What Grokipedia Says About Science
Wikipedia's value is in its shared, if imperfect, factual reality, with edits tracked in public and editors who are, for all their biases, still human.
In my everyday work and personal life, I have started to more frequently see Grokipedia appearing in research results delivered by AI models not affiliated with Elon Musk's xAI, now SpaceXAI. Wikipedia has long been the dominant resource of this kind. Besides being one of the most visited websites on the internet, it sits at the top of Google's results for a vast range of queries. Then it was ingested by AI models, and now when we search, we often get AI-generated answers built on top of it.
Now, in some queries, the mediator is no longer Wikipedia but Grokipedia.
Wikipedia is, for all its problems — including ones we have written about ourselves — remarkably well-functioning. Despite documented biases, it has given us something close to a shared online reality, a coordination mechanism. Universities taught with it, and several generations of students have developed working intuitions about when and how much to trust it, and how to chase its citations. That kind of communal calibration takes time, and it does not transfer automatically to whatever replaces it.
Grokipedia is not the first attempt at this kind of cloning. An analog has already played out with Russian Wikipedia: a state-aligned clone launched in 2023, Ruwiki. At the time Ruwiki was built, the alterations required substantial human labor. Today, with LLMs, cloning and rewriting a Wikipedia at scale is trivial. Maybe because of that, Ruwiki closed editing to unverified users as of January 2026. Grokipedia itself launched in October 2025 with roughly 885,000 articles, many lifted directly from Wikipedia under its Creative Commons license, others rewritten by Grok.
Wikipedia is trying to defend itself from this kind of cheap alteration — in March 2026, its editors voted to prohibit LLM-generated or rewritten article content. But enforcement is unclear as AI's capacity to mimic human writing improves. Wikipedia's own human traffic has already fallen by roughly 8% year-on-year, which the Wikimedia Foundation attributes to AI summaries and chatbots answering directly from its content.
We may be heading toward a world of many forked, altered reference sources, with no shared frame for online facts.
What concerns us here is the representation of science. Here we briefly look at a few of Grokipedia's scientific articles.
Climate change. Wikipedia's article on climate change mentions misinformation, compares it with the tobacco industry and links out to a separate companion article — Climate change denial — where it names the actors. The denial movement, Wikipedia says, "originated from fossil fuel companies, industry groups, conservative think tanks, and contrarian scientists". On the science itself, Wikipedia reports a very strong consensus — over 99% in recent literature surveys — and notes that the public substantially underestimates this agreement.
Grokipedia's climate change article does not mention misinformation. Instead, extensive space is given to Bjorn Lomborg and the Copenhagen Consensus cost-benefit framing — net-zero costs running into the hundreds of trillions, low returns per dollar in avoided damages, and so on. On consensus, in one place it just states: “Some critiques question the methodology of consensus surveys.” without any reference. In another part of the article, it questions consensus again: "claims of near-unanimous scientific consensus... overstate agreement, owing to selective categorization in literature reviews… deeming the figure unreliable.[205]". Notably, reference [205] here is completely irrelevant. Well, say hello to a hallucinated reference.
Glyphosate. We wrote previously about corporate influence on research around glyphosate. Wikipedia's article structures its industry coverage around the Monsanto Papers: documents released through litigation showing that "Monsanto had planned a public relations effort to discredit the IARC report", ghostwritten op-eds under outside authors' names, and Bayer's $9.6 billion in settlements.
Grokipedia's glyphosate article treats Bayer in an entirely different mode.Innovation in alternatives section frames the company as an R&D funder: "Herbicide Innovation Partnership between Bayer Crop Science and the Grains Research and Development Corporation". Elsewhere in the same section, Bayer's collaboration with the USDA's GROW network on integrated non-chemical weed management appears alongside BASF, Corteva, and Syngenta as evidence of an industry-led research ecosystem.
Raw milk. Wikipedia reports the position of HHS, CDC, and other public health agencies plainly: they "strongly recommend that the public not consume raw milk". The agencies are cited as authoritative.
Grokipedia's raw milk article characterizes the same agencies in opposite terms: "agencies like the FDA and CDC, drawing from academia with documented left-leaning tendencies" (again, without a reference), echoing Musk’s stance on Wikipedia when he called it "Wokepedia".
These are surface examples. Across them we see how the two encyclopedias draw from overlapping pools of named actors, but the framing and emphasis function applied is different. A more in-depth study could reveal further peculiarities of Grokipedia. But Grokipedia and its successors may be a moving target — rewriting themselves in real time, tailoring their framings to cater to the worldviews of their creators, or even individual readers. Against that, Wikipedia's value is in its shared, if imperfect, factual reality, with edits tracked in public and editors who are, for all their biases, still human.
