Grok Is Sending Users Word Salad Instead of Answers
A bug in xAI's Grok chatbot is filling some users' screens with meaningless strings of words. Here is what is happening, who is affected, and what to do if it hits you.
Key points
- Grok Lite, the lighter version of xAI's AI chatbot, began producing gibberish responses for some users as early as Wednesday morning, August 20, 2026.
- One affected user asked Grok to create a PDF and received: "match it without and your they and two for planets can practical and often cheese."
- The Grok account on X confirmed the bug Thursday morning, calling it "a rare temporary generation glitch."
- XAI's status page showed all services operational and suggested starting a fresh chat to clear the problem.
- XAI has lost at least 50 researchers and engineers in recent months, according to a May 2026 report by The Information.
Something broke inside Grok this week, and users noticed before xAI said a word.
Affected users of Grok Lite, the lower-powered tier of xAI's chatbot, began posting screenshots of bizarre responses on Wednesday. One person asked the chatbot to generate a PDF and got back a string of disconnected words: "match it without and your they and two for planets can practical and often cheese," followed by several more paragraphs of the same. Another user's response was packed with links to reinforcement learning research sites. Reinforcement learning is a training method where an AI system learns by trial and error, similar to teaching a dog with rewards.
The Grok subreddit, a community forum for Grok users on Reddit, filled up with complaints. TechCrunch first reported the issue after speaking with affected users.
Is this serious?
For now, the bug looks contained. It appears limited to direct conversations on Grok.com and hasn't touched the Grok account on X, which kept posting normally. XAI's status page listed all services as operational throughout.
The Grok team confirmed the problem on X Thursday morning: "That pure word salad is a rare temporary generation glitch. Official status at https://status.x.ai shows all Grok services fully operational with no incidents. Start a fresh chat or regenerate, it usually clears right away. Sorry about the gibberish."
Refreshing the chat session fixed things for many users. Some reported the nonsense kept coming even after several refreshes, though it's unclear how many people hit that wall. TechCrunch couldn't reproduce the issue in its own testing, which suggests only a slice of users were affected. XAI didn't respond to a request for comment on the cause.
What does this mean for Grok users?
If you see random word strings in a response, start a fresh chat. That single step works for most people. If it persists, check status.x.ai for updates.
This glitch lands at a rough moment. As we reported on 12 August, xAI has been pushing Grok aggressively into workplace automation. A chatbot that randomly produces gibberish is a harder sell when you're asking people to trust it with their tools. The company also lost most of its founding team and at least 50 researchers and engineers in recent months, per The Information's May 2026 report, and released a new foundation model in July describing it as faster and more cost-efficient than its predecessor.
AI chatbots can produce strange output when something goes wrong on the server side. It isn't your device, it isn't your account, and it's usually not permanent. What I'd watch here isn't the glitch itself but whether xAI explains the cause: a lab pitching always-on agents for serious work needs to say more than "sorry about the gibberish."
