New research highlights risks of Generative Engine Optimization in recommendation agents

A new research paper introduces SafeGEO, an evaluation suite designed to understand the risks associated with Generative Engine Optimization (GEO) in recommendation agents. GEO allows content owners to rewrite web content to improve their visibility in generative systems, potentially leading to flawed products being favored. The study found that GEO attacks can increase the promotion of flawed products by up to 83.2%. While simple defenses like defensive prompting and structured evidence checks can reduce this harmful promotion by up to 39.2%, they do not fully restore performance to the level seen without GEO, indicating that it remains a significant risk. AI

IMPACT Highlights potential manipulation of AI recommendation systems, impacting user trust and product discovery.

RANK_REASON The cluster contains a research paper detailing a new evaluation suite and findings on a specific risk in AI systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• CORE Recommender
• Generative Engine Optimization
• Hugging Face
• SafeGEO

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →