Grokipedia stopped updating, at least in the public evidence available to researchers. An analysis by Renée DiResta and Ronald Robertson for Lawfare found that xAI’s machine-written encyclopedia appears to have ceased making observable article changes and processing reader corrections around April 24. The finding does not establish whether xAI paused the service, is rebuilding it, or abandoned it.
That distinction matters. Grokipedia is supposed to be maintained by xAI’s Grok model, with readers able to propose corrections through a public status queue. A suggestion begins as “in review,” then can be approved or rejected and, if approved, implemented. Lawfare says that mechanism once reached decisions quickly, sometimes within minutes.
Why does Grokipedia appear to have stopped updating?
The researchers examined 34,519 pages containing at least one suggestion, covering 225,496 proposed edits. They found no accepted or rejected corrections dated within the preceding three months. On Grokipedia’s ChatGPT page, 12 suggestions submitted April 24 were approved that day; suggestions submitted in May, June and July remained in review.
Lawfare counted 13,002 unresolved suggestions. Human contributors were still submitting corrections after April, at an average of 216 per week across the pages studied. In other words, the suggestion box remained open while the visible decision process appears to have stopped moving.
The analysis also found that the automated accounts called “grok” and “Grok Editor,” which accounted for 57.8% of all edit requests, stopped submitting model-generated requests in stages between March and mid-April. After late April, all submissions in the observed data came from people and entered the unresolved queue.
Grokipedia launched in October 2025 as xAI’s machine-written alternative to Wikipedia. The underlying large language model generates and revises text rather than relying on a conventional community-editing process, though readers can submit proposed fixes. For a plain-language account of that machinery, see how LLMs generate answers.
The audit trail had already become unreliable
The apparent April halt is separate from a problem in Grokipedia’s edit history. Lawfare reported that a broad rewrite around March 14 appears to have broken the links, or anchors, between suggested changes and the passages they targeted.
After that rewrite, Grokipedia’s live logging system appears to have retroactively displayed some edits previously recorded as accepted as rejected, with a “Highlighted section not found” error. In examples reviewed by the researchers, the affected text still appeared in the article. That leaves contributors with a public record that can say their correction failed even where the change seems to have landed.
The practical risk is not mysterious: a reference site that is no longer visibly revising entries can retain outdated claims, while an unstable edit record makes it harder for readers to check what changed or challenge an error. Gizmodo reported that xAI did not immediately respond to its request for comment. No supplied reporting establishes why the observed pipeline stalled or what xAI plans to do with Grokipedia next.
This story draws on original reporting from Techdirt.