Generative Engines Need Mechanism Design for Win-Win Outcomes

Generative AI platforms are increasingly relying on citations to attribute information and value to content providers. This creates a conflict where content creators might optimize their material solely for AI citation, potentially compromising quality and accuracy, while platforms strive to maintain high-quality, trustworthy answers. This paper models this interaction as a strategic game, showing how such tensions can escalate into "citation wars" where content is rewritten to seek citations, degrading overall quality. To address this, the researchers introduce a novel platform-creator mechanism called Verifiable-Content Rewards (VCR). Instead of merely penalizing suspicious content, VCR actively credits creators for rewrites that surface verifiable factual substance. This approach aims to align the incentives of content creators with the platform's goal of providing trustworthy information. Empirical evaluations across three benchmarks demonstrate that VCR consistently achieves superior defense utility, significantly outperforming current baseline methods. The mechanism fosters a mutually beneficial outcome, ensuring both content providers and generative engine platforms achieve their objectives without compromising information integrity.