1、Characterizing Web Search in The Age of Generative AIElisabeth Kirsten,Jost Grosse Perdekamp,Mihir UpadhyayKrishna P.Gummadi,Muhammad Bilal Zafarelisabeth.kirstenrub.deRuhr University Bochum,UAR RC Trust,MPI-SWSGermanyAbstractTheadventofLLMshasgivenrisetoanewtypeofwebsearch:Gen-erative search,where

2、LLMs retrieve web pages related to a queryand generate a single,coherent text as a response.This outputmodality stands in stark contrast to traditional web search,whereresults are returned as a ranked list of independent web pages.Inthis paper,we ask:Along what dimensions do generative searchoutputs

3、 differ from traditional web search?We compare Google,atraditional web search engine,with four generative search enginesfrom two providers(Google and OpenAI)across queries from fourdomains.Our analysis reveals intriguing differences.Most genera-tive search engines cover a wider range of sources comp

4、ared to websearch.Generative search engines vary in the degree to which theyrely on internal knowledge contained within the model parametersv.s.external knowledge retrieved from the web.Generative searchengines surface varying sets of concepts,creating new opportuni-ties for enhancing search diversi

5、ty and serendipity.Our results alsohighlight the need for revisiting evaluation criteria for web searchin the age of Generative AI.1IntroductionWhile traditional web search has been ever-evolving over the lastthree decades,its most basic mechanics have remained largelyunchanged.In response to a user

6、 query,web search engines suchas Brave,Bing,DuckDuckGo,and Google return a ranked list ofroughly ten search results.These results are ranked based on theirrelevance,source authority,and other criteria like diversity andalignment with the users profile 33.A vast body of research hasfocused on optimiz