生成式人工智能时代的网络搜索特征研究（英）-2026.pdfVIP

CharacterizingWebSearchinTheAgeofGenerativeAI

ElisabethKirsten,†‡JostGrossePerdekamp,†‡MihirUpadhyay‡

KrishnaP.Gummadi,⊺MuhammadBilalZafar†‡

elisabeth.kirsten@rub.de

†RuhrUniversityBochum,‡UARRCTrust,⊺MPI-SWS

Germany

Abstractmanuallynavigateinordertoviewthelower-rankedresults.Prior

TheadventofLLMshasgivenrisetoanewtypeofwebsearch:Gen-researchindicatesthatusersoftendonotnavigatebeyondthetop

erativesearch,whereLLMsretrievewebpagesrelatedtoaquerytensearchresults[8,32],oreventhetopthreeresults[2].Onthe

andgenerateasingle,coherenttextasaresponse.Thisoutputotherhand,generativesearchcanpotentiallysynthesizecontent

5modalitystandsinstarkcontrasttotraditionalwebsearch,wherefromfarmorethantensourcesintoasinglecoherentresponse.(iii)

2Relianceonexternalv.s.internalknowledge:Traditionalweb

0resultsarereturnedasarankedlistofindependentwebpages.Insearchlargelyreliesonexternalsources,thatis,webpages,tode-

2thispaper,weask:Alongwhatdimensionsdogenerativesearchliversearchresults.Generativesearch,ontheotherhand,combines

outputsdifferfromtraditionalwebsearch?WecompareGoogle,a

t