Who gets Grokked the most?

Every day, tens of thousands of X users turn to Grok with a request to fact-check something they saw on the platform. 

This is a particularly bad idea during breaking news events. The AI chatbot has confidently spread misinformation about the Bondi Beach attack, the killing of Alex Pretti, and US strikes on Iran. (It’s gotten to the point that even X’s head of product is correcting Grok on main.)

Still, lots and lots of people continue to ask: “@grok is this true?” 

While I strongly advise against trusting the chatbot for a one-off evaluation on fast moving news stories and conflict imagery, research shows that Grok’s assessments may be more right than wrong at scale and improve over time. Regardless of Grok’s actual answer, a high presence of “@grok is this true”-style replies is a good sign that a tweet could contain viral misinformation or a newsworthy claim.

So Indicator built a public dashboard to track these tweets. The Grok-is-this-True Tracker displays data about the tweets that generated the most English-language Grok fact-checking requests. This includes how many Grok fact checks were requested for each tweet, the median Grok reply, and whether Community Notes are attached or proposed to the tweet.

The dashboard relies on methodology developed by researchers Thomas Renault, Mohsen Mosleh, and David Rand in a recent preprint.¹

Whenever you structure and visualize big sources of data, stories emerge. One thing our dashboard confirms is that Grok continues to rely on mainstream sources that its creator Elon Musk has disparaged, like Wikipedia and Reuters. At a more granular level, these tweets can help expose patterns of deception by high-impact X users, including one case of successful subversion of the Community Notes rating process.