xAI’s Grok chatbot has been producing gibberish in response to users, sparking confusion in its Grok Lite tier. Users report bizarre outputs like “match it without and your they and two for planets can practical and often cheese…” instead of useful content. The issue emerged early Wednesday and seems limited to a smaller group of users—the normal experience remains for most.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
What’s happening and who’s affected
Many of the affected are Grok Lite users sending direct requests on Grok.com, not via external platforms. One user received nonsensical text spanning multiple paragraphs; another found their output cluttered with links to research on reinforcement learning—clearly not what was asked.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
Some managed to reset things by refreshing the session. Yet for others, neither refreshing nor starting new chats helped—the disruptions persisted. Meanwhile, Grok’s account on X (formerly Twitter) appears unaffected.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
Official response and system status
The Grok account on X acknowledged the “word salad” problem, calling it a rare generation glitch and suggested starting a fresh chat or regenerating responses to fix it.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/)) It also pointed to the status page, which claims all services are operational with no reported incidents.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
It’s worth noting xAI has gone through substantial staff changes in recent months. Much of its founding team has departed, along with around 50 researchers and engineers.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/)) Just last month, the company released a new foundation model it described as “Opus-class, but faster, more token-efficient and lower cost.”([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
At least one user suggests the issue surfaced Wednesday morning. Tests by external observers (including tech outlets) have so far failed to reproduce the problem. That implies the glitch may stem from specific edge conditions—certain accounts or user flows.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
Errors seem to exclusively hit direct chats with Grok on its own site. There’s been no reported distortion for users interacting with the AI via social media channels.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
Context: Grok’s journey and recent changes
Grok launched from xAI as a challenger in conversational AI, aiming to offer fast, efficient models and direct engagement with users. Earlier this year, xAI’s Opus-class foundation model earned praise for balancing speed, token efficiency, and cost.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/)) But the departure of key engineering and research staff, along with leadership changes, has raised questions about long-term stability.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
Glitches aren’t unusual in large language models; generation errors, inconsistencies, and failures scale with complexity and user base. Yet frequency and scope matter. Grok’s issue appears minor in scale—small user pool—but it’s striking enough to create distrust over reliability.([techcrunch.com](https://techcrunch.com/2026/08/20/grok-keeps-sending-gibberish-responses-to-users/))
What this means going forward is simple: xAI needs patches, transparency, and stable infrastructure. If not addressed, even rare glitches can fracture user confidence—especially in a competitive field where ChatGPT, Claude, and others chase every fraction of performance and coherence.
