Role-Augmented Intent-Driven Generative Search Engine Optimization

Abstract

Generative Search Engines (GSEs), powered by Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG), are reshaping information retrieval. While commercial systems (e.g., BingChat, Perplexity.ai) demonstrate impressive semantic synthesis capabilities, their black-box nature fundamentally undermines established Search Engine Optimization (SEO) practices. Content creators face a critical challenge: their optimization strategies, effective in traditional search engines, are misaligned with generative retrieval contexts, resulting in diminished visibility. To bridge this gap, we propose a Role-Augmented Intent-Driven Generative Search Engine Optimization (G-SEO) method, providing a structured optimization pathway tailored for GSE scenarios. Our method models search intent through reflective refinement across diverse informational roles, enabling targeted content enhancement. To better evaluate the method under realistic settings, we address the benchmarking limitations of prior work by: (1) extending the GEO dataset with diversified query variations reflecting real-world search scenarios and (2) introducing G-Eval 2.0, a 6-level LLM-augmented evaluation rubric for fine-grained human-aligned assessment. Experimental results demonstrate that search intent serves as an effective signal for guiding content optimization, yielding significant improvements over single-aspect baseline approaches in both subjective impressions and objective content visibility within GSE responses.

Figures (7)

• Figure 1: Overview of Generative Search Engine (GSE) workflow. Upon receiving a user query, the system retrieves a set of relevant documents and feeds them into the large language model (LLM) to generate a synthesized response with source-level citations. Optimized content may increase its likelihood of being cited in the final response. Notably, in the black-box setting assumed in this work, the query is not visible to content creators.
• Figure 2: Overview of the Role-Augmented Intent-Driven G-SEO method. The method leverages search intent to guide the optimization process and further integrates reflection-based modeling from multiple roles to enhance generalizability to diverse user needs in complex GSE scenarios.
• Figure 3: Adaptability of G-SEO methods across diverse GSE retrieval scenarios. We evaluate each method’s adaptability by counting the number of optimized content instances that yield observable improvements in subjective and objective visibility across multiple retrieval tasks. This reflects the generalization capacity and real-world utility of each approach. Figure (a) shows a comparison between RAID G-SEO and other baselines, and figure (b) presents part of the ablation analysis.
• Figure 4: Distribution of optimization step preferences in RAID G-SEO. Each generated optimization step is structurally parsed to identify its corresponding optimization objective and operational strategy type, followed by semantic clustering. Subfigure (a) shows the distribution across intent-aligned target dimensions (e.g., enhancing content completeness, improving factual credibility, increasing clarity), indicating where the intent prioritizes refinement. Subfigure (b) presents the distribution over strategy categories (e.g., content restructuring, elaboration, redundancy reduction), capturing the model’s typical operational behavior under intent-driven guidance.
• Figure 5: Distribution of RAID G-SEO across multi-role perspectives. We perform semantic clustering on the user role descriptions generated by the 4W multi-role deep reflection module to characterize the types of cognitive perspectives involved during intent generalization. The results illustrate the relative frequency of each role category, reflecting the model’s response pattern to perspective distribution during optimization.