Grok Users Reported ‘Word Salad’ Responses in a Limited Chatbot Glitch

Some Grok users briefly received strings of unrelated words instead of coherent answers. The reports describe an intermittent problem affecting a subset of users—not evidence of a confirmed outage across the entire service.

What the gibberish looked like

The failure was more than an occasional typo or strange sentence. One user asked Grok to generate a PDF and received several paragraphs beginning:

“match it without and your they and two for planets can practical and often cheese…”

Other reported replies included phrases such as “App scale designs with that and to your horse,” while one user said purported source links led to a collection of reinforcement-learning research sites rather than useful references.

In other words, Grok was producing fluent-looking text fragments without a meaningful connection to the prompt.

When and where it happened

Users told reporters they noticed the problem as early as Wednesday morning this week. The reports appeared concentrated on direct prompts submitted through Grok.com, rather than the Grok service accessed through X.com. The X-based service was reported as unaffected.

The users who spoke with TechCrunch said they were using Grok Lite. That does not establish that every Grok Lite account was affected, or that other Grok versions were definitively immune; it identifies the version most commonly mentioned in the available reports.

How widespread and reproducible was it?

Complaints were substantial enough to overwhelm discussion in the Grok Reddit community, but the available evidence still points to a relatively small subset of users. TechCrunch said it could not reproduce the problem, making the incident difficult to verify consistently.

That combination—many user reports but no reliable reproduction—supports a cautious description: a real reported failure with uncertain scope, rather than a confirmed system-wide breakdown.

What users tried—and what worked

Several users said that refreshing the session restored normal replies. xAI’s Grok account recommended three similar steps:

• Start a fresh chat.
• Regenerate the response.
• Refresh the session.

The company said the problem usually cleared promptly after those actions. However, some users reported that the gibberish continued through multiple refreshes, so the workaround was not universal.

How xAI described the incident

Grok’s official X account called the output a “rare temporary generation glitch,” apologized for the gibberish, and directed users to the company’s status page.

At the time covered by the reports, that status page listed Grok’s services as “fully operational” and showed no incident. That status conflicted with the user complaints, but it also suggests the issue was not being treated publicly as a broad service outage.

Does it connect to xAI’s staff departures?

The glitch was reported during a period of significant turnover at xAI and its successor SpaceXAI. TechCrunch reported that more than 50 researchers and engineers had departed since February, including people associated with coding, world models, and Grok voice. Earlier reporting also documented the departure of six members of xAI’s 12-person founding team.

That context naturally raises questions about engineering capacity and operational reliability. But the available reporting does not show that employee departures caused the word-salad incident. The most defensible conclusion is that the staffing losses were surrounding context, not an established explanation for this particular failure.

The practical takeaway

The reported Grok problem looked like a temporary generation failure: coherent language fragments were assembled into long, meaningless answers, especially for some Grok Lite users on Grok.com. A new chat, regeneration, or refresh often helped, but not for everyone. xAI acknowledged the issue as a rare glitch, while its public status page continued to show normal operations.

Until the cause and scope are confirmed, users should treat any affected response—including links, citations, and generated files—as unreliable and verify it independently.