CC-GSEO-Bench: A Content-Centric Benchmark for Measuring Source Influence in Generative Search Engines

TL;DR

CC-GSEO-Bench addresses the shift from traditional ranked visibility to content-driven influence in Generative Search Engines by introducing a large, article-centered benchmark paired with a creator-focused evaluation framework. It defines three core influence dimensions—Exposure, Faithful Credit, and Causal Impact—plus two content-quality axes, Readability & Structure and Trustworthiness & Safety, and aggregates signals across article-associated query clusters using an offline retrieval-simulation setup with counterfactuals to estimate marginal contribution. The benchmark comprises over 1,000 source articles and more than 5,000 query-article pairs, enabling robust article-level analysis of influence dynamics; it also demonstrates through nine rewriting strategies how content design can trade off readability, safety, and influence, revealing rank effects and context-dependent optimal strategies. The work provides a principled, multi-objective measurement foundation for GSEO, informing content creators and researchers about how to optimize source visibility in AI-generated responses while highlighting limitations such as offline assumptions and potential ethical risks in manipulating search narratives.

Abstract

Generative Search Engines (GSEs) synthesize conversational answers from multiple sources, weakening the long-standing link between search ranking and digital visibility. This shift raises a central question for content creators: How can we reliably quantify a source article's influence on a GSE's synthesized answer across diverse intents and follow-up questions? We introduce CC-GSEO-Bench, a content-centric benchmark that couples a large-scale dataset with a creator-centered evaluation framework. The dataset contains over 1,000 source articles and over 5,000 query-article pairs, organized in a one-to-many structure for article-level evaluation. We ground construction in realistic retrieval by combining seed queries from public QA datasets with limited synthesized augmentation and retaining only queries whose paired source reappears in a follow-up retrieval step. On top of this dataset, we operationalize influence along three core dimensions: Exposure, Faithful Credit, and Causal Impact, and two content-quality dimensions: Readability and Structure, and Trustworthiness and Safety. We aggregate query-level signals over each article's query cluster to summarize influence strength, coverage, and stability, and empirically characterize influence dynamics across representative content patterns.

Figures (4)

• Figure 1: Shift from TSEs with ranked visibility to GSEs with synthesized answers, illustrating the new uncertainties for creators in measuring and influencing content display.
• Figure 2: Illustration of the CC-GSEO-Bench framework. We quantify influence through five dimensions: Exposure (E), Faithful Credit (F), Causal Impact (C), Readability (R), and Trustworthiness (T). In this example, Source 1 demonstrates high Readability and Trustworthiness which translates into dominant Exposure and Causal Impact as its logical framework shapes the final recommendation. Conversely, Source 2 exhibits low Trustworthiness and fails to positively influence the synthesized answer despite being retrieved.
• Figure 3: Metric correlation heatmap on gpt-oss-120b. Each cell reports the mean Pearson correlation ($r$) between two metric dimensions (E/F/C/R/T), computed at the item level and averaged over optimization runs. The matrix highlights generally weak cross-metric dependencies (e.g., near-zero E--C), alongside several moderate couplings (E--F, F--C, and R--T), supporting the need for multi-objective trade-off analysis.
• Figure 4: Position effect on gpt-oss-120b. We report the mean Exposure (E), Faithful Credit (F), and Causal Impact (C) across retrieval ranks for the None baseline and several high-performing optimization strategies; More Quotes remains comparatively robust at lower ranks, especially on F and C.