[5E1-GS-6d] Language media processing

9:15 AM - 9:30 AM JST(12:15 AM - 12:30 AM UTC)

[5E1-GS-6d-02] An Analysis on Citation Strategies of AI Summarization in Search Engines

〇Takumi Ogawa¹, Satoshi Kimura², Takafumi Koshinaka¹ (1. Yokohama City University, 2. CyberAgent, Inc.)

Keywords:

Deep Learning,Language Model,AI Overview

Search engine characteristics have been studied within the context of Search Engine Optimization (SEO). However, the behavior of "AI Overviews," increasingly prevalent in recent search results, remains underexplored. Consequently, a new trend known as Generative Engine Optimization (GEO) has emerged, aiming to understand the mechanisms by which AI recommends specific brands or references websites. This study constructs a search ranking prediction model based on Japanese ModernBERT, trained using a pairwise approach with queries and top-100 web content from standard organic searches, to analyze the semantic relevence between queries and web page contents. The results indicate that content referenced in AI Overviews exhibits high relevance to the query, surpassing that of highly ranked content in organic searches. These findings suggest that GEO places greater emphasis on query-content relevance than traditional SEO.