The latest ChatGPT model has begun using Grokipedia, Elon Musk's controversial AI encyclopedia, as a source of information. Tests revealed that the popular chatbot cites this source when answering questions about lesser-known topics, raising concerns about the spread of misinformation.
Grokipedia appears in ChatGPT responses
The Guardian conducted a series of tests in which the GPT-5.2 model cited Grokipedia nine times in response to more than a dozen different queries. Topics for which Grokipedia appeared as a source included political structures in Iran, such as the salaries of the Basij paramilitary force and ownership of the Mostazafan Foundation. ChatGPT also cited Grokipedia in response to questions about the biography of Sir Richard Evans, the British historian and expert witness in the court case against Holocaust denier David Irving.
Interestingly, ChatGPT did not cite Grokipedia when directly asked about topics on which the encyclopedia is known to spread false information. These include the January 6 events in the United States, alleged media bias against Donald Trump, and the HIV/AIDS epidemic. Instead, information from Grokipedia made its way into the model's responses to queries about lesser-known and specific topics.
Why is Grokipedia controversial?
Grokipedia was launched in October by Elon Musk's xAI as a competitor to Wikipedia. Unlike Wikipedia, it does not allow direct human edits. Instead, an AI model writes the content and responds to requests for changes. Musk has long criticized Wikipedia for alleged bias against conservatives. Shortly after its launch, journalists pointed out that while many articles had been copied directly from Wikipedia, Grokipedia also contained problematic content. For example, it claimed that pornography contributed to the AIDS crisis, offered "ideological justifications" for slavery, and used derogatory terms for transgender people.
When citing Grokipedia, ChatGPT repeated stronger claims than those found on Wikipedia. For example, regarding the Iranian company MTN-Irancell, ChatGPT claimed that the company has ties to the office of Iran's supreme leader, a claim that cannot be found in that form on Wikipedia. ChatGPT also cited Grokipedia when repeating information that The Guardian had previously debunked. Specifically, it concerned details about Sir Richard Evans's work as an expert witness in David Irving's trial.
ChatGPT is not the only one using Grokipedia
GPT-5.2 is not the only large language model that apparently cites Grokipedia. According to anecdotal evidence, Anthropic's Claude chatbot has also referenced Musk's encyclopedia on topics ranging from oil production to Scottish beer.
An OpenAI spokesperson said that the model's web search "aims to draw from a broad range of publicly available sources and viewpoints." The spokesperson added that safety filters are applied to reduce the risk of displaying links associated with serious harm and that ChatGPT clearly shows which sources informed an answer through citations. The company also said that it has ongoing programs to filter out low-trust information and influence campaigns.
Concerns about "LLM grooming"
The fact that information from Grokipedia is making its way into the responses of large language models is worrying misinformation researchers. Last spring, security experts expressed concerns that malicious actors, including Russian propaganda networks, were producing vast amounts of misinformation in an effort to fill AI models with lies. This process is known as "LLM grooming." In June, concerns were raised in the US Congress that Google Gemini was repeating the Chinese government's position on human rights abuses in Xinjiang and China's COVID-19 policies.
Nina Jankowicz, a misinformation researcher who has worked on the issue of LLM grooming, said that ChatGPT's citation of Grokipedia raises similar concerns. Although Musk may not have intended to influence large language models, the Grokipedia entries that she and her colleagues reviewed "rely on sources that are unreliable and poorly substantiated at best, and deliberate misinformation at worst."
The fact that large language models cite sources such as Grokipedia or the Pravda network may, in turn, improve the credibility of those sources in readers' eyes. "They might say, 'Oh, ChatGPT cites it, these models cite it, it must be a decent source, they must have vetted it'—and they might go there to look for news about Ukraine," Jankowicz said.
Once bad information makes its way into an AI chatbot, it can be difficult to remove. Jankowicz recently discovered that a major news outlet had included a fabricated quote attributed to her in an article about misinformation. She wrote to the news outlet asking it to remove the quote and posted about the incident on social media. The news outlet removed the quote. However, AI models continued to cite it as hers for some time. "Most people will not do the work necessary to figure out where the truth actually lies," she said.
xAI's response
When asked for comment, a spokesperson for xAI, the owner of Grokipedia, replied: "Traditional media lies." An OpenAI spokesperson emphasized that ChatGPT clearly displays sources through citations, allowing users to assess the credibility of the information. The company is also continuing its efforts to filter out untrustworthy sources and influence campaigns.
This development highlights the growing problem of information quality in the artificial intelligence landscape. While AI models are becoming increasingly powerful and accessible, the question of the credibility of the sources they draw from remains crucial to their future use.
Additional sources: tomshardware.com and techcrunch.com
