Paid Voices vs. Public Feeds: Interpretable Cross-Platform Theme-Based Analysis of Climate Discourse

[arXiv]

Abstract

Online climate discourse unfolds across communication environments with distinct production, distribution, and audience structures, yet these environments are rarely compared directly. We study climate-related Meta advertisements and public Bluesky posts from July 2024 to September 2025 to examine how they differ in what they discuss, how they communicate shared issues, and how they respond to major real-world events. To support this comparison, we propose an interpretable theme-based measurement framework that combines semantic clustering with large language model (LLM) assisted theme induction and assignment, requiring no predefined topic inventory or seed set. We also evaluate the utility of the discovered themes through downstream stance prediction and theme-guided retrieval tasks. Our analysis reveals systematic differences between the two environments. Meta advertising is distributed across a broader set of solution- and promotion-oriented themes, while Bluesky discourse is more concentrated around institutional, corporate, and systemic critique. Importantly, we distinguish differences in thematic composition from differences in rhetorical expression: although crisis-versus-solution framing is largely explained by the topics emphasized on each platform, differences in formality, authority, tone, and scientific framing persist even within shared themes. We also find divergent temporal responses to major events, including the 2024 U.S. presidential election, after which climate advertising contracts while public discourse expands. Together, these findings show that structurally different communication environments organize climate discourse differently not only in thematic agenda, but also in rhetorical presentation and responsiveness to external events.