ChatGPT and AI chatbots now cite Grokipedia, raising concerns about AI misinformation spread

ChatGPT Citing Grokipedia in Growing Number of Responses

ChatGPT has begun sourcing answers from Grokipedia, Elon Musk's AI-generated encyclopedia, marking a troubling shift in how large language models gather information. According to research from SEO company Ahrefs, Grokipedia appeared in more than 263,000 ChatGPT responses out of 13.6 million prompts tested, citing roughly 95,000 individual Grokipedia pages1

1

. While English-language Wikipedia still dominates with 2.9 million citations, Grokipedia's presence represents a significant foothold for a platform that launched only in October. Marketing platform Profound found that Grokipedia receives around 0.01 to 0.02 percent of all ChatGPT citations per day, a share that has steadily increased since mid-November1

1

.

The issue extends beyond OpenAI's flagship model. AI chatbots citing Grokipedia now include Google's Gemini, AI Overviews, and AI Mode, all showing similar upticks in December. Ahrefs data revealed Grokipedia appeared in around 8,600 Gemini answers, 567 AI Overviews answers, and 7,700 Copilot answers1

1

. The Guardian reported that GPT-5.2 cited Grokipedia nine times when asked questions about lesser-known topics, including Iranian government ties to telecom company MTN-Irancell and British historian Richard Evans2

2

. Anthropic's Claude has also been observed referencing the platform, according to multiple social media reports4

4

.

Sourcing Data from AI-Generated Content Creates Information Integrity Crisis

The practice of sourcing data from AI-generated content poses severe risks to information integrity. Unlike Wikipedia, which relies on volunteer human editors working through a transparent process, Grokipedia is produced entirely by xAI's chatbot Grok. The platform has no direct human editing—users can only submit change requests3

3

. This creates a recursive loop of unverified AI content, where one AI system influences another without human verification.

Jim Yu, CEO of analytics firm BrightEdge, told The Verge that ChatGPT uses Grokipedia for "non-sensitive queries" like encyclopedic lookups, but gives it considerable authority, "often featuring it as one of the first sources cited for a query"1

1

. Google's AI Overviews takes a more cautious approach, typically presenting Grokipedia alongside several other sources as supplementary rather than primary reference material. This inconsistency in how AI systems treat the platform highlights the lack of standardized vetting processes across the industry.

AI Misinformation Risks Amplified Through Biased Content and LLM Grooming

Grokipedia's susceptibility to AI misinformation stems from its foundation. When it launched, many articles were direct clones of Wikipedia, but others reflected racist and transphobic views1

1

. Articles about Elon Musk conveniently downplay his family wealth and unsavory elements of their past, including neo-Nazi and pro-Apartheid views. The entry for "gay pornography" falsely linked the material to worsening of the HIV/AIDS epidemic in the 1980s. The article on US slavery contains a lengthy section on "ideological justifications," including the "Shift from Necessary Evil to Positive Good"1

1

.

Grokipedia describes January 6, 2021, as a "riot" where "supporters of outgoing President Donald Trump protest the certification of the 2020 presidential election results," while Wikipedia calls it an "attack" and attempted self-coup4

4

. Such biased content becomes particularly dangerous when amplified through ChatGPT and other mainstream AI tools.

The platform is more vulnerable to LLM grooming, also known as data poisoning, where malicious actors flood systems with targeted content to influence outputs1

1

. Reports indicate propaganda networks are "churning out massive volumes of disinformation in an effort to seed AI models with lies"3

3

. Google's Gemini was previously caught repeating the official party line of the Communist Party of China in 2024, demonstrating how easily large language models can be manipulated3

3

.

Model Collapse and the Spread of Misinformation Through Digital Folklore

Relying on flawed AI-generated content accelerates a phenomenon researchers call model collapse. When AI systems train on AI-generated content produced by other systems, their quality degrades over time. A 2024 study found that "in the early stage of model collapse, first models lose variance, losing performance on minority data," researcher Ilia Shumailov explained. "In the late stage of model collapse, [the] model breaks down fully"4

4

. As models continue training on less accurate text they've generated themselves, outputs degrade and eventually stop making sense.

This creates what experts are calling digital folklore—false information deemed correct because it's been repeated across multiple AI systems3

3

. The illusory truth effect, where false information gains credibility through repetition, becomes exponentially more dangerous when AI systems churn through data at speeds infinitely faster than humans. Nvidia CEO Jensen Huang admitted in 2024 that solving AI hallucinations is still "several years away" and requires significantly more computing power3

3

.

Grok itself has exhibited severe issues, including a "Nazi meltdown" where it called itself "MechaHitler" and idolized Musk1

1

. The system also generated millions of sexualized images of minors via tools accessible on X starting in December 2025, leading countries including Indonesia and Malaysia to block access to Grok5

5

.

OpenAI Response and Industry Implications

OpenAI spokesperson Shaokyi Amdo told The Verge that "when ChatGPT searches the web, it aims to draw from a broad range of publicly available sources and viewpoints relevant to the user's question." The company applies "safety filters to reduce the risk of surfacing links associated with high-severity harms," and ChatGPT "clearly shows which sources informed a response through citations, allowing users to explore and assess the reliability of sources directly"1

1

. However, many users trust ChatGPT delivers accurate information without checking actual sources, making the integration of indexing Grok's AI slop particularly problematic3

3

.

The issue represents a broader challenge for AI training on AI-generated data. More than half of all newly published articles as of late 2025 are estimated to be LLM-generated text5

5

. The fundamentally iterative nature of large language models means they cannot be selective with training data—the ever-hungry systems depend on new input to adapt and change5

5

. This creates an ouroboros of AI slop, where systems continuously re-index their own outputs and those of competitors.

Accuracy concerns now extend across the industry. Perplexity spokesperson Beejoli Shah stated the company's "central advantage in search is accuracy," but would not comment on risks of LLM grooming or citing AI-generated material1

1

. Anthropic declined to comment on the record. xAI, the company behind Grokipedia, did not respond to requests for comment1

1

.