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

TL;DR

The paper investigates Grok, a public, platform-integrated LLM, operating on X, through three months of observational data to quantify how users call Grok, how Grok responds, and what social roles it adopts in public discourse. Employing a mixed-methods approach, the authors quantify usage (e.g., a $62 ext{%}$ reply rate and English prompts dominating at $51 ext{%}$), qualitatively derive a ten-category role taxonomy, and use LLMs and topic modeling to classify thousands of Grok interactions and user bios. They find that Grok predominantly acts as an information oracle but also fulfills dispute-related roles such as Truth Arbiter, Advocate, and Adversary, with roles and engagement shaped by user interests and conversation context. The work delivers a foundational quantitative and qualitative portrait of human–LLM interactions in a social, one-to-many environment, with implications for AI literacy, platform transparency, and policy design as LLMs become embedded in online social spaces.

Abstract

xAI's large language model, Grok, is called by millions of people each week on the social media platform X. Prior work characterizing how large language models are used has focused on private, one-on-one interactions. Grok's deployment on X represents a major departure from this setting, with interactions occurring in a public social space. In this paper, we systematically sample three months of interaction data to investigate how, when, and to what effect Grok is used on X. At the platform level, we find that Grok responds to 62% of requests, that the majority (51%) are in English, and that engagement is low, with half of Grok's responses receiving 20 or fewer views after 48 hours. We also inductively build a taxonomy of 10 roles that LLMs play in mediating social interactions and use these roles to analyze 41,735 interactions with Grok on X. We find that Grok most often serves as an information provider but, in contrast to LLM use in private one-on-one settings, also takes on roles related to dispute management, such as truth arbiter, advocate, and adversary. Finally, we characterize the population of X users who prompted Grok and find that their self-expressed interests are closely related to the roles the model assumes in the corresponding interactions. Our findings provide an initial quantitative description of human-AI interactions on X, and a broader understanding of the diverse roles that large language models might play in our online social spaces.

Figures (15)

• Figure 1: Four forms of focal post chains in our data.
• Figure 2: Time series of English-language "@grok" mention posts and Grok reply posts per day during our data collection period (August 18–November 17, 2025). About 62% of "@grok" mention posts receive a Grok reply, indicating user interest that exceeds actual interaction.
• Figure 3: As observed in the cumulative distribution plots, engagement with Grok Reply posts is lower relative to Conversation Root posts across all engagement metrics.
• Figure 4: Distribution of categories of interaction with Grok. Note that each interaction can be assigned to more than one category. Information seeking and fact-checking were the most common categories.
• Figure 5: Ten roles of Grok, plotted on axes of allocentric vs. egocentric and supportive vs. constraining, shaded yellow if they reflect an LLM role emergent on social media.