ChatGPT cites Grokipedia: OpenAI’s AI uses Musk’s

When I discovered that ChatGPT was now citing Grokipedia, I thought Elon Musk had made another big move. The Guardian has just revealed a phenomenon as surprising as it is worrying: OpenAI’s artificial intelligence now relies on the 100% AI encyclopedia created by xAI, Elon Musk’s company.

The irony of the situation? OpenAI cites its direct competitor, led by the former co-founder who has become a fierce rival. Even more troubling: Grokipedia has generated over 6 million fully automated articles, and tests conducted by The Guardian detected 9 citations from this source in ChatGPT’s responses.

This case raises fundamental questions about the reliability of sources used by AI, the commercial war raging in this sector, and especially the dangerous circularity of information when one AI cites another AI. Let’s break down together what that really means.

Grokipedia discreetly infiltrates ChatGPT’s responses

The Guardian’s discovery is no coincidence. Journalists conducted methodical tests by asking ChatGPT questions on various topics, from the most mainstream to the most obscure.

Result: 9 citations from Grokipedia appeared in the responses provided by OpenAI’s AI. But beware, the observed pattern is revealing: these citations mainly concern niche or little-known historical topics, not general public queries.

Specifically, if you ask ChatGPT “Who is Elon Musk?”, you won’t get a reference to Grokipedia. However, for specific questions about obscure historical events or little-known personalities, xAI’s encyclopedia makes an appearance.

The practical implication is simple but troubling: the average user notices absolutely nothing. The source is indeed there, integrated into the response, but without anyone questioning its source reliability. It’s subtle, but it’s there, and that’s precisely what the problem is.

Grokipedia, the AI encyclopedia that challenges Wikipedia

Rapid growth

Launched in October 2024 by xAI, Grokipedia shows growth that defies understanding. In just a few months, the encyclopedia has generated 6 million articles, all created by artificial intelligence.

To put this figure into perspective: this represents over 80% of the volume of English Wikipedia, which took over twenty years to build its knowledge base. No human intervention, no manual verification, just Grok generating content continuously.

A fundamentally different model

The difference between Wikipedia and Grokipedia is not limited to volume. It’s the entire philosophy that radically diverges.

Wikipedia relies on human moderation, collaborative editing, and permanent cross-checks. Thousands of contributors debate, correct, and source every piece of information.

Grokipedia? It’s 100% automated. Automatic generation by Grok, the AI developed by xAI, without human moderation behind it. xAI even claims a “bias-free” encyclopedia, a claim I strongly qualify.

Spoiler: a bias-free AI doesn’t really exist. Each model inherits biases present in its training data. The central question remains: can we truly trust content never reviewed by a human? The questionable reliability of this model legitimately raises concerns.

Claude and other AIs are also affected

The phenomenon extends far beyond OpenAI. Claude (Anthropic), the other major competitor in the LLM universe, also cites Grokipedia in its responses. This is therefore not an isolated case, but a fundamental trend.

The technical challenge is simple to understand: large language models must diversify their training sources to improve their performance. Source diversification becomes a necessity, and Grokipedia, with its massive volume, establishes itself as an essential “public” source.

When faced with questions, radio silence from Anthropic, while OpenAI offers an evasive response: “we use a wide range of public sources available on the web.” Technically correct, but it carefully avoids the core debate.

The real risk? Data circularity. One AI citing another AI, which itself draws inspiration from web data potentially generated by AI. We are entering a loop whose implications for LLM training no one truly understands. This is a real fundamental problem for the AI ecosystem.

The source war in Elon Musk’s universe

xAI’s response to the criticism was swift, and it is straightforward: “Legacy media lies” (traditional media lies). A communication strategy that perfectly aligns with Elon Musk’s vision.

Let’s recall the historical context: Elon Musk is a co-founder of OpenAI, which he left in 2018 to become its fiercest rival. Since then, he has multiplied attacks against the company and its CEO Sam Altman.

His strategy becomes clear: to create his own end-to-end information ecosystem. X (formerly Twitter) for dissemination, Grok for artificial intelligence, Grokipedia for knowledge. Data control from the beginning to the end of the chain.

This approach also recalls the one deployed with the Powerwall and the Tesla energy ecosystem. Musk applies the same logic as with his cars: total vertical integration, control from A to Z.

The ultimate irony? OpenAI ends up using the data of its competitor and former co-founder. A situation that raises a fascinating question: are we heading towards a balkanization of information sources according to tech ecosystems? Each giant would create its own universe of knowledge, with its own rules and its own biases.

Let’s recap: ChatGPT cites Grokipedia, it’s confirmed, documented, and it also concerns Claude. The concrete implications for us, users? It’s becoming increasingly difficult to know what source truly lies behind the answers we get.

My practical advice remains simple: always verify critical information with multiple independent sources. Never take an AI’s answer at face value, especially when it may come from unverified sources generated automatically.

This is just the beginning. The debate on the reliability of artificial intelligence and the transparency of its sources is just beginning. Meanwhile, Elon Musk must be having a good laugh seeing his old company cite his new creation.