This research introduces a novel approach combining network analysis with LLM prompting to detect organized political disinformation campaigns under extreme class imbalance conditions.

• Leverages social network structures to inform prompt engineering for large language models
• Addresses the critical challenge of extreme class imbalance in real-world disinformation detection
• Demonstrates how network-derived insights can significantly improve LLM performance on security tasks
• Provides a framework that maintains effectiveness while reducing computational demands

This approach matters for security professionals as it offers a practical method to identify coordinated inauthentic behavior at scale, potentially enabling earlier intervention against influence operations that threaten democratic processes and information integrity.

Network-informed Prompt Engineering against Organized Astroturf Campaigns under Extreme Class Imbalance