26September2025

Qualitativeresearch:

UserexperiencesofGenerativeArtificialIntelligence(GenAI)Search

PreparedbyYouGovQualitative

Classification:CONFIDENTIAL

Classification:CONFIDENTIAL

3

1Introduction

4

Background

YOUGOV

In2025,OfcomcommissionedYouGovtoconductqualitativeresearchtoexploretraditionalonlinesearchexperiencesandGenAIsearchexperiencesamongUKadultinternetusers.Wedefinetheseexperiencesasfollows:

?TraditionalonlinesearchoccurswhenauserentersaqueryonlineusingtoolswhichreturnarankedlistofURL(UniformResourceLocator)linksthatuserscanvisittofindinformation(e.g.,Bing,Google,Yahoo).

?GenAIsearchoccurswhenauserentersaqueryonlineandreceivesdirect,natural-languageresponses.Thisincludes:

?AIsearchchatbots(‘AIchatbots')whichprovideaconversationalsearchexperiencebyansweringquestionsusingselectedsourcesandsupportingfollow-upquestions(e.g.,ChatGPT).

?AIsearchsummaries(‘AIsummaries')whichprovidedirectresponsestouser

questionsthroughasummaryofselectedsources,typicallyshownabovearankedlistofURLlinkswithinatraditionalonlinesearchtool(e.g.,GoogleAIoverviews).

Notes:ItwasnotpossibletodeterminewithcertaintywhetherparticipantswereconsistentlyreferringtoAIsearchchatbotinteractionsthatretrievedreal-timeonlineinformation,orthosethatreliedsolelyontheAI

chatbot’spre-trainedmodeldata.Thisambiguityarisesbecausebothtypesofinteractionsareencompassedwithinparticipants’typicalsearchbehaviours,whichmadeitdifficulttodistinguishbetweentheminthestudy.

ThisresearchwascommissionedforOfcomtounderstandthemotivationsandconcernsofeverydayuserswhenusingsearchtools,whichcanhelpinformOfcom’sregulatorywork.However,giventhatGenAIsearch

broadrangeofproducts,thisresearchdoesnottakeaviewonwhetheranynamedservicesorparticulartypesofservicesreferencedinourmaterialscouldbeconsidereda'searchservice'withinthemeaningoftheUKOnlineSafetyAct2023(OSA),orwhethertheyfallwithinthescopeoftheOSA.

coversa

TheopinionsexpressedbyparticipantsaretheirownanddonotrepresentOfcom’sviewsorpositions.

5YOUGOV

searchexperiences.

ResearchObjectives

ResearchQuestions

Throughthisresearch,weaimedtoexploreandunderstand:

?Currentbehavioursandpreferencesofuserssearchingonline;

?Howbehavioursandpreferencesduringonlinesearchesmaydifferfordifferentsubjectmatters;

?Whatfactorsinfluencepeople’sdecisionstochooseGenAIsearchversustraditionalsearchforfindinginformationonline(andvice

versa);

?WhatnegativeexperiencesareencounteredwhilstusingGenAIsearch,includingonbothAIsearchchatbotsandAIsearch

summaries;

?People’sgeneralattitudesandopinionsonusingAIsearchchatbotstosearchforinformationonline,includingwhethertheyareawareofanysafetymechanisms;and

?AnyunpromptedviewsorconcernsthatpeopleraiseaboutGenAI

6YOUGOV

MethodologyandGuideStructure

Methodology

YouGovQualitativeconducted24onlinein-depthinterviewslastingbetween30to45minutesbetween17thofJuneand3rdofJuly2025.

Sampleframeincluded:

oAmixofages(17+),genders,socialgrades,regionsandeducationlevels

oInclusionofethnicminorities(5outof24)

oEqualsplitbyself-reporteddigitalconfidence–veryhigh,highandmedium

oMixofbrowserandsearchtooluse

oMixoffavourabilitytowardsandunderstandingofAI

Theanalysiswasundertakenbyresearcherswhoconductedthefieldwork.Datawasanalysedusingthematicanalysisandkeyrecruitmentcriteria–demographicsand

levelofdigitalconfidence.Therewereonlyafewareaswheredifferenceswere

observedbetweenthesesubgroups–theyhavebeenstatedinthereportwhereverapplicable.Findingsthatarediscussedwithoutsubgroupdifferencesrefertoviewsthatwereexpressedbythewholesample.

InterviewStructure

Priortotheinterview,participantscompleteda

pre-taskbyconductingfiveonlinesearchesusingthetoolstheynormallyaccess.Threeofthe

searcheswerespecifiedandtwowereoftheirchoice.Thespecifiedtaskslisted3-4questions,andparticipantswereaskedtoselectoneto

research.Thespecifiedtopicscovered:

oHealthandwellbeing

oFinanceandmoneymanagement

oNewsandcurrentaffairs

Theinterviewsstartedwithparticipantsreflectingontheirpre-task,followedbyadiscussionaboutparticipants'preferredonlinesearchingtools,

experienceandviewsontraditionalandGenAIsearch.

7

2ExecutiveSummary

8YOUGOV

?ParticipantsexpectedthatGenAIsearchexperienceswillbecomemorepopularwithtimeasthetechnologydevelopsandasGenAIisincreasinglyintegratedintotools.Theybelievedthatregulation,greatertransparencyofdatauseandgrowingfamiliarityofGenAI

wouldalsohaveanimpactonadoptionofGenAItoolsforbothsearchandnon-searchpurposes.

Theonlineworldisrapidlychanging,especiallyconsideringtheemergenceoftechnologiessuchasGenAIsearchtools.Consumers’

attitudesandbehaviourswhensearchingforinformationonlinearelikelytochangeasaresultofthesecontinuoustechnologicaldevelopments.Basedonthe24UKinternetusersinterviewedinourqualitativestudyinJuly2025,weobservedthat:

?MostparticipantsreportedusingGoogletofindinformationonlineduetofamiliarityandperceivedreliability.SomeparticipantsalsomentionedusingAIsearchchatbotslikeChatGPTforspecifictasksandsearchqueries.

?ParticipantshadstartedtouseGoogleAIoverviewsmorefrequently,withtheexceptionofsomewhodidnottrustAIsearchsummariesingeneral.

?Participantsoptedfordifferentsearchexperiencesbasedonthecomplexityandnatureofthequery.GenAIsearchtoolswere

favouredforlow-stakesquerieswheretheconsequencesofinaccuracywerelesssignificant,suchasquickandstraightforwardtasks.Ontheotherhand,traditionalsearchwasmorelikelyusedforhigh-stakesqueries,suchashealthandfinancematters,wherethe

consequencesofinaccuracyweremoresignificant.ParticipantswerecautiousaboutAI-generatedcontentandreportedthattheywouldoftenverifyitusingresultsfromtraditionalsearchtools.

?WhilstsearchingonAIsearchchatbotswaspopularamongsomeparticipantsduetoitsperceivedspeedandefficiency,concernsaboutaccuracy,transparency,dataprivacy,andethicalimplicationswereraised.

?Generally,participantslackedawarenessofsafetymechanismsduringGenAIsearchexperiences,butsaidtheywouldexpectthesetobeinplace,especiallytoprotectuserswhoarechildrenorwhohavementalhealthconditions.

9

3Pre-task–results

10YOUGOV

Priortotheinterviews,participantswereaskedtoconductarangeofonlinesearchtasksastheywouldnormallydo

Aspartofthepre-taskandbeforetheonlineinterviewstookplace,participantsconductedonlinesearchesacrossfivedaysonarangeoftopicsoftheirchoice,alongsidesearchesonspecifiedtopics.

?Day1一Healthandwellbeing.Participantsweregiventheoptiontochooseoneofthefollowingquestions:

oWhataresomeeffectivetechniquesformanaginganxietyindailylife?Whichmethodsseemtobethemostpopularandwhy?

oWhatarethehealthbenefitsoftheMediterraneandiet?WhatarethechallengestopracticingitforthoselivingintheUK?

oWhataretheprosandconsofusingfitnesstrackerstomonitorhealth?Inyourview,whataresomeofthemostpopularfitnesstrackerchoicesintheUKforareasonablebudget?

oWhataretheprosandconsoftheMediterraneanandKetogenicdietsintermsoftheirhealthbenefits,risks,andsuitabilityfordifferentlifestyles?

?Day2一Financeandmoneymanagement.Participantsweregiventheoptiontochooseoneofthefollowingquestions:

oHowdoesacreditscoreworkandhowcanIimprovemine?

oWhatwouldbeagoodapproachtostartinvestinginstocksandshares?WhatthingsdoIneedtobeawareofbeforedoingso?

oFindsomedifferentstrategiesforimprovingcreditscoresovertime.Whataretheprosandconsofeach?

oFindsomedifferentapproachestogettingstartedinvestinginstocksandsharesintheUK.Whataretheprosandconsofindexfunds,individualstocksandrobo-advisors?

?Day3一Freechoice

?Day4一Freechoice

?Day5一Newsandcurrentaffairs.Participantsweregiventheoptiontochooseoneofthefollowingquestions:

oWhatarethemainnewsheadlinesintheUKtoday?

oWhatisthebiggestnewsheadlineinUK’ssportsnewstoday?

oWhatisthebiggestnewsheadlineinUK’sentertainmentnewstoday?

11YOUGOV

Participantsweremorevigilantwhensearchingforinformationtheyconsideredtobe‘highstakes’

Theapproachtakentosearchestendedtovaryslightlyforeachparticipant.However,manyweregenerallymorecautiouswhensearchingforinformationthattheyperceivedas‘highstakes’.Thisiswhentheriskofanadverseoutcomefrominaccuracywouldbehigher,suchasfortopicsrelatingtofinance,moneymanagement,news,currentaffairs,andhealthandwellbeing,asopposedto'lowstakes'searcheswheretheriskofanadverseoutcomefrominaccuracywouldbelower,includingtopicslikehobbiesorleisure.

‘Highstakes’information

Actions:

?IfusingGenAIsearchtools,participantstendedtocheckthesourcestoensuretheywerelegitimateand/orofficialbeforetrustingtheinformationgeneratedbyAIsearchchatbotsorAIsearchsummaries.

?Ifoptingfortraditionalsearch,participantsmostlyclickedonfamiliarlinksandsourcestheyperceivedtobetrustworthy.

?Outsideofthat,whilstusingbothGenAIandtraditionalsearchtools,participantstrustedtheinformationifitalignedwiththeirownpriorknowledgeofthetopic.

‘Lowstakes’information

Actions:

?TherewasahighertendencytotrusttheinformationprovidedbyAIsearchchatbotsorAIsearchsummarieswith‘lowstakes’information,althoughsomereportedtheywerecautiousstill.

?Whilstthelegitimacyofsourceswasimportant,participantsweremoreopentoaccessinglesser-knownsourcesforlowstakestopics,alongsideblogsorsocialmediasiteswhichtheywouldnotnormallyuseforhighstakessearches(e.g.,TikTok,Reddit),duetoperceivingthesesourcesaslesslegitimate.However,theywouldoftensense-checkthe

informationeitherbyconfirmingitalignedwiththeirunderstandingorbycross-checkingitwithothersources.

12YOUGOV

Thefree-choicesearchwasconductedonavarietyof

topicsandpredominantlythroughGoogle’straditional

Thefree-choicesearchwasconductedonarangeoftopics,including:

?Travel–e.g.,researchingaholidaydestinationandactivities.

?Healthandwellbeing–‘Whataretheeffectsofvaping?’,‘WhatisPyloricStenosis,itssymptomsanditstreatment?’,‘Agymroutineforabeginner’.

?Hobbies–e.g.,‘Howdoyougrowcitrusinthe

UK?’,‘Howtomakealemontart?’,‘Thebeste-readersfor2025’,‘WhatistheclubWorldCup?’.

?TVandmovies–e.g.,researchinginformationaboutTVshowsormovies.

?Other–‘know-how’questions,suchasaguidetousingasoftwareordevelopingcertainskills.

search,withmostreportingpositivesearchexperiences

?Whenconductingtheirfree-choicesearch,participantsreliedmostly

ontraditionalGooglesearchaswellasAIsearchsummariesinGooglesearch,alongsidewebsitesthattheytrustedandreviewplatforms

(e.g.,Reddit,TripAdvisor,BBCNews).

?TheyfoundtheGooglesearchresultstobemostlytrustworthyandnotedthatduringtraditionalsearchexperiences,theyweremostly

presentedwithrankedwebpagesthattheyfeltwerelegitimate.

?AfewparticipantswereconcernedabouttheinformationgiveninAIsearchsummariesifthesourcelinksincludedwebpagesthattheydidnotfullytrust(e.g.,Wikipedia).Whilsttheydidnotreportany

inaccuracies,thismadethemlesslikelytotrustit.

?AminoritywerecautiousaboutresultsfromtraditionalGooglesearchontopicsthatwereconsidered‘lowstakes’,suchashobbiesand

leisure.Itwasfeltthatsearchresultsonthesetopicstendedtoshowunofficialsources(e.g.,blogs)thatwereseenaspotentiallybiased.Asaresult,theseparticipantsnotedthattheywouldbemorelikelyto

compareinformationacrossdifferentsourcesbeforetrustingit.

13YOUGOV

QuestionsthatparticipantssearchedusingAIsearchchatbotsincluded:

?Healthandwellbeing–e.g.,‘Ismatchagoodforyou?’and‘Howtohealelbowtendons?’

?Hobbiesandleisure–e.g.,‘MostpopularXboxGamePassGames’,‘Movieswith

IMBDratingabove9in2025’and‘HowcanIgetinvolvedlocallyinbowls’?

?Technicalquestions–e.g.,reasonsaspecificExcelformulaisnotworking.

informationtheyaccessedpreviouslyknowledge,leadingthemtotrustit.

ThosewhousedAIsearchchatbotstendedtoappreciatethelevelofdetailandperceivedaccuracy

?AminorityusedAIsearchchatbots,suchasChatGPTandGemini.ThechoicetousetheseAIsearchchatbotsinthepre-taskwaspredominantlybasedontheirhabits.

?TheexperienceswithAIsearchchatbotsweremostlypositive:participantstrustedtheinformationastheylikedthelevelofdetailandfelttheinformationwasaccurate,whichwasmostlybasedontheirownunderstandingof

thesubject.Onlyoneparticipantmentionedthatthe

informationprovidedbytheAIsearchchatbotwaslimited,leadingthemtodoubtitsaccuracy,andtheytherefore

turnedtotraditionalsearchtosupplementtheAIchatbotsearch.

?Thosewhosearchedforinformationrelatedtohealth,notedthatitappearedtobeinlinewithother

and/ortheir

14YOUGOV

Participantsfeltmostlyconfidentinonlinesearchresultsonthetopicofhealthandwellbeing

?Whencarryingoutanonlinesearchonthetopicofhealthand

wellbeing,participantsmostfrequentlyusedGooglesearch,whilstoneparticipantusedDuckDuckGo.

?ManyparticipantspaidattentiontoAIsearchsummariessuchasthe

GoogleAIOverviewandfeltthattheyweregenerallyhelpfuland

accurate,basedontheirownunderstandingofthesubject.Therewasatendencytotrustthemifthecitedlinkswereseenaslegitimate.

However,despitebeinghelpful,theAIsearchsummarywasnot

consideredsufficient–theyoftencheckedtheinitialresultsandcitedlinkstoaccessoriginalsources.

?ThosewhohadpreviousnegativeexperienceswithAIsearch

summariesshowinginaccurateinformationtendedtoignorethem

altogetherandinsteadwentdirectlytothewebpageslistedonthefirstresultspage.

?AminoritywhousedAIsearchchatbotssuchasChatGPTtosearch

abouthealthandwellbeingnotedthattheycouldrelyontheinformationprovided,astheyperceivedthetopictobenon-controversialandtheresultsalignedwiththeirunderstandingofthesubject.However,theystillpaidattentiontothesourcesusedandwantedtoknowthatthe

informationtheyreceivedwasreliable.

“Idon'ttakewordforwordtheinfosummarisedbyAI,however,ifthe

linkswheretheinformationwasacquiredfromarereputable,thenI

don'tmindbelievingit.”–Male,37,Veryhighdigitalconfidence

“TheaccuracyoftheAIsearchresultwasthemainconcern,butmoreforthereputationithasinbeinginaccurate,ratherthanhavingwitnessedinaccuracies.”–Female,26,Mediumdigitalconfidence

“ItrusttheinformationIamgettingafairamount,howeverIwould

liketoknowwheretheyaregettingtheinformationfrom…On

ChatGPTespeciallyitneverusuallysayswheretheinformationis

comingfrom.”–Male,20,Mediumdigitalconfidence

“Idotrust[ChatGPT]asthesourcesoftheinformationwerepresented,andIcouldvisitthemtoassesstheirtrustworthiness.

[Themediterraneandiet]isalsonotacontroversialtopic,soIfelt

safeinthisregardanddidn'tfeeltheneedtointerrogatethe

sources.”–Male,40,Highdigitalconfidence

15YOUGOV

Therewasmorecautionaboutthesourcewhensearchingforinformationonnewsandfinance

?Ingeneral,participantsweremorecautiousaboutsourceswhenitcameto

searchingforinformationinrelationtofinanceandmoneymanagementornewsandcurrentaffairs.Theytendedtocheckthelinkstoassessitstrustworthiness.

?Whensearchingforinformationonthesetopics,mostparticipantsusedtraditional

search–mostlyGoogle,withoneparticipantusingBingforcomparisonwithGooglesearchandoneusingDuckDuckGo.Mosttendedtonavigateandclickonpagesthattheyalreadyknewandtrusted,ratherthanlookthroughavarietyofresults.Amongsome,therewasatendencytodirectlyvisitwebsitestheywerefamiliarwithand

foundtobetrustworthy(e.g.,MoneySavingExpert,BBCNews).

?AmongparticipantswhousedbothGenAIsearchandtraditionalsearch,mostnotedthattheGenAIsearchresultswereofteninlinewiththeresultstheyobtainedvia

traditionalsearch.However,aminoritynotedthatthenewsheadlinesshownin

traditionalsearchdifferedfromtheheadlinesinAIsearchchatbots.SomealsofoundthattheseAIsearchchatbotsusedsourcestheywereunfamiliarwith,leadingthemtotrusttheresultsless.

?Therewasasenseamongmanyparticipantsthattraditionalsearchtoolstendedtobringupresultsfromsitesthattheyalreadytrusted,makingthesearchprocess

sources.

easier,astheydidnotneedtodoafurthersearchorverifyinformationfromdifferent

“Thelinksitprovidedgavemeaccesstobank

sitesIhaveusedandtrustedbefore.Imightbe

moresuspiciousofAIanswerswhichwerealso

listed.Buttheseseemedconsistenttotheother

answersIhadreceived.”–Male,67,Veryhigh

digitalconfidence

“IlookedonnewswebsitesanddidaGoogle

searchtoo…h(huán)erewereplentyofnewswebsites

tolookatanditwaseasytoseeallthestories

beingreportedonthefirstpage.Itwaseasyto

searchfornews.”–Female,39,Mediumdigital

confidence

“IusedGoogle,LloydsBank,NatWestBank

andMartinLewiswebsitestofindoutthe

informationbecauseIfeelmorecomfortablein

usingthesewebsitesasIfeeltheyarevery

informativeandestablished.”–Female,71,Very

highdigitalconfidence

16

4Onlinein-depthinterviews–

results

17

4.1Preferredonline

search

experience

18YOUGOV

Searchtoolpreferences

?Googlewaspreferredbecauseitsinterfacewasconsideredmoreuser-friendly,anditstraditionalsearchfunctionality

wasfelttogeneratemoreaccurateandrelevantresults.

However,thisviewwasnotunanimous,andaminorityfoundthatGooglesometimesincludeduntrustworthylinks.

?OneparticipantusedDuckDuckGobecauseitwasfelttohavefeweradverts,comparedtoothersearchtools.

?DespitegenerallyhighlevelsofpreferenceforGoogle,a

minorityfeltthattraditionalGooglesearchdidnotalways

bringacomprehensivelistofresults,soiftheywantedto

accessawiderrangeofwebpages,theymentionedbeinglikelytodoanadditionalsearchviaothertools,suchasBing.

MostparticipantsfavouredusingGoogleastheyperceivedittobeuser-friendlyandfeltitproducedrelevantresults

Useofsearchtoolsandbrowsers

?ThevastmajorityofparticipantsusedtraditionalGooglesearchin

GoogleChromeandotherbrowsers,suchasMozillaFirefoxandMicrosoftEdge,wheretheysetitasadefaulthomepageornavigatedtotheweb

pagemanuallyonopeningthebrowser.AminorityusedBing,duetohavingtheMicrosoftEdgebrowserinstalledwithBingasthedefault

searchtool.OnepersonusedDuckDuckGointheMicrosoftEdge

browser.Youngerparticipants(under30yearsold)tendedtomostlyuseGooglesearch,regardlessofthebrowsersinstalled.

?Aminoritywhousedmorethanonesearchtooldidsoduetohaving

differentbrowsersinstalledontheirdeviceswiththedefaultsearch

settingsactivated.Thismeantthattheuseofsearchtoolsdependedon

thedevicetheywereusing,withafewparticipantsoccasionallycomparingresultsacrossthesearchtoolstoaccessalargervarietyofinformation.

“It'sfamiliarity[withGooglesearch]…Itgivesmethe

answersIwantbecauseitknowsenoughaboutmysearch

history…youstillgotBingasaseparateenginesearch

engine…[but]itdoesn'tgivethemorecuratedresults.”–

Male,40,Highdigitalconfidence

“Googlehelpstofind…informationinlesstime…but

atthesametimeitdoesn’t[mean]theinformationthat

yougetisaccurateanditistrue.Sosometimesyou

find[in]Googlesomeinformationwhichiscompletely

fake.”–Male,53,Mediumdigitalconfidence

“I'msousedtohowGooglelooks

andhowitappearsthatanything

elsesortofseemslikelessuser

friendlytome.”–Female,29,

Highdigitalconfidence

19YOUGOV

WhilstAIsearchchatbotswerehelpfulingenerating

“It'sveryquicklyleaningtowardsmoreuseofChatGPT

becausealotofwhatIsearchsometimeswillbejustdata

stuff.AndIneedthedataquickly…Idon'twanttoscroll

throughlotsofsitesforit.IfIjustwantsomequickfacts,

basicallyit'saloteasierformetogetthatinformation.”–

Male,37,Highdigitalconfidence

“ChatGPT–sometimesIfeeltheymakeuptheirsources

becausewhenItrytofind[them]directly,likewhenthey

givemesourcesorwheretheygottheinformationfrom,it's

wrongorit'snotcorrect.”–Female,24,Veryhighdigital

confidence

“WhatIlikeaboutthePerplexityisintermsofthe

informationthatitgivesyou,youcanseeyourquestionthat

you'vetypedinandthenobviouslywhattheresponseis,

butalsoyou'vegotlittlesubheadingsunderneathtoshow

youwhereitgotitssourcesfrom,inordertogiveyouthat

answer.”–Female,70,Highdigitalconfidence

resultsforspecificsearches,theywerenotalwaystrusted

UseofGenAIsearch

?AIsearchchatbotswereusedbyasmallerproportionofparticipants,comparedtotheuseoftraditionalonlinesearchtools.

?ChatGPTwasthemostfrequentlyused,withasmallernumberofparticipantsusingGemini,Copilot,PerplexityandGrok.ThereasonsforusingvariousAIchatbotsmostlydependedonfamiliarityandsuitabilityforotherpurposes,mostlyoutsideofonlinesearch(e.g.,iftheyusedtheseAIchatbotsforworkpurposesorimagecreation,theyusedthemforonlinesearchesaswell).

?AIsearchchatbotswereconsideredmostusefulforspecificquestionsthatrequireahighlevelofdetail,whichwouldotherwiserequirealengthiersearchviatraditionalsearchtools.

?AIsearchchatbotswerefelttoretrieveinformationquicklyandwereabletoanswerspecificquestions.However,ifparticipantswantedtobesureaboutthesourceorwantedtofurtherverif