SAGEO Arena: A Realistic Environment for Evaluating Search-Augmented Generative Engine Optimization

TL;DR

SAGEO Arena addresses the lack of end-to-end evaluation for Search-Augmented Generative Engine Optimization by providing a realistic, reproducible benchmark that couples a full generative search pipeline with a structurally rich web-corpus. It demonstrates that optimizing body text alone often harms retrieval and that structural information signals are crucial for stage-wise visibility, particularly at retrieval, reranking, and generation. The paper introduces stage-level visibility metrics, a target-document framework, and ten optimization strategies, plus a stage-aware SAGEO approach that tailors changes to each pipeline stage. Findings reveal substantial, domain-dependent effects and demonstrate that current GEO/SEO strategies must be adapted to the full generative pipeline to improve AI-generated answers. The benchmark and insights establish a path toward realistic, generalizable SAGEO research and practical optimization guidance for search-augmented generation systems.

Abstract

Search-Augmented Generative Engines (SAGE) have emerged as a new paradigm for information access, bridging web-scale retrieval with generative capabilities to deliver synthesized answers. This shift has fundamentally reshaped how web content gains exposure online, giving rise to Search-Augmented Generative Engine Optimization (SAGEO), the practice of optimizing web documents to improve their visibility in AI-generated responses. Despite growing interest, no evaluation environment currently supports comprehensive investigation of SAGEO. Specifically, existing benchmarks lack end-to-end visibility evaluation of optimization strategies, operating on pre-determined candidate documents that abstract away retrieval and reranking preceding generation. Moreover, existing benchmarks discard structural information (e.g., schema markup) present in real web documents, overlooking the rich signals that search systems actively leverage in practice. Motivated by these gaps, we introduce SAGEO Arena, a realistic and reproducible environment for stage-level SAGEO analysis. Our objective is to jointly target search-oriented optimization (SEO) and generation-centric optimization (GEO). To achieve this, we integrate a full generative search pipeline over a large-scale corpus of web documents with rich structural information. Our findings reveal that existing approaches remain largely impractical under realistic conditions and often degrade performance in retrieval and reranking. We also find that structural information helps mitigate these limitations, and that effective SAGEO requires tailoring optimization to each pipeline stage. Overall, our benchmark paves the way for realistic SAGEO evaluation and optimization beyond simplified settings.

Figures (7)

• Figure 1: Optimizing for ranking alone (SEO) or citation alone (GEO) causes failure at either retrieval/reranking or generation. SAGEO jointly optimizes structural information and body texts across the full pipeline to succeed at both stages.
• Figure 2: Overview of SAGEO Arena. During document corpus construction, we extract structural information (title, meta description, headings, schema) and body text from web documents. During SAGEO evaluation, we first execute the pipeline with each test query and randomly select a target document from those that reach the generation stage (top-$k$ after reranking) as the baseline. The target document is then optimized, re-indexed into the corpus, and the pipeline is re-executed with the same query, tracking whether the document maintains, gains, or loses visibility at each stage (retrieval, reranking, generation).
• Figure 3: Reranking case studies illustrating two factors that improve document ranking after optimization. Case A shows that adding content directly addresses the query's informational need boosts rank. Case B shows that placing the direct answer in early paragraphs improves ranking position.
• Figure 4: Effectiveness of each optimization strategy across nine domains, measured by change in citation rate at the generation stage. Positive % indicate increased citation rate.
• Figure 5: Win rate comparison across three backbone models (GPT-5-mini, LLaMA-3.3-70B, Qwen3-80B) at each stage. A model is considered a win when its optimized document achieves the highest rank among the three at a given stage.