xAI's Grok chatbot spouts gibberish in user queries
Users reported that xAI's Grok chatbot began returning nonsensical strings of text instead of coherent answers, prompting an official response.
On Wednesday morning, users of xAI’s Grok chatbot began reporting that the service was returning nonsensical strings of text instead of coherent answers.
Chronology of the glitch
According to Gizmodo, the issue was first noticed early on Wednesday when a handful of users on the dedicated Grok subreddit complained that even simple prompts yielded “pure word salad.” TechCrunch corroborated that its reporters could not reproduce the failure themselves, noting that the glitch could not be replicated in their own testing.
Media additions
The official Grok Twitter account later weighed in, calling the phenomenon a “rare temporary generation glitch.”
What users saw
Several subreddit posts quoted in Gizmodo illustrate the range of garbled outputs. A user who asked Grok to generate a PDF received a reply that began:
“match it without and your they and two for planets can practical and often cheese…”
Another query for the single word “balls” produced the response:
“app scale designs with that and to your horse choose batch.”
Further examples posted by community members included fragments such as “people stable stage right professional on steel” and a lone “skin.” at the start of a sentence. One user also pointed out that source links supplied by the bot redirected to reinforcement‑learning research sites instead of the expected references, raising concerns about the bot’s retrieval layer.
Official reaction
The Grok account on X responded to the complaints with a brief statement that the nonsensical output was “rare temporary generation glitch.” The wording of that reply, quoted verbatim, is the only official comment from xAI in the public record so far.
TechCrunch noted that its engineers were unable to recreate the problem, leaving the cause unresolved at the time of reporting. The outlet did not provide a technical diagnosis, merely stating the difficulty in reproducing the user‑observed behavior.
Business impact and market perception
Gizmodo also highlighted a YouGov UK poll that placed Grok at the second‑worst user‑satisfaction rating among major AI platforms, with a positive score of 24.6 %. Only Siri, which is not a general‑purpose chatbot, scored lower at 23.7 %.
Recon Analytics’ data for the second quarter of 2026 showed that a minuscule 0.174 % of Grok users considered the service worth paying for. The same analysis linked the low willingness‑to‑pay figure to a broader trend of shrinking user bases and challenges in entering enterprise markets.
The combination of a public‑facing glitch and already weak satisfaction metrics intensifies scrutiny of Elon Musk’s ownership of xAI. Both Gizmodo and TechCrunch reminded readers that Musk’s track record includes a prior confirmed incident where he “personally broken the X platform on at least one occasion.”
Industry context: reliability concerns across AI
Forbes published an analysis of Anthropic’s new text‑watermarking feature for its Claude model, explaining how hidden watermarks are embedded in generated text by subtly altering word choices. The piece clarified that only AI‑generated responses carry the watermark, not the original user‑submitted material, and that minor edits such as typo fixes typically do not embed a watermark.
The Grok glitch, which turned coherent prompts into incomprehensible strings, serves as a stark reminder that even well‑known conversational agents can falter dramatically.
