How ChatGPT’s Latest Model Relies on Elon Musk’s Grokipedia and Why It Matters

OpenAI’s newest ChatGPT iteration, GPT-5.2, has been found to cite Grokipedia, an AI-generated encyclopedia launched by Elon Musk, as a source for various queries. Tests by The Guardian revealed that GPT-5.2 referenced Grokipedia nine times across over a dozen questions, covering topics from Iranian political entities to biographies of historians involved in Holocaust denial trials. Grokipedia, introduced in October, distinguishes itself from Wikipedia by relying solely on AI-generated content without human editing, aiming to offer an alternative online knowledge base. However, it has faced criticism for promoting right-wing narratives on sensitive subjects such as gay marriage and the January 6 US Capitol insurrection.

Interestingly, ChatGPT did not cite Grokipedia when directly asked about some of the most controversial misinformation topics associated with the platform, such as the January 6 events or HIV/AIDS misinformation. Instead, Grokipedia’s influence surfaced in responses to more obscure or nuanced questions. For example, GPT-5.2 repeated claims from Grokipedia about the Iranian government’s connections to the telecommunications company MTN-Irancell, including assertions about ties to Iran’s supreme leader’s office that are stronger than those found on Wikipedia. Similarly, it echoed debunked details about historian Sir Richard Evans’ role as an expert witness in a libel trial against Holocaust denier David Irving.

This phenomenon is not unique to OpenAI’s model. Anecdotal evidence suggests that Anthropic’s Claude also references Grokipedia on diverse topics, from petroleum production to Scottish ales. OpenAI has stated that its web search feature aims to incorporate a broad spectrum of publicly available sources and viewpoints, applying safety filters to mitigate high-risk misinformation. They emphasize ongoing efforts to filter out low-credibility content and influence campaigns. Anthropic has not publicly commented on their use of Grokipedia. Nonetheless, disinformation experts express concern over the subtle integration of Grokipedia’s content into large language models (LLMs).

Experts highlight a growing threat termed “LLM grooming,” where malicious actors flood the internet with disinformation to seed AI training data with falsehoods. This tactic was previously noted with Russian propaganda and more recently with Google’s Gemini reportedly echoing Chinese government narratives on sensitive issues. Nina Jankowicz, a disinformation researcher, warns that Grokipedia’s reliance on untrustworthy or deliberately misleading sources could misinform AI outputs. Moreover, when LLMs cite Grokipedia, it may inadvertently boost the encyclopedia’s perceived credibility among users who assume AI models vet their sources thoroughly.

The persistence of misinformation in AI models poses significant challenges. Jankowicz recounts an incident where a fabricated quote attributed to her was published by a major news outlet and subsequently removed after her intervention. Despite this, AI models continued to cite the false quote for some time, illustrating how difficult it is to correct entrenched errors in AI knowledge bases. This underscores the broader issue that most users lack the resources or inclination to verify the accuracy of AI-generated information.

In response to these concerns, xAI, the company behind Grokipedia, dismissed criticism by labeling legacy media as dishonest. The situation highlights the complex interplay between AI-generated content, source credibility, and the risk of misinformation propagation. As AI language models increasingly influence public knowledge, ensuring the integrity of their source material remains a critical challenge for developers, researchers, and users alike.