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

Researchers developed an interpretable AI pipeline to analyze climate discourse across paid Meta advertisements and organic Bluesky posts from mid-2024 to mid-2025, revealing fundamental differences in messaging: paid platforms emphasize solution promotion in formal tones, while public social media centers on systemic critique with scientific grounding. The framework demonstrates how LLM-powered thematic analysis can surface structural differences in communication across heterogeneous platforms.

This research addresses a critical gap in understanding how climate messaging differs between controlled, commercial environments and organic public discourse. The study's significance lies not in climate communication alone, but in demonstrating how AI can systematically compare messaging strategies across fundamentally different platform architectures. By developing an unsupervised thematic discovery pipeline that requires no predefined topics, the researchers created a replicable methodology applicable far beyond climate analysis.

The divergence between paid and organic discourse reflects deeper structural incentives: advertisers on Meta optimize for conversion and brand positioning, naturally leading toward forward-looking solution narratives. Conversely, Bluesky's organic conversations reflect unmediated public concern, producing more crisis-oriented, scientifically rigorous discussion. This distinction matters because it reveals how algorithmic curation and monetization models systematically shape information environments. The framework's performance on downstream tasks like stance prediction validates that discovered themes capture meaningful semantic patterns rather than artifacts.

For developers and AI practitioners, this work demonstrates the practical value of interpretable clustering combined with LLM labeling—moving beyond black-box topic modeling toward human-verifiable thematic structures. The generalizability across communication environments suggests applications in policy analysis, market research, and misinformation detection. Investors tracking AI infrastructure should note the increasing sophistication of large-scale comparative text analysis, particularly for regulatory compliance and reputation monitoring use cases.

• →Unsupervised LLM-powered thematic discovery can systematically compare messaging across structurally different platforms without predefined topic inventories.
• →Paid advertising emphasizes solution promotion with formal tone while organic social media centers on systemic critique with scientific grounding.
• →The framework generalizes beyond climate communication to any heterogeneous communication environment requiring comparative analysis.
• →Downstream task performance validates that discovered themes meaningfully capture semantic patterns for stance prediction and retrieval.
• →Platform architecture and monetization models systematically shape information environments in measurable, analyzable ways.