Think Before Writing: Feature-Level Multi-Objective Optimization for Generative Citation Visibility

Researchers Zikang Liu and Peilan Xu have published a paper introducing FeatGEO, a novel framework designed for the emerging field of Generative Engine Optimization (GEO). Unlike traditional SEO, which targets search engine rankings, GEO aims to optimize content for visibility within AI-powered answer engines like ChatGPT or Perplexity, which cite sources directly. The key innovation is moving beyond simple token-level text rewriting to a feature-level, multi-objective optimization approach. FeatGEO first abstracts a webpage into interpretable features across three categories: structural (like organization), content (like factual density), and linguistic (like readability).

Instead of editing text directly, the framework optimizes over this feature space to find the best configuration for citation likelihood and content quality. A separate language model then generates the final text based on these optimized features, decoupling high-level strategy from low-level writing. Experiments on the GEO-Bench dataset across three different generative engines demonstrated that FeatGEO consistently improved citation visibility while maintaining or even improving content quality, substantially outperforming existing token-level baselines.

The analysis revealed a critical insight: citation behavior in AI models is more strongly influenced by these holistic document-level properties than by isolated keyword swaps. Furthermore, the optimized feature configurations proved generalizable, working effectively across language models of different scales. This research provides a more controlled and interpretable methodology for creators and publishers who need their work to be surfaced and cited by the next generation of AI information interfaces.