Paper Type
ERF
Abstract
Generative AI (GenAI) search engines are changing digital marketing by giving users direct answers and reducing traffic to original content sources. In response, generative engine optimization (GEO) has emerged to restore brand visibility. However, most existing GEO methods rely on adversarial changes that can lower content quality and increase consumer skepticism. To address this limitation, this study proposes a responsible GEO framework grounded in persuasion knowledge model (PKM) theory. We model GEO as a constrained optimization problem and use Langevin dynamics to generate text perturbations that improve ranking while reducing perceptions of manipulative intent. To preserve content quality and factual consistency, the framework combines evidence-based initialization, multi-objective energy optimization, and gradient masking. This approach offers a human-centered and theory-driven way to improve brand visibility in GenAI search without weakening user trust.
Paper Number
1678
Recommended Citation
Huang, Tianle; Cheng, Xusen; and Zeng, Ang, "Responsible Generative Engine Optimization: Optimizing LLM Rankings via Trustworthy Text Generation" (2026). AMCIS 2026 Proceedings. 3.
https://aisel.aisnet.org/amcis2026/ai_sigculture/ai_sigculture/3
Responsible Generative Engine Optimization: Optimizing LLM Rankings via Trustworthy Text Generation
Generative AI (GenAI) search engines are changing digital marketing by giving users direct answers and reducing traffic to original content sources. In response, generative engine optimization (GEO) has emerged to restore brand visibility. However, most existing GEO methods rely on adversarial changes that can lower content quality and increase consumer skepticism. To address this limitation, this study proposes a responsible GEO framework grounded in persuasion knowledge model (PKM) theory. We model GEO as a constrained optimization problem and use Langevin dynamics to generate text perturbations that improve ranking while reducing perceptions of manipulative intent. To preserve content quality and factual consistency, the framework combines evidence-based initialization, multi-objective energy optimization, and gradient masking. This approach offers a human-centered and theory-driven way to improve brand visibility in GenAI search without weakening user trust.
Comments
SIG CULTURE