A new test reported by The Guardian suggests that ChatGPT is increasingly pulling information from Elon Musk’s Grokipedia, an online encyclopedia created by xAI. During the investigation, ChatGPT reportedly cited Grokipedia while answering questions on a range of subjects, including Iran’s political structures and biographical details about historical figures.
That trend is raising fresh concerns about the reliability of AI-generated answers, especially when an AI system leans on sources that don’t follow the same verification standards as traditional editorial outlets or large, community-moderated references. The core worry: if the underlying source material isn’t carefully vetted, the final response can look polished and authoritative while still being misleading, incomplete, or wrong.
Grokipedia was positioned as an alternative to Wikipedia, with Elon Musk framing it as a project designed to pursue truth and neutrality. He has criticized Wikipedia for what he describes as ideological bias. Critics, however, argue that Grokipedia can swing in the opposite direction, sometimes presenting politically right-leaning narratives or treating controversial topics in a one-sided way. Whether a reader agrees with those critiques or not, the debate highlights a broader issue: perceived neutrality is hard to guarantee when a platform’s editorial model is fundamentally different from a human-driven, community-reviewed knowledge base.
One of the biggest differences is how Grokipedia content is produced. Instead of relying primarily on human contributors and ongoing public edits, the encyclopedia’s entries are largely generated by xAI’s in-house model, Grok. User edits reportedly aren’t part of the process, and quality control is said to be handled internally by xAI employees. That structure can make updates and publishing faster, but it also concentrates decision-making and reduces the kind of broad, transparent peer review that many users associate with established reference sites.
The bigger technical concern comes when one AI uses another AI’s machine-written text as a “source of truth.” If models like the latest version of ChatGPT treat AI-generated encyclopedia entries as factual references, it can create a feedback loop where AI systems effectively train each other’s outputs through citations and repeated exposure. Over time, this can lead to a “garbage in, garbage out” scenario: weak claims get repeated, errors get reinforced, and biases can become harder to detect because they’re echoed across multiple systems.
The Guardian’s test suggests this problem isn’t theoretical. It reportedly found instances where ChatGPT cited Grokipedia while repeating claims that go beyond established knowledge or that have previously been disputed. When AI tools cite each other, misinformation can gain an illusion of credibility: it looks validated simply because it appears in multiple places, even if the original material was never properly fact-checked. For everyday users, that makes spotting false or slanted information even more difficult—especially when the response is delivered in a confident, conversational tone.
OpenAI responded to the report by emphasizing that ChatGPT’s web search is designed to reflect a broad mix of publicly available sources and viewpoints. The company also said it has safety filters aimed at reducing the risk of surfacing content with a high potential for harm. OpenAI further pointed to citations and sourcing as a transparency feature, and noted ongoing efforts intended to reduce reliance on low-credibility sources.
As AI search and AI-assisted writing become more common, this situation underscores a key lesson for anyone seeking accurate information: always pay attention to what’s being cited, compare multiple reputable sources, and remember that a well-written answer isn’t the same thing as a verified one.
