AI-generated Grokipedia articles are longer, less readable, and cite fewer sources than their Wikipedia counterparts

A recent study published in the Proceedings of the National Academy of Sciences provides evidence that automated encyclopedias differ from human-edited platforms in both structure and political leaning. The research suggests that rather than uniformly removing bias, these automated systems tend to favor longer, more complex narratives while introducing rightward shifts in certain topic areas. These findings raise questions about how artificial intelligence shapes public knowledge and source verification.

In October 2025, the American technology company xAI, founded by Elon Musk, launched Grokipedia. The platform was presented as the world’s first artificial intelligence-written encyclopedia. Musk promised the platform would fix left-leaning biases alleged to exist in the widely used online encyclopedia Wikipedia.

Wikipedia’s content is written and maintained by volunteer editors. Grokipedia generates and reviews its content using a large language model, which is a type of artificial intelligence trained on vast amounts of text to predict and generate human-like language. Visitors can suggest edits, but the automated system reviews and implements the changes without traditional human editorial oversight.

To evaluate these claims, researchers at Trinity College Dublin and Technological University Dublin conducted the largest academic analysis of Grokipedia since its launch. Scientists Saeedeh Mohammadi and Taha Yasseri set out to conduct a large-scale computational comparison to map structural and ideological differences objectively. They wanted to determine whether an automated platform could actually correct the perceived biases of human-edited websites.

The team analyzed articles on the same topic across Wikipedia and Grokipedia, using computational text analysis and machine learning methods. They focused on the 20,000 most-edited English Wikipedia pages, ensuring they were analyzing substantive articles by excluding lists and calendar dates. They then downloaded the corresponding 17,790 matching articles from the automated platform.

The authors extracted the main text from each article pair, stripping away menus, sidebars, and formatting code. They analyzed each pair for readability, vocabulary usage, and writing style. To measure reading difficulty, they used a standard formula that estimates the United States school grade level required to understand a text.

The team also calculated structural differences across the collected web pages by counting the exact number of references and hyperlinks present per one thousand words. To measure how closely the automated articles resembled their human-edited counterparts, they combined several similarity metrics into a single test. This score allowed them to directly compare the two versions of the same historical event or public figure.

The researchers found a profound split in how the automated system handled the existing content. While many Grokipedia articles closely mirror Wikipedia, a substantial proportion of the analyzed articles are more extensively rewritten. Roughly 66 percent of the entries diverge markedly in style, sourcing, and political leaning.

These rewritten Grokipedia entries were substantially longer than the Wikipedia versions, and the automated text proved much more difficult to read. On average, Grokipedia articles required a reading comprehension level of 14.5, compared to Wikipedia’s 10.7. The artificial intelligence platform also featured far fewer citations to back up its claims, providing an average of 20 references per one thousand words compared to Wikipedia’s 35.

To assess political bias, the scientists analyzed the external websites cited as references in the articles. They mapped these hyperlinks to an established dataset that assigns political leanings to news media sources based on social media sharing patterns. As a whole, articles on Grokipedia show a similar political leaning to those on Wikipedia, drawing predominantly on left-leaning news sources.

However, the scientists discovered noticeable changes within the highly rewritten group of articles. When it comes to politically and culturally sensitive topics, such as religion, history, literature, and art, Grokipedia shows a consistent shift toward referencing more right-leaning news sources compared to Wikipedia. The findings suggest the automated system provides a localized, topic-specific adjustment rather than a complete overhaul of political bias.

“Rather than systematically ‘correcting’ Wikipedia’s alleged biases, as claimed when first launched, our findings suggest that AI-generated encyclopedias such as Grokipedia selectively reshape existing knowledge,” said Taha Yasseri, director of the joint Centre for Sociology of Humans and Machines at Trinity College Dublin and Technological University Dublin, and principal investigator of the study. “This creates a patchwork system in which some content is copied, while other content is reinterpreted in ways that are less transparent and harder to scrutinize.”

The researchers worry about the long-term impact of relying on automated knowledge generation. “Online encyclopedias are central to public knowledge,” said Saeedeh Mohammadi, lead author of the study and a doctoral candidate at the Centre for Sociology of Humans and Machines and Research Ireland’s Centre for Research Training in Foundations of Data Science. “They are also being used to train future generations of large language models.”

“Our findings raise important questions about how public knowledge is produced, reproduced, verified, and governed,” Mohammadi added. “Unlike Wikipedia, where biases are visible and contested through human editing, AI-generated systems operate largely opaquely. This means shifts in perspective or sourcing may occur without clear accountability or editorial oversight.”

“Simply put AI generation does not remove bias – it changes how and where bias enters the system, often making it less visible.”

While the research provides useful evidence of emerging differences between AI-generated and human-edited encyclopedic knowledge systems, the researchers acknowledge that focusing on Wikipedia’s most-edited English-language pages likely overrepresents high-profile and contentious topics.

The automated similarity metrics also assess textual form and vocabulary alignment, but they cannot verify factual accuracy or detect hallucinated claims. The opaque nature of the automated platform’s training data limits the ability to determine exactly why these specific ideological shifts occur. Yasseri noted that these findings point to broader societal concerns.

“There is a dire need for transparency, oversight, and regulation in this space,” Yasseri said. “Our information landscape is changing rapidly. We have already seen how the lack of editorial responsibility on social media platforms has enabled the generation and circulation of misinformation and disinformation, often with catastrophic consequences for elections, public health, and social stability.”

“Now, we are witnessing the large-scale, black-box regeneration of information by large language models that remain largely closed to public scrutiny,” Yasseri continued. In future research, scientists might explore the underlying mechanics of how these models select information to cite and display. They could also study the inclusion of academic and governmental sources to build a more comprehensive picture of automated knowledge generation.

The study, “Selective divergence between Grokipedia and Wikipedia articles,” was authored by Saeedeh Mohammadi and Taha Yasseri.