about how they differ from human-created platforms such as Wikipedia. This study provides one of the first large-scale empirical comparisons between Wikipedia and an AI-generated encyclopedia. The work is particularly timely because AI systems are increasingly used not only to answer questions, but also to generate and reshape public knowledge itself. Our findings show that AI-generated encyclopedias may introduce new forms of hidden bias and opacity, especially in politically and culturally sensitive topics, while lacking the transparent editorial processes through which biases in Wikipedia can be debated and contested. By examining nearly 18,000 matched articles, the study helps illuminate how AI systems may transform the future production, governance, and reliability of online knowledge.
Perspectives
What I find most interesting about this work is that it moves the discussion about AI bias away from chatbots and toward something potentially more consequential: AI-generated knowledge infrastructures. Wikipedia’s biases are visible, contested, and constantly negotiated by human communities. AI-generated encyclopedias, by contrast, can appear authoritative while obscuring how content is selected, rewritten, or framed. What surprised me most was not simply that differences emerged, but how selective they were. Many articles remained extremely close to Wikipedia, while others diverged substantially in style, sourcing, and political orientation. This suggests that AI systems may not replace existing knowledge uniformly, but instead reshape specific parts of it in uneven and difficult-to-detect ways. As AI-generated content becomes part of search engines, recommendation systems, and even future AI training data, understanding these dynamics is becoming increasingly important.
Taha Yasseri
University of Dublin Trinity College
Read the Original
This page is a summary of: Selective divergence between Grokipedia and Wikipedia articles, Proceedings of the National Academy of Sciences, May 2026, Proceedings of the National Academy of Sciences,
DOI: 10.1073/pnas.2603294123.
You can read the full text:
Resources
Contributors
The following have contributed to this page
