Title: Defining Web Search in the Era of Generative AI

Abstract:

The emergence of Large Language Models (LLMs) has ushered in generative search, a novel paradigm where models fetch web-based information pertinent to a user’s query and consolidate it into a unified, coherent answer. This approach stands in stark contrast to conventional web search, which delivers results as a ranked sequence of distinct web pages. This study investigates the specific dimensions along which generative search diverges from traditional methods. We perform a rigorous comparison between Google’s organic search results and five generative search platforms provided by three major vendors: Google, OpenAI, and Perplexity. Our findings highlight significant disparities among these engines regarding their dependence on internal versus external knowledge sources, the diversity of their citations, and the consistency of their outputs. Although generative systems frequently match traditional search in terms of topical breadth, they accomplish this through distinctly different retrieval patterns and synthesis techniques. Additionally, we demonstrate that generative search outputs are subject to variability over time and across multiple executions, presenting fresh challenges for system robustness. These results indicate that generative search introduces new evaluation criteria not addressed by current frameworks, underscoring the need for assessment methods that explicitly measure retrieval behavior, synthesis quality, and stability in generative search applications.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC