xAI’s Grok chatbot began spitting out gibberish this week in response to direct queries, TechCrunch reported. According to TechCrunch, one user reported that they asked Grok to generate a PDF, receiving a reply beginning “match it without and your they and two for planets can practical and often cheese…” (It just sort of went on from there, apparently.) Another user found that source links provided by Grok led to reinforcement learning research sites instead. Users told TechCrunch they first noticed the issues on Grok Lite, beginning on Wednesday morning.
Complaints on the Grok subreddit included a reply to the word “balls” that began, “app scale designs with that and to your horse choose batch.” Other users posted similar gibberish from Grok in recent days, like “people stable stage right professional on steel” or this one just starting out with the sentence “skin.” Truly a blow to the “jailbreaking Grok for nonconsensual sexual imagery” community, as well as any other users that may theoretically exist.
Grok is primarily useful within the context of the X platform itself, though Tesla also sells drivers a Grok-powered voice assistant. Its biggest hits include a Wikipedia knockoff known for Holocaust denial and AI-generated “anti-woke” content. A recent YouGov UK poll clocked it as having the second-worst user satisfaction scores of any major AI platform at 24.6% positive. It managed to beat only Siri’s 23.7% positive rating, even though Siri isn’t even a general-purpose AI. Other reports have indicated Grok is shedding users, failing to generate revenue or penetrate enterprise markets, and has convinced only a miniscule number of users it’s worth paying for (0.174% in Q2 2026, according to Recon Analytics). X/xAI owner Elon Musk doesn’t have the best track record, having personally broken the X platform on at least one confirmed prior occasion.
TechCrunch noted it wasn’t able to replicate the technical issues experienced by users… but how could they tell, am I right folks?
The official Grok account responded to one user’s complaints by characterizing the “pure word salad” as a “rare temporary generation glitch,” although it should be noted that asking a chatbot why it’s malfunctioning is like asking a dog to do math.