New study reveals how 169K users interact with Grok on X
76.8% of users invoke Grok exactly once — here's why
A new arXiv paper titled "Asking Grok: AI-Assisted Sensemaking in Social Media Conversations" (arXiv:2605.19720) provides the first large-scale empirical analysis of user interactions with xAI's Grok assistant on X. The researchers collected 169,137 posts that explicitly invoked Grok, categorizing request types and usage contexts. Key findings show Grok is predominantly used reactively — users summon it to verify claims or obtain contextual information mid-conversation. Despite rapid response times, Grok's reach is limited, with the typical response viewed by only a small audience. Adoption patterns reveal a stark power-law: 76.8% of users invoked Grok just once, indicating shallow engagement rather than habitual use.
Comparing Grok interactions with X's Community Notes system, the study found limited overlap — only a small fraction of Grok-invoked posts also received Community Notes. When overlap occurred, it concentrated on verification-oriented and high-visibility content. Critically, Grok interactions consistently happened earlier than Community Note corrections, and Grok's presence did not predict subsequent fact-checking activity. The authors conclude that AI assistants like Grok function as an early complementary sensemaking layer on social media, not a replacement for crowd-based fact-checking. This suggests platforms should design for AI-human collaboration rather than substitution.
- 169,137 posts analyzed; 76.8% of users invoked Grok only once, showing shallow adoption
- Grok used primarily for reactive information verification, not proactive exploration
- Overlap with Community Notes is limited; Grok interactions occur earlier and don't predict fact-checks
Why It Matters
Grok's role as an early sensemaking layer, not a replacement, shapes how platforms design AI fact-checking