Simulating and Experimenting with Social Media Mobilization Using LLM Agents

TL;DR

The paper tackles how peer influence in online networks shapes political mobilization by presenting LLM-SocioPol, an agent-based simulator that fuses census-grounded demographics, authentic Twitter topology, and heterogeneous LLM agents (GPT-4.1 variants) to model voting behavior. It reproduces the qualitative findings of the classic Facebook mobilization study, showing stronger effects from social messages and observable spillovers, while enabling counterfactuals and sensitivity analyses beyond real-world constraints. The study discusses quantitative differences in magnitude due to the isolated online setting and extended pre-election exposure, highlighting both the fidelity and limitations of high-fidelity LLM-based simulations. By providing a publicly available codebase, it offers a reproducible ex-ante platform for testing intervention designs and causal estimators that complement field experiments. The framework thus bridges rigorous empirical designs with flexible computational modeling for political mobilization research.

Abstract

Online social networks have transformed the ways in which political mobilization messages are disseminated, raising new questions about how peer influence operates at scale. Building on the landmark 61-million-person Facebook experiment \citep{bond201261}, we develop an agent-based simulation framework that integrates real U.S. Census demographic distributions, authentic Twitter network topology, and heterogeneous large language model (LLM) agents to examine the effect of mobilization messages on voter turnout. Each simulated agent is assigned demographic attributes, a personal political stance, and an LLM variant (\texttt{GPT-4.1}, \texttt{GPT-4.1-Mini}, or \texttt{GPT-4.1-Nano}) reflecting its political sophistication. Agents interact over realistic social network structures, receiving personalized feeds and dynamically updating their engagement behaviors and voting intentions. Experimental conditions replicate the informational and social mobilization treatments of the original Facebook study. Across scenarios, the simulator reproduces qualitative patterns observed in field experiments, including stronger mobilization effects under social message treatments and measurable peer spillovers. Our framework provides a controlled, reproducible environment for testing counterfactual designs and sensitivity analyses in political mobilization research, offering a bridge between high-validity field experiments and flexible computational modeling.\footnote{Code and data available at https://github.com/CausalMP/LLM-SocioPol}

Figures (9)

• Figure 1: Overview of the LLM Social-Political Mobilization (LLM-SocioPol) simulator. The framework integrates real demographic and network data with heterogeneous LLM agents to model online voter mobilization. U.S. Census and Twitter data are first used to construct a realistic population and social graph. Each agent’s profile is then enhanced with demographic attributes and political-stance scores before being assigned an LLM model tier (GPT-4.1, GPT-4.1-Mini, or GPT-4.1-Nano) reflecting individual sophistication. Within the simulation environment, agents manage their follow relationships, engage with and create posts, process social-influence cues, and continuously update their voting intentions. The example panel shows a representative user profile, a treated feed containing a social-message prompt, and the 0–4 voting-likelihood scale for outcome measurement.
• Figure 2: Voting turnout and Difference-in-Means (DM) estimates across experimental conditions. Left: Average turnout across five random seeds (19,785 agents each; hollow circles show individual seeds). Middle and right: Estimated effects of informational and social messages on voting turnout. Social messages boost turnout far more than informational ones, mirroring the qualitative pattern in bond201261.
• Figure 3: Normalized voting intentions (to 0--1 scale) on final pre-election period (period 30). Left: mean intentions. Middle and right: Estimated effects of informational and social messages on voting intention.
• Figure 4: Mean voting intentions (0--4 scale) over pre-election periods. Vertical lines mark treatment phase transitions. The social message consistently demonstrates higher intentions than the informational message.
• Figure 5: Treatment effects on normalized voting intentions aggregated across five seeds. Left: informational message. Right: social message. Vertical lines mark treatment phase transitions.