Grok in the Wild: Characterizing the Roles and Uses of Large Language Models on Social Media

Grok responds to the majority of requests but engagement remains limited: The model responds to 62% of requests on X, yet half of Grok's responses receive 20 or fewer views within 48 hours, suggesting that while the tool is widely available, its actual impact on social conversations may be modest.

Public social media use differs fundamentally from private AI interactions: Unlike one-on-one settings where LLMs primarily provide information, Grok on X frequently assumes roles in dispute management—acting as a "truth arbiter," "advocate," and "adversary"—raising questions about the appropriateness of AI systems mediating public disagreements and conflicts.

LLMs are taking on diverse and sometimes problematic social roles: The researchers identified 10 distinct roles that Grok plays in social interactions, with the model most commonly serving as an information provider, but also functioning in ways that could amplify polarization or misinformation in public discourse.

User interests predict how they deploy AI in social spaces: The study found that users' self-expressed interests closely correlate with the roles they assign to Grok, suggesting that people may be using AI tools to amplify their existing viewpoints rather than to gain balanced perspectives—a potential concern for responsible AI deployment.

Public AI deployment requires new governance frameworks: This research highlights that AI systems designed for private use may behave differently and raise distinct risks when deployed in public social spaces, suggesting the need for platform-specific guidelines and oversight mechanisms.