Grok answers with word salad — xAI calls it a rare glitch
Since Wednesday morning, users have reported strings of meaningless words instead of answers. Grok Lite was hit hardest; the status page reported normal operation throughout.

Illustration · AI-generated (AI IN LIFE)
At a glance
- First reports on Wednesday morning, 19 August 2026
- Primarily affected the Grok Lite model
- Sample output: "match it without and your they and two for planets can practical and often cheese"
- xAI: rare, temporary generation glitch; advice is to start a fresh chat or regenerate
- Official status page reported normal operation throughout; TechCrunch could not reproduce the fault
Since Wednesday morning, reports have piled up from Grok users receiving strings of unconnected words instead of an answer. One user who asked for a PDF received a reply that, according to TechCrunch, began "match it without and your they and two for planets can practical and often cheese" and continued that way for several paragraphs.
Similar cases collected on the Grok subreddit — word chains such as "app scale designs with that and to your horse choose batch". Based on the available reports, the smaller Grok Lite model was primarily affected.
xAI describes the pure word salad as a rare, temporary generation glitch and recommends starting a fresh chat or regenerating the response, saying it usually clears right away. In parallel, the official status page listed all Grok services as fully operational, with no incident reported.
One detail matters for context: TechCrunch reports it could not reproduce the error in its own testing. That suggests only a subset of users was affected — consistent with the description of a sporadic fault in generation.
The real point sits beside the bug itself. There was a gap between what users experienced and what the status page showed. For anyone wiring assistants into a workflow, that is the practically relevant observation: a green status indicator is not evidence that the output is usable.
FAQ
What exactly happened?
Users received meaningless strings of words instead of answers. xAI describes it as a rare, temporary generation glitch, mainly on the Grok Lite model.
What helps?
xAI recommends starting a fresh chat or regenerating the response, which it says usually resolves the issue immediately.
Were all users affected?
Apparently not. TechCrunch could not trigger the fault in its own testing, and the status page reported normal operation throughout.


