AIANDTHEFUTUREOFNEWS

OCTOBER2025

GenerativeAIandNewsReport2025:

HowPeopleThinkAboutAI’sRolein

JournalismandSociety

FelixSimon,RasmusKleisNielsen,andRichardFletcher

ThomsonReutersFoundation

Contents

AbouttheAuthors

Acknowledgements

4

4

ExecutiveSummaryandKeyFindings

5

Introduction

10

Methodology

12

1.PublicAwarenessandUseof

GenerativeAI

14

2.AI-GeneratedSearchAnswers

30

3.PublicAssessmentofScaleand

ImpactofAIUseinDifferentSectors

37

4.PublicOpinionsontheUseof

GenerativeAIinJournalismandNews

44

Conclusion57

References59

ThisreportispublishedbytheReuters

InstitutefortheStudyofJournalismaspartofourworkonAIandtheFutureofNews,supportedbyfundingfromTheNational

CenterforAIinSocietyinDenmark.The

AIandtheFutureofNewsprojectismadepossiblebycorefundingfromtheThomsonReutersFoundation.

DOI:10.60625/risj-5bjv-yt69

THEREUTERSINSTITUTEFORTHESTUDYOFJOURNALISM

4

AbouttheAuthors

DrFelixM.SimonistheResearchFellowinAIandNewsattheReutersInstitutefortheStudyofJournalismandaresearchassociateattheOxfordInternetInstituteattheUniversityof

Oxford.Hisresearchlooksattheimplicationsofachangingnewsandinformationenvironmentfordemocraticdiscourseandthefunctioningofdemocracy,withaspecialfocusonartificial

intelligence.

ProfessorRasmusKleisNielsenisaprofessorattheDepartmentofCommunicationofthe

UniversityofCopenhagenandaseniorresearchassociateattheReutersInstitutefortheStudyofJournalism.Hisresearchisfocusedonthechangingroleofnewsandmediainoursocieties.Hehaswrittenextensivelyaboutjournalism,digitalmedia,thebusinessofnews,political

communication,misinformation,andrelatedtopics.

DrRichardFletcherisDirectorofResearchattheReutersInstitutefortheStudyof

Journalism.Heisprimarilyinterestedinglobaltrendsindigitalnewsconsumption,theuseofsocialmediabyjournalistsandnewsorganisations,and,morebroadly,therelationshipbetweencomputer-basedtechnologiesandjournalism.

Acknowledgements

WewouldliketothankCaryhsInnes,XhoanaBeqiri,andtherestoftheteamatYouGovfor

theirworkonfieldingthesurvey.Wearegratefultotheothermembersoftheresearchand

editorialteamatRISJ,aswellasMichelleDisser,fortheirinputonthequestionnaireand

helpwiththeinterpretationoftheresults.WearealsothankfultoMasaharuBan,Gretel

Kahn,PriscilleBiehlmann,MagnusBredsdorff,andTaniaMontalvofortheiradviceonthe

translations,andtoMitaliMukherjee,KateHanneford-Smith,AlexReid,andRebeccaEdwardsforhelpingtomovethisprojectforwardandkeepingusontrack.

GENERATIVEAIANDNEWSREPORT2025:HOWPEOPLETHINKABOUTAI’SROLEINJOURNALISMANDSOCIETY

5

ExecutiveSummaryandKeyFindings

Oursurveyinsixcountries(Argentina,Denmark,France,Japan,theUK,andtheUS)exploredhowpeopleusegenerativeAIintheireverydaylives,whattheythinkitsimpactwillbeondifferent

areasofsociety,andwhattheythinkaboutitsuseinnewsandjournalismspecifically.Itisa

follow-uptothesurveythatweconductedinthesamesixcountriesin2024(FletcherandNielsen2024).Basedontheresultsofthissurvey,andthepreviousone,wefindthefollowing.

FindingsontheawarenessanduseofgenerativeAI

Thepublic’suseofgenerativeAIhasincreasedsubstantiallyinthelastyear.TheproportionwhosaytheyhaveeverusedastandalonegenerativeAIsystemsuchasChatGPTjumped

from40%to61%,andweeklyusagenearlydoubledfrom18%to34%.ChatGPTisthesinglemostpopularAIsystemwith22%weeklyusage,thoughadoptionvariessignificantlybyage.

Information-seekinghasbecometheprimaryuse-caseforAI,morethandoublingto24%weeklyusageandsurpassingmediacreation,whilespecialiseduseslikenewsconsumptionremainlimitedat6%.Trustisconcentratedamongmajorbrands,withChatGPTleadingthefieldagain,thoughmostusersremainoccasionalratherthanregularadopters.

Inmoredetail,wefind:

?Awarenesshassurgedyear-on-year.Theshareofpeopleacrosscountrieswhohaveheardofatleastoneof13AItoolsrosefrom78%(2024)to90%(2025);only10%saytheyhaveheardofnone.ChatGPTremainsthemostwell-knownbrand,withpronouncedcountryvariationforothers.

?Useisexpandingrapidly–especiallyweeklyuse–thoughnotuniformly.Across

countries,theproportionofpeoplewhosaytheyhaveeverusedanyAIsystemrosefrom40%(2024)to61%(2025);weeklyusenearlydoubledfrom18%to34%.

?Again,ChatGPTdominatesactiveuse.Onaverage,22%reportusingChatGPTinthelastweek,wellaheadofothertools.SinceMay2024,thecoreuserbaseforeverymajorsystemhasroughlydoubled.Yetmostpeoplearenotregularusers:forthefourmostpopular

systems,manyusethemonlymonthlyoronceortwiceever,andlargeshareshaveneverusedornotheardofthem.

?UseofgenerativeAIskewsyounger.Inthe18–24category,59%usedanygenerativeAIinthelastweek,asopposedto20%amongthoseaged55andover,althoughthisagegapisdrivenmainlybyChatGPT.AgedifferencesaresmallerforGoogle’sGemini,andminimalforCopilot,MetaAI,andGrok–likelybecausethey’reembeddedinwidelyusedproducts.

?Information-seekingisnowtheleaduse-caseacrosscountries.WeeklyuseofAIfor

gettinginformationmorethandoubled(from11%to24%),overtakingcreatingmedia(up7percentagepointsto21%).Socialinteractionisnascentbutnotable(7%overall;13%of18–24sversus4%of55+).

THEREUTERSINSTITUTEFORTHESTUDYOFJOURNALISM

6

?Whilespecificinformation-retrievaltasksarebroadening,withansweringfactual

questionsrisingfrom6%to11%,mediacreationremainsnicheexceptforimages.Imagegenerationrosefrom5%to9%weekly;video(3%)andaudio(2%)werebasicallyflat;

codingusewasalsoflat,suggestingearlyadopterswerealreadyonboardin2024.

?GettingnewsviaagenerativeAIsystemhasdoubledbutremainsaminorityactivity.

Weeklyuserosefrom3%to6%,drivenmainlybyusersinJapanandArgentina;itis

strongestinArgentinaandtheUSAandamong18–24s(8%)comparedto55+(5%),

andhigherfordegree-holders.AmongAI-for-newsusers,‘latestnews’(54%)andhelp

withsummarising,evaluating,orrewritingaremostcommon.YoungerusersleanmoretowardsusingAItohelpthemnavigatethenews:48%of18–24susedAItomakeastoryeasiertounderstandcomparedto27%of55+(a21percentagepointgap).

?Trustisconcentratedinafewbrands.Onaverage,29%saytheytrustChatGPT,aheadofGemini(18%),Copilot(12%),andMetaAI(12%);mostotherbrandsaretrustedbyfewerthan10%,mostlyduetolowawareness.InmostcountriesChatGPTismoretrustedthandistrusted,withtheexceptionoftheUK.

FindingsonpublicviewsonAI-generatedsearchanswers

AI-generatedsearchanswershavebecomecommonplaceinthesixcountriesstudied.User

engagementwiththeseAIanswersismixed,asonlyonethirdsaytheyconsistentlyclick

throughtosourcelinkswhile28%rarelyorneverdoso,withyoungerusersmorelikelyto

engagefurther.Trustlevelsaremoderateat50%amongthosewhoencounterAIanswers,withusersvaluingtheirspeedandinformationaggregationcapabilities–althoughtrustbecomesconditionalinhigh-stakesareaslikehealthandpolitics,wheremanyverifyanswersthroughtraditionalsources.

Inmoredetail,wefind:

?PeopleregularlyseeAI-generatedsearchanswers.Acrosscountries,54%saytheysaw

anAI-generatedanswertooneoftheirsearchesinthelastweek.Reportedweekly

encountersarehighestinArgentina(70%),followedbytheUK(64%)andtheUSA(61%),andlowestinFrance(29%),whereGoogle’sAIoverviewfeaturehadnotbeenrolledoutatthetimeoffieldwork.

?Self-reportedclick-throughbehaviourismixed.AmongthosewhosawAIanswers,aboutonethird(33%)saytheyalwaysoroftenclicklinksintheoverview,37%saytheydososometimes,and28%rarelyorneverclickthrough.Youngerpeoplearemorelikelytosaytheyclickthrough,althoughitisimportanttoacknowledgethatself-reportedbehaviourmaydifferfromactualbehaviourforsomerespondents.

?AmongthosewhohaveencounteredAIanswers,50%saytheytrustthem,witha

significantminorityremainingneutral.Whilethereislittledifferencebygender,youngeradultsshowslightlymoreoutrighttrust.Respondentsemphasisedtheirspeedand

convenienceandthefactthatAIaggregatesvastamountsofinformationasreasonstotrustthem.

GENERATIVEAIANDNEWSREPORT2025:HOWPEOPLETHINKABOUTAI’SROLEINJOURNALISMANDSOCIETY

7

?TrustinAIanswersinsearchisconditional,especiallyinhigh-stakesdomains.Many

saytheyverifyanswers,especiallyforhealthorpolitics,treatingAIasafirstpassbeforecheckingnon-AIsources.

FindingsonpublicopinionabouttheuseofgenerativeAIindifferentsectors

ThepublicwidelyperceivesgenerativeAIasalreadyprevalentacrosssectors,with41%

believingit’sused‘a(chǎn)lwaysoroften’onaverage,risingto67%forsearchengines,and51%for

newsmedia.PublicsentimenttowardAI’simpactismixedbutslightlyoptimisticregarding

howAIwillimprovetheirinteractionswithvarioussectors,thoughthisvariessignificantlybydomain.Whileoptimistsoutnumberpessimistsforsectorslikehealthcare,science,andsearchengines,thereverseistruefornewsmedia,government,andespeciallypoliticiansandpoliticalparties.Personalversussocietalexpectationsdiverge,withoptimistsoutnumberingpessimistsregardingindividualbenefitsinfourofsixcountries,butpessimistsdominatewhenconsideringsocietalimpactinthreecountries,includingtheUnitedStates,withwomenconsistently

showinglowerexpectationsforbothpersonalandsocietalbenefitsfromAI.

Inmoredetail,wefind:

?ThereisawidespreadpublicperceptionthatgenerativeAIisalreadyeverywhere,atleasttosomeextent–thenumberofrespondentswhobelievegenerativeAIisusedalwaysoroftenindifferentsectorsis41%onaverageacrosscountriesandsectors,farexceeding

thosewhosayitisusedrarelyornever(15%).Thefigureismuchhigherfornewsmedia(51%)and,especially,socialmediacompanies(68%)andsearchenginecompanies(67%).

?Askedhowmuchbetterorworsetheythinkdifferentactors’useofgenerativeAIwillmakepeople’sexperienceofinteractingwiththem,amajorityofthepublicexpressajudgement,evenasmanyanswer‘neitherbetternorworse’orsimplythatthey

don’tknow.Onaverageacrossallcountriesandsectors,29%areoptimisticand22%pessimistic.

?Generally,therearemoreoptimiststhanpessimists–especiallyforsectorslike

healthcare,science,andsearchengines.Onlythreesectorsseethepessimistsoutnumbertheoptimists–newsmedia,government,and,especially,politiciansandpoliticalparties.

?ComparedtotheaverageshareofrespondentswhosaytheybelievegenerativeAIis

usedalwaysorofteninagivensectorandtheaverageshareofrespondentswhosay

theybelievetheuseofgenerativeAIwillmaketheirexperienceofinteractingwitha

givensectorbetter,wecanidentifysectorsthatstandout–sectorswherepeoplethinkgenerativeAIisparticularlywidelyusedandwheremanyexpectthiswillimprovetheirexperienceincludesearchandsocialmedia.Sectorswhereexpectationsareparticularlylowincludegovernmentuseandusebypoliticiansandpoliticalparties.

THEREUTERSINSTITUTEFORTHESTUDYOFJOURNALISM

8

?AskediftheythinkthatgenerativeAIwillmaketheirlifebetterorworse,andwhetheritwillmakesocietybetterorworse,theoptimistsoutnumberthepessimistsinfourofthesixcountriescoveredintermsofpeople’sownlives,butwhenitcomestosocietytherearesignificantlymorepessimiststhanoptimistsinthreeofthesixcountries,includingtheUS.

?ExpectationsaroundAIalsodifferalongsocio-economicvariables.FemalerespondentsaresignificantlylesslikelytoexpectthatgenerativeAIwillmaketheirlivesbetter.Theyarealsosignificantlylesslikelytosaytheyexpectittomakesocietybetter,andmore

likelytoexpectitwillmakeitworse.

FindingsonpublicopinionabouttheuseofgenerativeAIinjournalismandnews

Asignificant‘comfortgap’existsbetweenAI-andhuman-lednewsproduction,withonly12%ofrespondentscomfortablewithfullyAI-generatednewscomparedto62%forentirelyhuman-madecontent,althoughacceptanceincreaseswithhumanoversightandwhenhumanslead

withAIassistance.Publiccomfortalignswithperceivedusagepatterns,aspeoplearemost

acceptingofback-endapplicationslikegrammareditingandtranslation,whileshowingmoreresistancetofront-facinguseslikeartificialpresenters.ExpectationsaboutAI’simpactonnewsremainmixed,withpeopleanticipatingbenefitslikecheaperproductionandfasterupdates

whilealsohavingconcernsaboutreducedtransparencyandtrustworthiness,whereasonly33%believejournalistsroutinelycheckAIoutputsbeforepublication.DespiteagrowingawarenessofAIuseinnews,mostpeople(60%)don’tyetregularlyencounteraudience-facingAIfeatures,andonly19%seeAIlabellingdaily,despite77%consumingnewsdaily,suggestingadisconnectbetweenAIimplementationandpublicvisibility.

Inmoredetail,wefind:

?Thereisaclear‘comfortgap’betweenAI-andhuman-lednews.Onaverage,only12%arecomfortablewithnewsmadeentirelybyAI;thisrisesto21%witha‘humanintheloop’,43%whenahumanleadswithsomeAIhelp,and62%forentirelyhuman-madenews(anincreaseof4percentagepointssince2024).Thisgapcanbefoundacrossdemographicsandcountries.

?Peoplecontinuetobemostcomfortablewithback-endusesofAIinnews.Comfort

ishigherforeditingspellingandgrammar(55%)ortranslation(53%),andlowerfor

rewritingcontentfordifferentaudiences(30%),creatingarealisticimagewhennophotoexists(26%),andartificialpresenter/authors(19%).

?MorepeoplethinknewsroomsuseAI.Theproportionofpeoplesayingjournalists‘a(chǎn)lways’or‘often’usegenerativeAIareupbyatleast3percentagepointsacrosstasks.

GENERATIVEAIANDNEWSREPORT2025:HOWPEOPLETHINKABOUTAI’SROLEINJOURNALISMANDSOCIETY

9

?People’sperceptionsofwhereAIisusedbroadlyalignwiththeirownpreferencesforwhereitshouldbeused.PeoplebelievejournalistsmostoftenuseAIfortasksthey’remostcomfortablewith(e.g.grammarediting:55%arecomfortableand51%thinkit’susedregularly)andrarelyfortasksthey’releastcomfortablewith(artificialpresenters:

19%arecomfortableand20%thinkit’susedregularly).

?PeoplecontinuetohavemixedexpectationsaboutwhatAIwilldotonews.Aswith

lastyear,manyassumeAIwillmakenewscheapertoproduce(+39percentagepoint

differencebetweenthosethatsaidmoreandthosethatsaidless)andmoreuptodate

(+22),butlesstransparent(-8)andlesstrustworthy(-19);viewshavehardenedsince

2024(nodecreases).PeopleinJapanandArgentinaaregenerallymorepositiveaboutAI’simpactonnewsqualities,whilepeopleintheUKaremorenegative.

?PeoplehavelimitedconfidenceinroutinehumanoversightofAIinnews.Only33%thinkjournalists‘a(chǎn)lways’or‘often’checkAIoutputsbeforepublishing–thisishigherinJapan(42%)andArgentina(44%),andlowerintheUK(25%)–withsmallshiftssince2024.

Trustinnewsstronglycorrelateswiththis:57%ofthosewho‘stronglytrust’newsthinksuchcheckshappen,asopposedtojust19%amongstrongdistrusters.

?Thepublicacrosscountriesexpectsresponsibleusetovarybyoutlet.Forty-threepercentforeseelargedifferencesinhowresponsibledifferentnewsoutletswillbeintheiruseofAI,comparedto28%whoexpectsmalldifferences.

?Mostpeopledon’tyetrecallseeingaudience-facingAIfeaturesofferedbynewsoutlets.SixtypercentsaytheydonotregularlyseeAIfeaturesonnewssitesorapps.Most

commonareAIsummaries(19%)andAIchatbots(16%).

?SeeingAIlabellingonnewsisinfrequentrelativetodailynewsuse.Only19%seeAI

labelsdailyand28%weekly–alownumberconsideringthat77%saytheyusenewsdaily.

?OnlyaminorityassumesthatAIisusedinnewswithoutlabelling.About15%say

theyoftenoralwayssuspectAIwasusedwithoutlabelling(Argentina30%;US17%;approximately10%inJapanandEurope).ClearcommunicationofAIpoliciesremainsimportant.

Wealsonotethatacrossmostofourattitudinalmeasures,bothconcerningAIinnewsandAIingeneral,respondentsinJapanandArgentinatendtobemorepositiveandoptimisticaboutthepotentialimpactofgenerativeAIthanrespondentsinDenmark,France,theUK,andtheUSA.

THEREUTERSINSTITUTEFORTHESTUDYOFJOURNALISM

10

Introduction

ThelastyearhasseenarapidgrowthinthepublicuseofgenerativeAItoolsandsystemsandcontinueddevelopmentofthefeaturesandperformanceonoffer.Manybusinessessaythey

havenotyetseenanyclearbenefitsfromtheirattemptstousethetechnology,whileexamplesofproblematicperformancecontinuetosurface,andsomeworryaboutthescaleofinvestmentpotentiallycreatinganew-typebubble,theriskoffurtherentrenchingdominant

technologycompanies,andtheenvironmentalimpactofenergy-hungrynewtechnologies.

Inthisreport,weusesurveydatatodocumentandanalysethepublicuseof,perceptionsof,

andexpectationsaroundgenerativeAIwhenitcomestonewsspecifically,andsocietymore

broadly.Buildingonasimilarreportcoveringthesamesixcountriesthatwepublishedlastyear(FletcherandNielsen2024),welookatdevelopmentsinArgentina,Denmark,France,Japan,

theUK,andtheUS,whichtogetherrepresentarangeofdifferentmediasystemsandhigh/uppermiddle-incomecontextsacrossfourcontinents.

WhilemanydevelopmentsinandaroundgenerativeAIconcernspecificorganisational

adaptations,acentraldrivercontinuestobepublicuptake,andacentralissuecontinuestobehowthepublicmakessenseofthesetechnologies.Wefocusontheseissueshere,andexaminethepublic’suse,perceptions,andexpectationsofgenerativeAI.

Overall,ouranalysisdocumentsacomplicatedsituationwherepeopleincreasinglyseeAIas

beingusedeverywhere,whilehavingverynuancedexpectationsaroundwhattheimpactis

acrossdifferentsectors,forsocietyasawhole,andforthempersonally.Thereareplentyof

misgivingsandreservationsaboutsomeaspectsofgenerativeAIanditseffects.Atthesame

timepublicusecontinuestogrowrapidlyand,overall,thepercentageofrespondentsacross

thesesixcountrieswhosaytheyhaveusedatleastonegenerativeAIsysteminthelastweek

hasnearlydoubled,from18%to34%injustoneyear.Toputthatintoperspective,ittookaboutthreeyearsforinternetusetogrowinasimilarfashioninthesecountriesinthelate1990sandearly2000s.1ThereisplentyofunsubstantiatedhypearoundAIperformanceandadoption,

andmanyconcernsabouttheimplicationsofitall,butourdataclearlydocumentsrapidpublicuptakeacrossallcountriescovered–generativeAIuseseemstobegrowingroughlythree

timesasfastastheinitialspreadofinternetuse.

Inthisreport,afterexplainingtheonlinesurveymethodologywerelyon,wefirstexamine

publicawarenessanduseofgenerativeAI.Next,welookspecificallyatAI-generatedsearchengineanswersbeforeturningourattentiontopublicassessmentoftheuseandimpactof

generativeAIacrossdifferentsectorsofsociety.WethenlookatpublicopinionontheuseofgenerativeAIinjournalismandnewsspecifically.Finally,weendthereportwithaconcludingdiscussion.

1

/grapher/share-of-individuals-using-the-internet?tab=line&country=DNK~ARG~JPN~USA~FRA~GBR

&mapSelect=DNK~ARG~JPN~USA~FRA~GBR

GENERATIVEAIANDNEWSREPORT2025:HOWPEOPLETHINKABOUTAI’SROLEINJOURNALISMANDSOCIETY

11

Itisimportanttostressthat,aswithallsurvey-basedwork,wearereliantonpeople’sown

understandingandrecall.Thisisparticularlyimportanttokeepinmindwithanewand

sometimesnebulousphenomenonlikegenerativeAI,andwhenconsideringwhatpeoplethinkothers(whetherindividualsororganisations)dowiththesetechnologies.Butevenincases

wherethisdataisatoddswithbehaviouralandpassivetrackingdatafromothersources,itisstillanimportantsourceofinformationaboutpopulation-levelpractices,andpublicopinionoverall.

Wehopethedataandanalysiswepublishherewillbeusefulformediaprofessionals,scholars,andothersinterestedingenerativeAIanditssocietalimpact.

THEREUTERSINSTITUTEFORTHESTUDYOFJOURNALISM

12

Methodology

ThereportisbasedonasurveyconductedbyYouGovonbehalfoftheReutersInstitutefortheStudyofJournalism(RISJ)attheUniversityofOxfordandtheDepartmentofCommunicationattheUniversityofCopenhagen.Themainpurposeistounderstandifandhowpeopleuse

generativeAI,andwhattheythinkaboutitsapplicationinjournalismandotherareasofworkandlife.

Thesurveyandthisreportarefollow-upstoasurveycarriedoutin2024usingthesame

methodology(FletcherandNielsen2024).TheyarebothpartoftheAIandtheFutureofNewsprojectatRISJ.

The2025datawerecollectedbyYouGovusinganonlinequestionnairefieldedbetween5Juneand15July2025insixcountries:Argentina,Denmark,France,Japan,theUK,andtheUS.

YouGovwasresponsibleforthefieldworkandprovisionofweighteddataandtablesonly,andRISJwasresponsibleforthedesignofthequestionnaireandthereportingandinterpretationoftheresults.

Samplesineachcountrywereassembledusingnationallyrepresentativequotasforagegroup,gender,region,andpoliticalleaning.Thedatawereweightedtotargetsbasedoncensusor

industry-accepteddataforthesamevariables.

Samplesizesareapproximately2,000ineachcountry.Theuseofanon-probabilitysamplingapproachmeansthatitisnotpossibletocomputeaconventionalmarginoferrorforindividualdatapoints.However,differencesof+/-2percentagepointsorlessareveryunlikelytobe

statisticallysignificantandshouldbeinterpretedwithaveryhighdegreeofcaution.We

typicallydonotregarddifferencesof+/-2percentagepointsasmeaningful,andasageneralrulewedonotrefertotheminthetext.

GENERATIVEAIANDNEWSREPORT2025:HOWPEOPLETHINKABOUTAI’SROLEINJOURNALISMANDSOCIETY

13

Nationallyrepresentativesamplesizes

2024

2025

Country

Sample

Fieldwork

Sample

Fieldwork

Size

Dates

Size

Dates

Argentina

2,018

9-23April2024

2,010

6June-4July2025

Denmark

2,011

9-22April2024

2,004

6June-11July2025

France

2,056

9-22April2024

2,132

6June-15July2025

Japan

2,007

16-30April2024

2,288

9June-14July2025

UK

2,113

28March-5April2024

2,041

5-23June2025

USA

2,012

28March-5April2024

2,090

6-19June2025

Itisimportanttonotethatonlinesamplestendtounder-representtheopinionsand

behavioursofpeoplewhoarenotonline(typicallythosewhoareolder,lessaffluent,andhavelimitedformaleducation).Moreover,becausepeopleusuallyoptintoonlinesurveypanels,theytendtoover-representpeoplewhoarewell-educatedandsociallyandpoliticallyactive.

Somepartsofthesurveyrequirerespondentstorecalltheirpastbehaviour,whichcanbe

flawedorinfluencedbyvariousbiases.Additionally,respondents’beliefsandattitudesrelatedtogenerativeAImaybeinfluencedbysocialdesirabilitybias,andwhenaskedaboutcomplexsocio-technicalissues,peoplewillnotalwaysbefamiliarwiththeterminologyexpertsrelyon,orunderstandthetermsthesameway.Wehavetakenstepstomitigatethesepotentialbiasesandsourcesoferrorbyimplementingcarefulquestionnairedesignandtesting.

Somefiguresinthisreportdonotdisplayallofthepercentages.Allpercentagescanbeviewedintheinteractivefiguresat:

https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-

report-2025-how-people-think-about-ais-role-journalism-and-society

THEREUTERSINSTITUTEFORTHESTUDYOFJOURNALISM

14

1.PublicAwarenessandUseofGenerativeAI

AwarenessofgenerativeAIsystems

Incomparisontolastyear,awarenessofAItoolsandsystemshasgrownsignificantly.Theshareofrespondentswhosaytheyhavenotheardofanyofthe13toolsorsystemsweaskedabout

shrunktojust10%,meaningthat90%haveheardofatleastoneofthemostwidelyusedAItools.In2024thefigurewas78%.

Althoughitisstillnotuniversallyrecognised,Figure1revealsthatOpenAI’sChatGPTisstillbyfarthemostwidelyrecognisedgenerativeAIsystem.NootherbrandcomesclosetoChatGPT’srecognition,butgenerativeAIproductstiedtoincumbenttechnologycompanieshavegrownevenmoreinrecognitionincomparisontolastyear’sdata(seeFigure1),closingthegap

betweenChatGPTandtherestofthepack.

Figure1.ProportionwhosaytheyhaveheardofeachAIsystem

Onaverageacrosssixcountries,awarenessofAItoolsandsystemshasgrownsignificantly,withChatGPTthemostwidelyrecognisedsystem.

Anyofthese

ChatGPT

GoogleGeminiMetaAI

MicrosoftCopilotDeepSeek

SnapchatMyAIGrok

Claude

2%

2%

MidjourneyPerplexity.ai

Mistral

YManusAI

20242025

i78%||90%

>

>

>

%

%

>

>

0%20%40%60%80%

54%

>73%

17%

14%

>50%

46%

15%

32%

16%1

20%

7%

3

3

>15%>7%

6%7%

>6%

4%

3%

2%

0%20%40%60%80%

AI_brandheard.HaveyouheardofanyofthefollowinggenerativeAIchatbotsortools?Base:TotalsampleacrossArgentina,Denmark,France,Japan,theUK,theUSAineachyear≈12,000.

ChatGPTisthemostwidelyrecognisedAItoolineachofthesixcountries,rangingfrom

79%awarenessinDenmarkto68%inArgentina.BeyondChatGPT,awarenessshowsstrong