Polarizing knowledge?

Background

On October 27, 2025 xAI launched “Grokipedia,” an AI-powered online encyclopedia and alternative to the most widely used source of peer-produced knowledge, Wikipedia. According to Elon Musk, founder and CEO of xAI, Grokipedia’s seeks “to purge out the propaganda” allegedly prevalent on Wikipedia. Technically, Grokipedia is described as a fork of Wikipedia whose content is automatically altered through xAI’s large language model (LLM), Grok. It sources and integrates web-accessible content through processes that remain largely opaque. Since its launch, journalistic coverage has raised concerns about factual inaccuracies, right-wing biases, and unattributed reproduction of Wikipedia content. Systematic investigations of these issues are missing so far.

Motivation

Neutrality and accuracy are foundational norms of encyclopedias and are explicitly institutionalized within Wikipedia’s principles. Although Wikipedia has been criticized for the misrepresentation of certain social groups, its collaborative production model is generally regarded as a mechanism that mitigates systematic bias. By contrast, the use of an LLM such as Grok, which has been reported to produce factually inaccurate and even problematic content, to construct an encyclopedia raises substantial epistemic concerns. The move by xAI to publish an alternative, explicitly right-leaning knowledge platform takes shape in the midst of fundamental challenges faced by liberal democracies and democratically organized digital knowledge infrastructures. Against this backdrop, an empirical evaluation of the political orientation, representational patterns, and potential biases of both knowledge databases appears urgent.

Objectives

This project advances our understanding of how AI-powered platforms may reshape public information environments. We aim to systematically assess the degree of content overlap between the online encyclopedias Wikipedia and Grokipedia and to compare their political leanings, measured by the visibility and evaluation of political actors. Methodologically, the study combines approaches from political science and computer science, integrating conceptual frameworks on bias and representation with computational techniques for large-scale text analysis.