New Delhi: Grokipedia, the AI-generated encyclopedia launched by Elon Musk’s xAI last October, has not processed a single edit in more than three months, according to a new investigation.
The findings, published this week by Renée DiResta and Ronald Robertson in Lawfare, found that Grokipedia’s editing pipeline halted around 24 April. The site allows readers to submit corrections through a form, which are then supposed to be reviewed by the AI itself and marked “in review,” “approved,” “rejected” or “implemented.”
That process, which once resolved most submissions within minutes, has produced no new approvals or rejections since late April, the researchers found.
“The encyclopedia stopped editing itself and processing suggested edits from humans,” DiResta and Robertson wrote in the article. “As far as we can tell, no entry has changed in more than three months.”
To test the finding, the researchers submitted a straightforward factual update to Grokipedia’s page on SpaceX, noting the company’s 12 June initial public offering. The submission and identical suggestions from other users went unprocessed. Expanding the search, the researchers used the site’s search-autocomplete feature to identify roughly 300,000 of its most-visited pages.
Grokipedia does not publish a traffic ranking of its own. Across a sample of more than 34,000 pages with a combined 225,000 suggested edits, they found no edit had been accepted or rejected in the prior three months. The pattern held true regardless of the page’s popularity, political sensitivity or subject matter.
The investigation also identified problems with the reliability of Grokipedia’s own editing record. A batch of edits that Columbia’s Tow Center for Digital Journalism had documented as “accepted” earlier this year now appear as “rejected” in Grokipedia’s public logs, an apparent side effect of a March system rewrite that broke the way edits are anchored to specific passages of text. In several cases, the researchers found, the edits appear to have been incorporated into the live article despite being logged as rejected — meaning the site’s audit trail can no longer be taken at face value.
A slowdown
According to the Lawfare analysis, the automated systems responsible for a majority of Grokipedia’s edits went down in stages between March and mid-April, before the queue for human-submitted edits stopped moving entirely. These edits were managed by bots referred to internally as “grok” and “Grok Editor.” A public feed that displayed live edits also disappeared sometime between January and March, based on archived snapshots of the site. This feed was used by the Tow Center to collect data for its own research.
Some of the platform’s most active contributors had already noticed the slowdown. Posts on Reddit and X beginning in late April described suggested edits piling up unresolved and the site’s public edit feed vanishing from its homepage.
One contributor said in mid-June that the review system had been stalled for more than 50 days. In another X post, a user mentioned that their contact at xAI had acknowledged the problem without giving a timeline for a fix. Musk has not posted publicly about Grokipedia since February, the report said.
Also Read: Is Grokipedia really a neutral Wikipedia? Elon Musk wants you to believe it
What next for Grokipedia?
According to an article published in Nieman Lab, the web analytics firm Similarweb estimated the site received 6.7 million visits in June, and some individual pages draw far larger audiences. Grokipedia’s internal statistics reportedly show more than 83 million views on its page about former President Barack Obama.
An analysis by Ahrefs found Grokipedia content had been cited roughly 356,000 times by other AI systems, including ChatGPT and Google’s AI Mode. This is a fraction of Wikipedia’s estimated 25 million citations but a significant number nonetheless.
DiResta and Robertson argue the episode illustrates a structural weakness in AI-run reference systems compared with Wikipedia.
“Wikipedia’s contestability occurs at multiple levels: The edit history, reversions, and talk-page arguments are public, and readers who disagree have a mechanism for intervening. For better or worse, it blocks AI submissions,” they noted. “Grok is a closed, top-down system; readers can submit a request in the interface, but now nothing is behind it.”
They cited this as a major challenge for Grokipedia, which was originally released as an anti-woke, AI-powered Wikipedia.
“At the moment, Grokipedia is neither (observable nor accountable),” they concluded.
xAI did not respond to the findings in the reports. It remains unclear whether the company is preparing a major update to Grokipedia or has shelved the project.
(Edited by Stela Dey)