Grok is now editing itself on Grokipedia.
Wikipedia is, at its heart, a massive group project. It’s edited by almost a million people each year. Grokipedia, Elon Musk’s AI-generated Wikipedia alternative, is, by contrast, only edited by a chatbot, Grok. The site, launched in 2025, does allow humans to suggest changes to its content. But according to a new analysis from the Tow Center, something has shifted: suggestions are increasingly being submitted from Grok to itself. Now the chatbot is primarily suggesting and approving edits to Grokipedia, without humans involved. Grok-supplied edits spiked in December and have overtaken human submitters, making up more than three-quarters of the suggestions.
At the same time that Grokipedia is editing itself more, people are increasingly relying on the site for information. A simple ChatGPT question can result in information from Grokipedia, and the site often appears in Google search results. But there are risks to relying on it. On January 31, one day after the latest batch of files concerning Jeffrey Epstein was released, Grok suggested adding that Musk had never been mentioned in the Epstein files to his entry on Grokipedia. (Musk was mentioned multiple times in the most recent batch.)
“Alone, [Grok] is not obviously something that I would trust with any type of fact-checking,” said Alexios Mantzarlis, a cofounder of Indicator.media and the director of the Security, Trust, and Safety initiative at Cornell Tech.
On Wikipedia, an army of humans provide clarification and updates, in dialogue with one another. On Grokipedia, edits initiated by “Grok Editor” seem to draw from trending topics or daily news on X. Grok does engage in a decision-making dialogue—with itself. Its choices can sometimes look like editorial restraint. We found that Grok often rejects its own suggestions, only accepting them two-thirds of the time (it accepts human suggestions 70 percent of the time). For example, after Alexandria Ocasio-Cortez, the representative from New York, refused to do an interview with Jesse Watters, the Fox News host, because of his “sexually exploitative rhetoric,” the bot suggested adding that update to Ocasio-Cortez’s page. Next, it responded to its own suggestion and rejected the addition, saying the add would increase “minor interpersonal conflict over empirical substance” and “dilute focus on notable events with transient social media engagement.”
Representatives from xAI, Grok’s parent company, could not be reached for comment.
Musk created Grokipedia as an alternative to Wikipedia because, he has said, he believes the latter is an extension of legacy media propaganda. But Ryan McGrady, who studies internet platforms and their social implications at the Initiative for Digital Public Infrastructure at the University of Massachusetts Amherst, believes that Grokipedia is far from neutral. Many of Grok’s processes are still largely invisible, despite Musk’s claims that they should be open-source. It’s also unclear how much Musk can affect its decisions. “It almost seems like Musk is creating a vertically integrated, centrally controlled knowledge production system,” McGrady said.
Musk previously manipulated Grok’s responses on a case-by-case basis, according to an analysis by the New York Times, and Grokipedia has been under fire since its launch for peddling far-right talking points. “It’s a return to top-down knowledge,” McGrady said. “It’s unambiguously one person’s encyclopedia project.”