<<

November 2017 November

The Impact of CrossCheck on & the Audience

Learning the lessons from a collaborative project fighting disinformation online during the French Presidential Election. 2017

By Nikos Smyrnaios (University of Toulouse) Sophie Chauvet (LSE-FirstDraft) and Emmanuel Marty (University Grenoble Alpes)

firstdraftnews.com/crosscheck-findings

Supported by : CrossCheck - November 2017 THE VIEWOFAUDIENCE how thepublic received, used EXECUTIVE SUMMARY and interpreted CrossCheck ABOUT FIRSTDRAFT METRICS / p46 / p28 7 4 1 / p6 / p3 DISCUSSIONS DID CROSSCHECK DISCUSSIONS TWITTER, INTHEMEDIA AND GENERATE ONFACEBOOK, ABOUT THEAUTHORS FROM THEAUDIENCE? RESEARCH METHOD INTRODUCTION WHAT KIND OF / p56 / p32 2 8 5 / p9 / p3 the project asseenby theparticipants THE CROSSCHECK THE VIEWOF JOURNALISTS CONCLUSION APPENDICES FOREWORD PROJECT * / p62 / p34 / p16 / p61 6 3 / p5

2 importance ofcollaboration. importance scales the andhelpschampionthe work Draft, ofFirst organizations, anduniversities. network The informs and global , technology companies, human rights Network that now hasover organizations, 100 including age. InSeptember 2016, itexpanded by creating aPartner address challengesrelating to trust andtruth inthedigital Founding inJune Partners 2015 to raiseawareness and formed asanonprofit ofnine coalition Draft First and publish content sourced from thesocialweb. It provides practical, ethical guidance onhow to find, verify ofinformationabout andsharing the online. reporting First isdedicated Draft to improving skillsandstandards First Draft First ABOUT interactions withpublic opinion andpolitical issuesonline. through lexicometrics, especiallymediaframes andtheir analysis new mediadiscourse practices, andits journalistic Institute of Grenoble University. Alpes Hisresearch concerns theories andpractices at theInformation andCommunication (UNS) from 2011 to 2017, andisnow teaching Cannes , University of Nice. Antipolis Sophia He taught digitaljournalism asanAssociate Professor at the munication Sciencesfrom theUniversity of Toulouse, France. receivedEmmanuel Marty hisPhD inInformation and Com- on journalism andsociety, andhow to advance solutions. She isnow interested intheimpact of onlinedisinformation for social projects related to mediainSouth America. online publications inFrench, English andGerman,sheworked February to September 2017. After contributing to various CrossCheck asaproject editor andresearcher from Science from McGill University. She collaborated on from theLondon of School Economics inPolitical andaB.A. ChauvetSophie inMedia holdsanMSc andCommunications controversies onsocialnetworking sites. business modelsof andpolitical journalistic startups, about reviewed andbookchapters articles inEnglish andFrench use of socialmedia.He haspublished numerous peer- His research focuses ononlinejournalism andthepolitical history, sociology, culture andeconomics of themedia. at the University ofToulouse, where heteaches theory, France. Since 2007, hehasbeenanAssociate Professor Communication Sciences from theUniversity of Grenoble, Nikos Smyrnaios received hisPhD inInformation and editorial onlinenews andpluralism, and diversity The authorsThe ABOUT

3 CrossCheck - November 2017 FOREWORD

4 of people’s voting experiences. onelection day overvirtually thestory journalists,tional 150 allcollaborating hub inNew York withanaddi - City the United States, andanewsroom over 400reporters located across over journalism 600 studentsand It was anambitious project involving presidential election. of voting problems duringtheUS monitoring socialreports project, ProPublica for theirElectionland 7. 201 Director inJanuary of First Draft Jenni Sargent, whowas Managing CrossCheck was thebrainchildof had partneredFirst with Draft By Claire Wardle, EXECUTIVE DIRECTOR * this project andprovided much neededresources. Lab andFacebook, whowere willing to work together on to our Devteam for staying upallnight,andto GoogleNews to alltheparticipating newsrooms andtechnology companies, editors, to AFPwhoacted as theoverall editorial , Karolina Johanssen,to theirtireless student team of 10 co-ordinator MarieBohner, to editors SamDubberley and success, from project manager JenniSargent, to project youThank to everyone whomade CrossCheck such a what should bechangedfor future collaborations. wanted to know what worked andwhat didn’t, andto learn new things. We commissioned thisresearch because we exampleperfect ofhow much we whenwe can learn try on testing assumptions inthefield, CrossCheck was the iswonderful. Forstory First anonprofit Draft, that works today,the industry thatof beingpart agoodnews unbelievable. There are somany depressing stories about proud of what was achieved andwant to continue, issimply The fact that theproject worked, that thejournalists are so early mistake would ruintheproject’s reputation forever. new ‘brand’ sufficiently, orbecause it would beimpossible to create a additional work, orthrough lack of time to planout theproject failed, eitherthrough newsrooms not beingableto support No-one knew ifitwould work anditcould quite easilyhave project wasThe always considered anexperiment. was built. day bootcamp was organized, new andabrand website Workflows were designedand technologies chosen,a 3 27, theproject was conceived, mappedout, andsignedoff. over 40journalists, to thelaunch of theproject onFebruary 6. InthesixweeksJanuary between that first meeting with David Dieudonné from GoogleNews Lab inParis on project, JenniSargent arrangedameeting withthehelpof rested onaFrench inworking with Draft First election when French journalists suggested they would beinte- Inspired by theproject, andfull of thelessonslearned, in such a short space oftime, insuch orbecause ashort an

5 CrossCheck - November 2017 Executive Summary 1

6 on theCrossCheck website. and finally with the public, who were invited to ask questions journalists from different newsrooms, thenbetween platforms, collaboration tidimensional — first and foremost between their platform, CrossCheck was aunique example ofmul- bysupported Facebook through promotional advertising on Organized by Lab, News First funded Draft, by and Google if onlypublished onthe CrossCheck site. debunk was higherthan considerably itwould have been ontheir sites, meaningthat the reachstories ofeach also published CrossCheck newsroomsThe partners process underpinningthatdebunk. particular the newsrooms that had participated intheverification ‘Fabricated’, etc.). Each was alsolabeledwiththelogosof disinformation that had beenverified (e.g. ‘Misleading’, by avisualicon whichwas designedto explain thetype of content (e.g. ‘True’, ‘False’, ‘Insufficient Evidence’, and etc.) panied by anoverall categorisation of theclaimorpiece of were67 sucharticles published. Each was accom article - CrossCheck website, and, over thecourse of theproject, became widelyshared, a debunk was published onthe on thesocialweb. Whenmisleading ormanipulated information rumours, aswell asfabricated images andvideos, circulating journalists inthirty-three newsrooms monitored claimsand From February 27 untilMay 5, 2017, more than onehundred between audiences andthenews industry. online disinformation, aswell asimprove levels of trust journalistic practices for themonitoring anddebunking of collaboration at thisscale mightimprovea whether test that consumed theoutput. CrossCheck was designedto journalists that participated intheproject andtheaudience This report seeksto analyse theimpact of theproject onthe up to the French 2017 election. presidential designed to fightdisinformation in the ten weeks leading CrossCheck was acollaborative project journalism studies withtwo sets of participants: This report isbased onthree research METHODOLOGY the project. to investigate what we learnedfrom and thisreport was commissioned several research questionsinmind, This project wascreated partly with were recorded and analysed thematically. 2017—a few weeks afterthe project ended guide, were conducted withjournalists andeditors inJune A total in-depth of interviews, 16 based onan ad hoctopic news andquestionssubmitted by articles theaudience public whoengaged withtheproject. content improve audiences’ mediaandcritical literacy skills? techniques usedby journalists to debunk aclaimorpiece of increase readers’ trust intheiroutput? room training? journalists’ verification skills to agreater degree thanclass the CrossCheck workflow improve of thequality verification? that participated inCrossCheck. 3 2 1 4 3 2 1 Did thecollaborative Did journalistic processes embeddedinto Adiscourse analysis of related tweets, Facebook posts, Asurveyandfollow-up interviews withmembers of the publishing Did astep-by-step demonstration of the Would apublic alliance of recognized mediaoutlets working Did onanactive debunking initiative improve Semi-structured interviews withjournalists andeditors 1 . Theinterviews -

7 did receive over 800,000 views. 2 _Whilethelimitations of Facebook videoview numbers are well known, onevideo Appendix3for1 _See theTopic Guideof theinterviews. CrossCheck - November 2017 family whowere sharingdisinformation. in thedebunks, andshared thesewiththeirfriends and admitted to learningnew skillsfrom theprocesses outlined visited thewebsite onadailybasis, andread entire stories; There isevidence that acore segment of theaudience commenting that they were sadto seetheproject end. to withsome the whoresponded survey), ofsample people were also positive (although there arelimitations to the oftheMembers audience that responded to the survey testing how itmightwork outside anelection campaign. project, andwould like to continue to thecollaboration, try who participated intheproject. Theyproud are very of the Overall, theproject was considered asuccess by those KEY FINDINGS CrossCheck. by members of thepublic, andnewsreferring articles to the nameof theproject, questionsthat were submitted comments onCrossCheck’s Facebook page, tweets containing cometric method was used uponfour different samples: Finally, adiscourse analysis based onthe lexi Reinert - when itcomes to theresults of thestudy. not includedinthesample. limitation Thisisanimportant only occasionally incontact withtheCrossCheck project are checking thantheaverage French citizen. Thosewhowere political orientation, although they were more aware of fact- diverse interms of age, area of residence, education and managed to reach asurprisinglywideaudience that was ristics of thosewhoresponded shows that CrossCheck interested intheproject andthedemographiccharacte- small, itisrepresentative of thosethat were particularly ofthemembers public. Thoughisrelatively thesample were7 semistructured interviews conducted with In total, 29responses were received. Inaddition, JOURNALISTS IMPACT ONNEWSROOMSAND new journalistic techniques, outside thetraining room. were noted asextremely powerful asways of embedding imagesfact-checking andvideos claims andverifying the frequent conversations onSlack about the process of techniques (for example andCrowdTangle) NewsWhip and provided astrong foundation, thedailyuseof new tools and reported learningnew skills. Whilethekick-off bootcamp who had worked infact-checking andverification previously, it should beconsidered apublic service. should not becompetitionof work, inthistype andinfact fact-checking anddebunking, participantsagreed that there alreadychecking Libération) unitat had strong reputations in (for example Le Monde’s team, andthe fact- Decodeurs collaborations. reminder of including theaudience injournalismnecessary) diverse, whichprovides hopefully anun (but animportant ignore. Participants noted thepower of thesecross news decisions about what to reportandwhat to strategically allowed competitive otherwise newsrooms to make joint journalism. slower than resulted inhighquality reporting, traditional acknowledgement that thecross-checking process, while the project was often slow to publish, there was ashared corrections had to beissued, andwhilesomestruggled that other to account. Participants were extremely proud that no with participants explaining that they were ableto holdeach seen ascompetitors resulted journalism, inhigherquality ‘show your work’ to newsrooms that would be otherwise online communities to wideraudiences. to give additional oxygen andto move themout of niche and fabricated rumours content, relying onthem amplify themselves are beingusedby ofdisinformation to agents room conversations, at particularly atimewhennewsrooms 2 1 5 4 3 Journalists whotook intheproject, part even those Even though someof thepartners inthisproject Overall, thepublic’s contributions were useful and Theprocess of collective editorial decision-making Theprocess of working transparently, andhaving to - -

8 FUTURE CONSIDERATIONS IMPACT ONAUDIENCES making short explainer videosformaking short Facebook. towards the endof theproject, CrossCheck editors started this needsto beresearched ingreater detail. Inaddition, (see Figure 3for anexample of cards). these Theimpact of alongside any image that referenced the fabricated content a graphictemplate whichallowed editors to usethese original piece of fabricated content. AFPtherefore designed posts) meantthatmedia CrossCheck was perpetuating the (whichposts, then got into automatically dragged social clear that including screenshots asthe ‘hero’ image onthe took place to theoriginalprocesses. For example, itbecame using images and videos. Astheproject evolved, changes reasons that itappealedto awidespectrum of people. The perceived of theproject impartiality was alsooneof the reached peopleacross the spectrum. political appears to have beenoneof thereasons why the project people’s mindsaround theirvoting choices. the election, andsomeeven claimedto have changed both onlineandoff,who were sharingdisinformation around CrossCheck stories andinformation withfriends andfamilies, content emotional withparticularly language orvisuals. evidence thatof peoplealsolearnedto bespecifically wary howlearn they could dothis work themselves. verified increased trust inthearticle, but alsohelpedthem how arumour orpiece of content was fact-checked or so many outlets. andcredible because itincluded independent, impartial reporting. Respondentsfelt that CrossCheck was more meant that respondents had increased levels of trust inthe 1 5 4 3 2 1 Undertaking additional research Undertaking oneffective debunks Thefact that theproject includedlocal outlets, Respondents explained how they had shared Aswell aslearningcritical reading skills, there is Respondents noted that thetechnique of explaining Having multiple newsrooms collaborate onstories

shared widely The metrics immediately showed that they were being why CrossCheck worked intheFrench context. newsrooms to collaborate. It’s we important understand have elections German beenlesssuccessful at getting around the tosimilar run projects attempts UKand While theresults of thisresearch havepositive, beenvery impact ontheUSpresidential election. been theactive conversations about disinformation andits senior editors more time to‘no’) say orifthere hadn’t just hadDraft had alongerlead time(which given wouldhave that CrossCheck would never have got offtheground if First bound contexts for collaborative projects. likely Itisvery rumour onaparticularly orpiece of fabricatedstory content. looking at before they decidewhether andhow to publish a tipping pointsits, andwhat metrics journalists should be More analysis needsto beundertaken about where this During CrossCheck, decisionswere taken collectively. and how of tostories. reportonthesetypes that newsrooms willneedto give additional thought to when tions, andthethreat of givingoxygen to rumours, means Reporting ondisinformation requires different considera- debunks andfact-checks. about themosteffective ways ofcreating videobased 3 2 Understandingof theimportance culturalandtime- the Understanding point’ ‘tipping 2 but more research needsto be undertaken

9 CrossCheck - November 2017 Introduction 2

10 several research questions: logistics were finalised by First anddesigned Draft to test projectjournalism to date. Theproject plan,processes and CrossCheck, ofSome theideasdiscussedat that meeting evolved into upcoming French election. prevent thespread of mis- anddis-information duringthe the morningdiscussingways that they could collaborate to 6, 2017, representatives from anumber of newsrooms spent At ameeting onJanuary facilitatedeffort. by Draft First From beginning, CrossCheck thevery was acollaborative of theirjournalism. ‘cross-checking’ each other’s work improved the quality and thoseinvolved intheproject felt that theprocess of It improved verification skills across French newsrooms, and wanted itto continue. content, were positive very about the initiative, and audience members whoconsumed CrossCheck Overall, we found that journalists involved intheproject, from CrossCheck. project andaudience members whoconsumed content impact ofCrossCheck involved onjournalists inthe The present research was commissioned to examine the content improve audiences’ mediaandcritical literacy skills? techniques usedby journalists to debunk aclaim orpiece of increase readers’ trust intheiroutput? than classroom training? improve journalists’ verification skills to agreater degree verification? into theCrossCheck workflow improve of thequality 4 3 2 1 Did publishing Did astep-by-step demonstration of the Would apublic alliance of recognized mediaoutlets working Did onanactive debunking initiative thecollaborative Did journalistic processes embedded the real-time mostambitious collaborative with their audience. help journalists develop better andrepair practices trust journalism asaprofession, apart andhow collaboration can tional circumstances allowed for anexploration of what sets they would normallyseeascompetitors. Theseexcepwhom - reconciling theirwork habits withthoseof theirpeers, foran opportunity journalists to explore by theiridentity the present research discovered that CrossCheck provided In addition to answering thefour questionsstated above, see theprocesses usedby thejournalists to verifycontent. they were more trusting of thecontent because they could had worked together onit.People alsosuggested that of CrossCheck’s content because multiple newsrooms research doessuggestthat audiences were more trusting While theaudience was sample not representative, our skills andtheirmissionof public service. to question andstrengthen theiridentity, their journalists response ofcollaborative fact-checking, can spur Our suggestionisrather that disinformation, andthe mation. only solutiontodisinfor nor provethatitisthe the crisisofmedia, collaboration cansolve is nottodemonstratethat The pointofthisresearch

-

11 produced withtheintention of causing harm. appropriate policy responses Disorder are: they identify The three of Information types satire (including is usedto refer to many, different andpractices very contexts mis- anddis-information. It’s avague andambiguous term First of all, itisinadequate to describethe complexities of this for term anumber ofreasons. pulated images andvideos. Inthisreport,we avoid using misleading orfalse information and/or mislabeledormani- The term '' colloquially refers to content that uses 1/ CrossCheck - November 2017 disorders allrefer to the of falseness concepts andharm. As shown inthefigure below, thesedifferent kinds of harm onaperson, socialgroup, organization, orcountry. Hossein Derakhshan typology proposedas definedinthe by Claire Wardle and In thisreport, we willinstead usethe disinformation, term to discredit citizen journalists andalternative news sites organization, orcountry. deliberately produced to harmaperson, socialgroup, to discredit thefree press Additionally, theterm hasbeenco-opted by somepoliticians Disorder theoretical framework. ONLINE DISINFORMATION ? 3 2 1

WHAT RESEARCHTELLSUSABOUT Malinformation: Misinformation: Disinformation: 3 ), andthus hinders theestablishment 7 , whichderives from their Information Information that isfalse and Information that isfalse, but not 5 accurate information usedto inflict , andby themediaestablishment 4 .

of of 6 . disciplinary framework for and policymaking”, research disciplinary Council of Europe, 2017) (: Claire Wardle, “Information Hossein Derakhshan, Disorder.Toward aninter INFORMATION DISORDER The three types The three types INTENDED of information TO HARM Figure n°1 FALSE disorder Hate speech Harassment Leaks information Mal Fabricated content Manipulated content False context information Dis Misleading content False connection information Mis -

12 news and thefact that socialmediaisthedominantsource of as awhole concentration, whichhasledto agrowing distrust of journalism issues of partisanship, bias, ethical standards andownership located inthe mainstream media’s to address inability be can ofthe roots The disinformation phenomena disinformation – itisnot new. However, thiscrisisisalsopolitical – andlike theissueof news balanced withtheconsumptionofhard when itisunderstoodassatirical and how the public perceiveseven politicians, the short duration ofthe short political communication’s impact these mechanismsismitigated by other findings showing lective memory Disinformation hasanegative effect onindividualand col- the digital mine" coal a symptom of thegeneralcrisisof media – in a"canary The explosion of onlinedisinformation can beviewed as individual on hasanegativeeffect "Disinformation popularity ofpopularity false news andrumours audience maximization, andthus economically valuethe advertising modelsanddistribution channelsthat dependon digitization. Newsrooms have beenforced to adapt to new and theirbusiness modelshave beenprofoundly affected by journalism to renew itself. Theglobal landscape of media even favour thespread of disinformation publishing standards that influence journalistic practices and capture of online asignificant portion revenue, andset online They includeconfirmation bias formation have beenresearched extensively. disin mechanisms at playconsumes whenanaudience with theconsumption of hard news cians, even whenitisunderstood assatirical andbalanced Indeed, distribution and selective exposure for only14% ofAmericans 11 13 . ." 12 , andonhow thepublic perceives politi-

platforms suchasGoogle andFacebook 8 16

– that creates for opportunities and collectivememory . But theultimate of importance 14 , belief perseverance 18 13 . 9 . . The psychologicalThe 10 . 15 17

, - we willexamine inthisreport the French presidential campaign, isonesuchstrategy that these phenomena.CrossCheck, aproject organized during significant enough to demandstrategies that can counter – by contaminating thepublic sphere withconfusion – is long-term damage to the societies fabric ofdemocratic Nonetheless, the ofdisinformation risk causing 18 _Allcott,18 H., & Gentzkow, M.(2017). persuasion effects from mass communication. Political Communication, 30(4), 521-547. Hill, _ S.,17 Lo, J., Vavreck, L., & Zaller, J. How quicklywe forget: (2013). Theduration of Communication Research, 41(8), 1042-1063. media: Endorsements source trump partisan affiliation whenselecting news online. _Messing,16 S., & Westwood, S. J. Selective exposure (2014). intheage of social findings from socialscience. New America Foundation. _Nyhan,B.15 & Reifler, (2012). J.A. Economic Perspectives, 211-236., 31(2) & Gentzkow, Media Social andFake M.(2017). News inthe2016 Election. Journal of political beliefs. American Journal of Political Science, 755-769 50(3), and Allcott, H., 413-439, Taber, C. S., & Lodge, M.(2006). Approaches to Credibility Evaluation Online. Journal of Communication,, (3) 60 Metzger,14 _See M.J., Flanagin,J., A. Medders, R. B. (2010). 41 Whenfake _Balmas,M.(2014). 13 news becomes real. Communication Research, Nature,memory. 168-170. (7644), 543 _Spinney,12 How Facebook, (2017). L. fake news andfriends are warping your mouth University. 'Fake news' presented by inquiry theFaculty for Media & Communication, Bourne- 179-196 andLilleker, D. Evidence to theCulture, (2017). Media andSport Committee Nip,11 _See J. Y. Thelastdays of . Journalism M.(2008). Practice (2) Fields Journal, no10. ‘Google _Smyrnaios, Infomediation10 andtheAlgorithmic N.(2015). of News’, Media The London of School Economics andPolitical Science. 9 _Tambini, D. & Goodman,E.(2017) March 2017. 8 _Beckett, C. ‘Fake (2017). news’: Thebestthingthat’s happenedto journalism. Polis, as defined by WardleDerakhshan. and the term disinformation inthisreport to refer to both disinformation andmisinformation framework for research andpolicymaking. Council of Europe. For convenience we use 7 _Wardle, C., H.(2017). Derakhshan, 6 _Idem. Project . 5 _Tambini, D. Whobenefits from using the term ‘fake (2017). news’? LSE Media Policy perspective. November 25). mediaandfake Social news from (2016, afree 19 4 _Article speech Abingdon: Routledge. S.In Allan, TheRoutledge (Ed.) companion to news andjournalism 374-383). (pp. 3 _Baym,G.(2010). Aimee Rinehart for Rinehart theirpreciousAimee contribution to thisreport. _Theauthors19 would like to thankClaire Wardle, NicDias,MarieBohnerand Journal of Economic Perspectives, 211-236., 31(2) 430-454. (3), 12 , and on , andon Real News/Fake News :Beyond thenews/entertainment divide. Misinformation andfact-checking: Research . Fake news: Public policy responses. Information Disorder. Toward aninterdisciplinary Social MediaSocial andFake News inthe2016 Elec 19 Motivated skepticism intheevaluation of . Social andHeuristic. Social

tion. tion.

13 IN THEFRENCHCONTEXT toward the end of hispresidency in2017. François Hollande reachedPresident thehistorical low andtheapproval rate parties, political of formertrusts In fact, ofthe onlyasmallminority French population mistrust of theFrench towards thepolitical system. These groups have beenstrengthened by theextreme of news consider socialnetworking sitessource to beanimportant to different studies, between15 and 20% of the French than eightmillionuniquedailyvisitors. Inaddition, according most popular website inFrance behindGoogle, withmore consumption, isconstantly rising: Facebook isthesecond At time, thesame socialmediause, for particularly news journalists sources that can bedifficult to navigate, even for professional in Europe, witha30% approval rate phenomenon: trust intheFrench mediaisamongthelowest information andxenophobic propaganda flooding the French ofmisleading withadiversity ‘fachosphère’– hasrecently beenespeciallysuccessful at – more orlessrelated to Front National andknown asthe of loosely-connected, far-right onlinegroups andwebsites Islamophobic andanti-immigrant discourses. Avastnetwork spread of conspiracy theoriesandincreasingly disinhibited multiethnic suburbs),France hasnot beenspared from the nepotism,nomic growth, tensions and within working-class socioeconomic issues(e.g., highunemployment, slow eco In acontext of recurring terrorist attacks andlongstanding social media linked decreasing trust levels inlegacy mediaandtheriseof taneously, risingsocialmediause. Indeed, scholars have by risingmistrust intraditional journalism and, simul- prevalence andimpact onaudiences have beenamplified Disinformation may not bearecent phenomenon,but its both of whichoccurred in2016. the US’s presidential election andtheUK’s EU referendum, The issueof onlinedisinformation hasbecomesince salient 2/ CrossCheck - November 2017 economic forces andpolitical majority ofthat thinks are influenced journalists by

THE PROBLEMOFDISINFORMATION 24 . 21 . France affected hasbeenparticularly by this 20 , whichhascreated anew landscape of news 23 . 22 andapopulation 25 . - election outcome.election impact ofamassive disinformation campaign onthe mainstream media’s to handlethe ability potential UK – the2017campaign represented atrial for French the precedents of disinformation campaigns intheUSand can provide afertile ground for disinformation – andgiven unsettledConsidering thisparticularly context, which BBC News, http://www.bbc.com/news/world-europe-39475635 25 _Astier, H.(2017, French 5April). election: Isonlinefar rightathreat to ? _ 24 la-parolepublique Available at :https://fr.slideshare.net/IpsosFrance/la-confiance-des-franais-lgard-de- _Ipsos,23 La confiance des Français àl’égard dela parole publique, 2015. NewsDigital Report 2017. 22 _Newman, N.,Fletcher, R., Levy, D., & Nielsen,R. (2016). Landscape. Data & Research Society Institute. andFontaine, _Madden,21 M.,Lenhart A., C. How Youth (2017). Navigate theNews Houndmills, Basingstoke, 39-56). (pp. Hampshire ;New York, NY: Palgrave Macmillan. through media.InCouldry, N.,Pinchevski, A.,& Ethics Madianou, of media M.(Eds.), 38-49).nalism (pp. London: Routledge andCouldry, Living well withand N. (2013). professional identities. S. InAllan, TheRoutledge (Ed.), Companion to News andJour Cambridge: Cambridge University Press, Donsbach, W. (2010). Practice, 373-382, 4(3), O'Neill, O.(2002). 20 _ in 2017." the endofhispresidency historical lowtoward Hollande reachedthe PresidentFrançois former and theapprovalrateof politicalparties, trusts of theFrenchpopulation "Only asmallminority See for See example Médiamétrie, étude Actu 24/7, 2016. See Phillips, A. (2010). Phillips,See A. Transparency andthenew ethics of journalism. Journalism Aquestionof lectures ;2002). trust (Reith TheReuters Institute Journalists andtheir -

14

15 CrossCheck - November 2017 The CrossCheckThe project 3

16 26 _https://crosscheck.firstdraftnews.com/france-en/ reports inboth English andFrench campaign, andpublished atotal of sixty-seven debunking related toand socialthe mediaposts presidential CrossCheck’s reviewed hundreds participants ofarticles as project editors. day, Every for two andahalfmonths, journalism studentswhoserved and 10 UK, the and France journalists in33regional100 andnational newsrooms in information through acollaboration between more than The project aimedto debunk andverify suspicious debate throughout the period. pre-electoral disinformation andfoster awell-informed democratic 2017. Itwas conceived to combat asajointeffort online until theendof theFrench presidential campaign inMay CrossCheck was launched inlate February 2017, andlasted 3/

THE CROSSCHECKPROJECT journalistes, plusNewsWhip par desétudiants/ Suivi manuel par lepublic viaHearken soumises Questions via GoogleTrends identi ées Questions 26 . de projet gérés par leschargés en attente Éléments RECHERCHES

vériées sur Slack desinformations automatiques Alertes collaborative surCheck Véri cation

sur Facebook ou Twitter rstdraftnews.com, en direct surcrosscheck peut suivre lesinformations Le public pour leur audience préparent dessujets participant au projet Les rédactions This is the original slide This istheoriginalslide used to explain the d’information de brèves ches publient endirect Projet project to the Figure n°2 participants.

à partager et desgraphiques crée desexplicatifs First Draft

17 the nuances of themis- anddis-information. misleading, misreported orsatire – to helpreaders understand was alsomade – manipulated, fabricated, misattributed, was markedIf astory asFalse, anadditional determination True, False, Caution, Insufficient Evidence and Attention. to thefollowing typology: were articles All marked according investigative for work additional (SeeAppendices examples). logos from thenewsrooms that participated its andconfirmed Each published ontheCrossCheck article website included CrossCheck - November 2017 of the top half of a story of thetop halfof astory icons that were added These were thevisual This isascreenshot on theCrossCheck to thestories. Figure n°4 Figure n°3 website.

MANIPULATED MISREPORTED SATIRE MANUFACTURED MISATTRIBUTED MISLEADING

18 which was enabled by aHearken formation by onthe askingquestions project’s website, CrossCheck’s audience was also invited to signaldisin- web & metrics). media social fortwo andhalfmonths (See the Appendix moredetailed gathered more than1.2 millionviews duringtheproject’s than 180,000 followers, videos explanatory anditsshort 30,000 itsFacebook profile visits, page was liked by more on socialmedia.ItsTwitter page attracted more than from the public. Furthermore, theproject active was very As aresult, CrossCheck received more than600 questions the graphicbelow. joined by ofmedia organizations, alllisted adiversity in Hearken, NewsWhip, Check, andSAM. Theseactors were and technical partners such asCrowdTangle,Economics; Journalistes), SciencesPo, andtheLondon of School top universities suchasCFJ (Centre deFormation des CrossCheck ofdifferent including unitedactors, arange PARTICIPANTS assumptions: CrossCheck was based onseveral in thegrowing problem of disinformation project inresponsejournalistic to critiquesabout theirrole to contributetheir to acommon collaborative rivalry also uniqueinthat both Facebook andGoogleset aside beyond circles. Assuch,thiscollaboration journalistic was on itsplatform, thus increasing the project’s visibility Facebook alsoprovided funds ads tofor support CrossCheck and thehiringof additional staff. creation of CrossCheck’s website, thetraining of participants News Lab’s input was fundamental, asitallowed for the a nonprofit, and funded by Google NewsLab. Google news literacy. CrossCheck was organized by Draft, First where andhow anonlineclaimwas rightorwrong improves withreaders. credibility each willboostits published article verification, whichbenefitsallnewsrooms. 3 2 1 Taking readers step-by-step through ananalysis of Including multiple logosof well-known to medianext Collaborative around journalism helpsbuild capacity 28 plug-in. 27 . tions by fellow news professionals. image search, andothertasks verification about also geo-location, learned participants reverse tatives While sometools were presented by theirofficial represen- verification tools could beused. workshops andwere given thechance to learnhow of variety At thisboot camp, journalists andstudentsattended a to athree-day boot camp inNormandie, France. Before theproject launched, allparticipantswere invited TRAINING Technology Partners: audience, andto integrate thosequestionsinto theireditorial output. 28 _Hearken isatool whichmakes iteasierfor newsrooms to askquestionsof their book”, 27 _ See forSee exampleTufekci, Zeynep, “Zuckerberg’s Preposterous Defense of Face , September 29, 2017. (e.g., CrowdTangle, NewsWhip and Check), (e.g.,and CrowdTangle, NewsWhip CrossCheck partners Figure n°5 from presenta-

-

19 CrossCheck - November 2017 Photograph taken theCrossCheck Figure n°6 bootcamp reached many people. in order to avoid providing that oxygen had not to astory silence, whilebeingmindful of thepopularity arumour Third, journalists had to observe apolicy ofstrategic been previously investigated. not had Second, journalists had to focus content onoriginal that specifically about the presidential campaign. considerations. First, journalists had stories to prioritize objective reporting. This translated into three important towards transparency andacommitment to balanced, fair and An overarching goal of CrossCheck was to demonstrate efforts determining whichstories would bedebunked by CrossCheck. In general, participantshad to follow aselection policy in CONTENT SELECTIONPOLICY the workflow outlined below. later participants were inthiswe describe report) trained in the project asdifferent cases of disinformation emerged (as flow. also introduced to CrossCheck’s methodology andwork during theproject. Duringthetraining, participantswere fully acquainted, asthey would beworking together online would work, but alsoto ensure participants were success andexplain how indetail CrossCheckparticipants goalofthisThe to level-up was training primarily all Whilesomeelementswere necessarily adapted during - -

20 Handbook, whichwas given to allproject participants. The following diagram was taken from theCrossCheck WORKFLOW If you have easilydebunked or confirmed anitem yourself : Add a short summary to thenotes section summary Add ashort Paste the Check linkandpaste itinthe Edit theheadline to become aquestion Add brief notes to each of thefive task a linkoruploading ascreenshot to the Create areportinCheckby pasting Change thestatus of thereportto #crosscheck channelinSlack. boxes explaining thechecksthat explaining why you have chosen and add necessary tags. and add necessary to investigate theitem. true, false orcaution. appropriate channel. you have made. If you needhelpto complete checks notes to any tasksyou have beenableto complete. to confirm ordebunk a report: #visualverification or#factcheck channels Return to your report inCheckandadd brief Add a short summary to thenotes section summary Add ashort Edit theheadline to become aquestion Leave untilall thestatus as'inprogress' a linkoruploading ascreenshot to the on Slack explanation. with ashort Create areportinCheckby pasting Paste the Check linkinto either explaining why you have chosen tasks have beencompleted. and add necessary tags. and add necessary to investigate theitem. CrossCheck workflow appropriate channel. as published in the Participant Figure n°7 Handbook

21

dashbor, tice ed bythe Ther wilaysbnmtofdjug publish reotadnmfyCck. sek toanwrquifmhpblchahev,d and eitorsfhmlvwche.Ipy, Crosheck wilnvtga,mdp Al French. found ​her.Tmislyavb own ebpagsuicdth found athis​lkembyr The Crosckqutinfmab team. @sam_firtd onSlckePjE,uwhp- urgency adlvofthpiqs.Thiwbk a postwas likely to receive over 24 thenext hours. participants to algorithmically predict how much engagement an unusual amount ofattention. NewsWhip alsoallowed were whichstories identify attracting significant and/or CrowdTangle andNewsWhip enabledjournalists to quickly ofquestions, technology. withthesupport Tools suchas This activity was powered by editors andtheaudience’s were spreading considerably. cases ofdisinformation thatsocial media,to identify and journalists monitoring theinternet, andspecifically The first step of the workflow consisted of project editors CrossCheck - November 2017 l responactdi a. b. Social mntrg Question Frm

A SCREENSHOTFROMNEWSWHIP

to seehow much engagement The platform allows you a piece of content has received, andislikely to receive inthenext

24-hour period. Figure n°8

n termsofpiy,badh

transfer themto journalists. relevant questionsaccording to theselection policy and real-time, the system published to questions aSlack channelin a link, andtheiremailaddress. Powered by Hearken, The form asked audience members to provide adescription, they were invited to reportitthrough aquestionform. they found online, orhad witnessed theriseof arumour, If members of theaudience had doubts about information to askquestionsonCrossCheck’sopportunity website. had begunto search arumour. Thepublic alsohad the the spread of rumours viaGoogleTrends by seeingifpeople of arumour.the visibility Additionally, editors could identify enabling CrossCheck’s andpredict participants to quantify which cases of mis- anddis-information to address by These tools facilitated thedailyeditorial decisionsabout so that editors could quicklyselect themost CrossCheck website, where audiences to submitclaims, rumours, images orvideos they wanted fact-checked a Hearken plug-in allows A screenshot from the Figure n°9 or verified.

22 Sample questionsfrom thepublic : for Macron? to citizensappealingforletters them to vote National back in1991? forhousing destined migrants? schools? primary Mélenchon? to finance hismovement En Marche?

Are there mayors inFrance whosent out Did Jean-Luc the Mélenchon praise Front Is aParis lighthouse beingreplaced by inFrench to bemadeIs Arabic compulsory Figaro surveysIs Le manipulating its against Is ittrue that Macron used taxpayers' money Sample questions from thepublic Figure n°10

23 the following questions: which asked journalists to answer through adefinedchecklist, Check allowed journalists to work verification process. journalists to clarify, structure andcollaborate about the card was created onCheck, anonlinetool that enabled a rumour was deemedworthy of adebunk. Averification The second step of theworkflow was initiated assoon CrossCheck - November 2017 3 2 1

When was the content created? Who created the content? Is this the form original ofthe content? Il ya7mois21notes PRUDENCE Tâche devéri cation5/5résolue(s) Résolue par Rajouté par Décrédibiliser legouvernementsocialiste publication ducontenuexaminé? Pouvez-vous identierlesmotifsderrièrela Motivation duPost signe-un-accord-pour-enseigner-larabe-en-primaire-78207 https://www/valeursactuelles.com/societe/najat-v-belkacem- évoqué lacréationd’uneécoletunisiennesituéeàParis. Les deuxministresdel’Educationontégalement l’arabe danslesécolesprimairesfrançaises. aurait étésignéentrelesdeuxpourenseigner d’information tunisien,BusinessNews,unaccord Néji Jelloulvendredi31mars.D’aprèslesite Belkacem, arencontrésonhomologuetunisien, ment supérieuretdelaRecherche,NajatVallaud La ministredel’Educationnationale,l’Enseigne- primaire? permettre l’enseignementdel’arabeen Najat VBa-t-ellesignéunaccordpour Chronologie devéri cation Commenter cerapport Slack show how Checkwas used by journalists to share These reproductions from the verification work they il ya7mois Nusinessnews (sitetunisien) L’article sourcede Tâche ‘qui?’résoluepar: il ya7mois tunisien. qui tiresasourced’unsite Un articledeValeursactuelles Tâche ‘Originale’résoluepar: il ya7mois primaire sera-t-ilobligatoire? L’enseignement del’arabeen Titre modišépar il ya7mois Google. RECHERCHE chercher descopiessur Cliquez sur«Recherche»pour plusieurs images. Ce médiacontientuneou were doingonpieces Figure n°11 of content. ENVOYER

24 Source secondaire Preuves Additionnelles Originale ? Qui ? http://nordpresse.be/najat-vallaud-belkcam-larabe-cp-ne-sera-obligatoire-cours-detude-coran/ Nordpresse reprendl’infodansunarticleparodique de naturecrédible? Le contenuexaminéfait-ilréférenceàdessourcessecondaires(citation,rapport,data), des-cours-d-arabe-obligatoires-dans-le-sud_5005537_4355770.html http://abonnes.lemonde.fr/les-decodeurs/article/2016/09/29/des-parents-delirent-sur- serait pasdetropencestempstroubles. ou l’espagnoll’allemand.LesujetadéjàétédebunképarAdrienenseptembremaisunepiqurederappelne VA etBusinessNewsneprécisentpasquel’enseignementdel’arabeserauneoption,aumêmetitrel’anglais dans lecontenuexaminé? Pouvez-vous apporteruneoudespreuvesindépendantesquicorroborentceestdit Résolue par Résolue par L’article sourcedeBusinessnews(sitetunisien) https://www.valeursactuelles.com/societe/najat-v-belkacem-signe-un-accord-pour-enseigner-larabe-en-primaire-78207 Un articledeValeursactuellesquitiresasourced’unsitetunisien Pouvez-vous garantirqu’ils’agitdupostinitialsurlesujet? larabe-dans-lesecoles-francaises,520,71290,3 http://wwwbusinessnews.com;tn/neji-jalloul-et-najat-vallaud-belkacem-signent-un-accord-pour-enseigner- Pouvez-vous établirl’authenticitédel’auteuroulapublicationàl’origineducontenu? Résolue par Résolue par

25 verification process behindit,alabelthat categorized the with thelogosof newsrooms whohad participated inthe both French andEnglish onCrossCheck’s website, along AFP journalists, the cards were published summary in and figures ontheimpact ithad. After afinal review by steps of theverification process, thesources of therumour, editors card would that thenwrite includedthe asummary receivedIf astory at leasttwo endorsements, project concluded, could review journalists the verification card and add their media’s logoif they astory. wished to endorse journalists could discussanddetermine whowas bestsuited to contribute to adebunk. Whenthe verification process was Any updates to theverification card were sent automatically to Slack. Once the verification process had been launched, CrossCheck - November 2017 Message in#crosscheck 6 replies A REPRODUCTIONFROMSLACK E Thread 1 1 project/590/media/4396 https://checkmedia.org/crosscheck-france/ veut-il supprimerlesallocationsfamiliales? Hello, besoind’uncrosscheckici:Macron Apr 28that9:21AM _ agreed withtheverification to add theirlogosifthey work undertaken onthis asks other newsrooms A Le Monde M particular story.particular lemonde Figure n°12 1 1 own platforms, aslong asthey credited CrossCheck. opportunity to publish CrossCheck’s content ontheir newsroomssocial media.Participating also had the videos,creating andpublished explanatory them on short In addition, astheproject advanced, project editors began social mediaandsentinemailnewsletters. and comprehensible to theaudience. Stories were shared on debunk, andapicture to makevisually appealing thestory 1 1 1 1

26

27 CrossCheck - November 2017 Metrics 4

28 TOTAL PAGE LIKESTOTODAY: 180,540 language site, 5% to theEnglish site. Unsurprisingly 95% of thetraffic wentUK. tothe the French Paris), 2% from Belgium,1.5% from theUS, and1.25% from other sites. 87% of thetraffic came from France (24% from and 42%ofthe trafficcame from linksembeddedon came to the16% website directly, 30% was from social returning visitors). 11% ofthat trafficcame from search, views, whichcame from 335,993 visitors (83,146 were Overall, theCrossCheck website received 590,443 page W totalled 276. partners published theirown CrossCheck andthey articles On theFrench site, 67stories were published. Newsroom ARTICLES PUBLISHED EBSITE METRICS 100K 150K 50K 0 likes over the3months MAR Total Facebook page Figure n°13 of theproject 07 14 21 28 APR 04 335 993visitors." views, whichcamefrom received 590443page "The CrossCheckwebsite government watch list. flour bombed by aman whoclaimed to beona The mostpopular was avideoofFrancois Fillon being wellAs discussedvideosperformed onFacebook. very graph. TheadonApril11. spendstarted spend (provided by Facebook) isclearwhenyou lookat this reactions and18,669 shares. Theimpact of theuseof ad are considered, CrossCheck content received 51,865 When CrossCheck postsonpartnerFacebook pages weeks and1,207,64210 videoviews intotal. followers whichincluded347,800 engagements over the CrossCheckThe Page onFacebook had 180,598 FACEBOOK METRICS 11 18 25 02 MAY Total PageLikes 00 16

29 housing reserved exclusively for migrants. demolished inFebruary was goingto bereplaced withsocial debunked aclaimthat aParis lighthouse that had been detailed informationIt includedvery about how CrossCheck Behind‘CrossCheck the Scenes: Learn how to Debunk.’ The first newsletter sentonApril 20 had thetitle subscribers and35% openedthenewsletter. one priorto each elective round. There were justover 500 The project sentout 2newsletters inFrench (both andEnglish), NEWSLETTERS and 6,158 retweets. while asmallnumber of followers, itreceived 1,919 mentions The CrossCheck Twitter page had 4.956 followers, and TWITTER METRICS CrossCheck - November 2017 T par précautionentre2012et2015, aprèsplusieursvoyages. l’homme quiacouvertFrançoisFillon defarinen’estpaschéS.Ill’aété Contrairement àcequ’armaitdansunpremiertempsValeursactuelles, CrossCheck: L’enfarineurdeFrançoisFillonest-ilŽchéS ? his videowasusedin1otherpost. L’enfarineur abienfaitl’objetd’une che S,maisentre2012et2015. popular CrossCheck Metrics for themost Figure n°14

video verification process: through thedifferent steps of the The newsletter walked readers from theproject. On May 4, asecond newsletter was sentout withhighlights Where? When? andWhy? 5 4 3 2 1 Publishing the story Crosschecking theinformation withnewsroom partners. Completing the verification Who? checklist: What? Contacting thoseconcerned directly. Researching theoriginalsource of theinformation. People Reached Audience andEngagment Top Location Top Audience Post Engagment Unique Viewers TOTAL VIEWS:800K Î le-de-France Men 25-34 1, 072,513 642, 259 496

30

31 CrossCheck - November 2017 Research Method 5

32 CrossCheck wascentraCrossCheck l editors of asproject students,journalism participation whose from zeroranged to 34years. experience ofthe seven women andninemeninterviewed were recorded andanalysed thematically. journalistic The students to seniormanagers. The experience levels of interviewees rangedfrom journalism news stations, journalistic start-ups orwire agencies. ticularly active intheproject andworked onlinefor television availability. We mostlyinterviewed journalists that were par experience, organizational position, type, and, ofcourse, media sample wereThe criteria degree ofparticipation, Facebook page. Though isrelatively thesample small, itis questions to CrossCheck orcommenting onCrossCheck’s hadAll interacted withtheproject by eithersubmitting Participants were contacted by emailorthrough Facebook. withmembers of7 semi-structured interviews thepublic. questions, whichreceived 29responses. We alsoconducted For theaudience research, we usedanonlinesurvey of 50 ACCOUNTS OFTHEAUDIENCE 2017—a few weeks aftertheproject ended guide, were conducted withjournalists andeditors inJune A total in-depth ofinterviews, 16 based onanad hoctopic IN CROSSCHECK ACCOUNTS OFMEDIAPARTICIPANTS studies withtwo sets of participants: This reportisbased ontwo research public whoengaged withtheproject. that participated inCrossCheck. 2 1 Asurveyandfollow-up interviews withmembers of the Semi-structured interviews withjournalists andeditors 30 .)

(These participants include participants (These 29 . Theinterviews - the project for several reasons. representative of thosethat were interested particularly in qualitative examination ofthe different samples. quantitativeThis analysis was augmented with amore of thepublic, andnews referring articles to CrossCheck. of theproject, questionsthat were submitted by members CrossCheck’s Facebook page, tweets containing thename It was usedonfour differentsamples: comments on analysis based ontheReinert lexicometric method. These two studies were completed by anautomated discourse results of thestudy. sample. limitation whenitcomes Thisisanimportant to the did notwith interact the project are not included inthe CrossCheck’s content, especiallythrough Facebook, and Naturally, thosewhowere onlyoccasionally incontact with abouttheproject. survey complete alongonline discuss theprojectorto the invitationtoeither Second, theyaccepted on Facebook. a questionorcommenting bysubmitting participated First, samplemembers of theanalysed text. on alexical co-occurrence criterion andallows for aclearview of themainthemes into clusters based onlexical similarity. Theanalysis proceeds to group units textual intotexts segmentsbased onpunctuation andsize criteria, andthenclassifiedthem 31 _For thequantitative analysis, we usedtheIRaMuTeQ software that breaks down Appendix4for30 _See afull listof interviewees. Appendix3for29 _See theTopic Guideof theinterviews. 31

33 CrossCheck - November 2017 Journalists: theJournalists: project the participants the ViewThe of the as seen byas seen 6

34 fact-checking cross-partisan consensus to increase thecredibility of Whereas research hasrecommended thecreation of a COLLABORATION A election outcome. disinformation, andonlinepropaganda posed to the dueelections, to the that real risk political misinformation, was especiallytrueThis inthe context ofthe French audience trust andjournalists’ role asgatekeepers. eventually across-partisan alliance withthegoal of restoring the mediathat translated into asenseof responsibility, and collaboration was asenseof interdependency between newsrooms. What seemsto have initiallyenabledthe on collaboration between ahundred journalists from different One of theparticularities CrossCheck was that itrelied 1/ system of checks andbalances between journalists: of debunksquality andfact-checking by establishinga collaboration created circle avirtuous that increased the project. Conversely, somejournalists thought the it would potentially reflect upontheimage of theentire mentioned that ifoneorganization committed amistake, level.and international Interms of image, interviewees media organizations asanecosystem, both onnational underpinned by asense ofinterdependency between Ultimately, thelegitimacy of theCrossCheck alliance was and anillegitimate claimto objectivity. interpreted homogenization asafurther of perspectives danger of amainstream mediaalliance, whichcould be of credibility. But other interviewees were aware of the dating adebunk was seenasasignificant demonstration To someinterviewees, thenumber of newsrooms vali- PROJECT fact-checkers whoclaim to beobjective orneutral linked to debate about the biases enduring of MODEL OFHORIZONTAL

A FUNDAMENTALLY COLLABORATIVE 32 , thisendeavour isdouble-edged and 33 . As mentionedby oneinterviewee: moment,or'kairos'.can beseenasanopportune Thus, thistimeof highstakes and institutional instability from beinginfluenced by disinformation. ought to betaken to prevent the French 2017 elections 2016 U.S. presidential elections asasignalthat initiatives tional level. Theinterviewees unanimously mentionedthe This interdependence was alsonoticeable at theinterna - mediaoutlet." every —it’s alsopositivefor well-done positive—very And byhavingsomething class initsentirety. reflects onthemedia behave haphazardlyit "If somejournalists

35 identified in the context of participatory journalism. in identified the context ofparticipatory their gatekeeping role – alimitthat hasbeenpreviously pation, basedfor onthenecessity journalists to maintain differed.Some expressed scepticism about audience- partici But journalists’ opinionsonthisextensive collaboration or should not do social groups inthediscussionabout what journalism should academic debate that recommends theinclusionof different collaboration between resonates mediaactors withthe senseThis ofshared responsibility, andconsequent previously mentioned, hadaroleto inCrossCheck.) play latter as groups, two and the social mediaplatforms. (The was notbility just theirs; italso belongedto the audience democratic debate. They alsohighlighted that thisresponsi - from theirduties asjournalists to contribute to ahealthy that they had caused disinformation’s spread, but rather act. However, thisresponsibility stemmed not from abelief Thus, theinterviewees felt encouraged by aresponsibility to COLLABORATION ANDGATEKEEPING integration oftraditionalpoliticalparties." scores fortheFrontnational,andadis- news —withtheriseofpopulism,crescendo offake potentially beexplosiveinterms standing onabreedinggroundthatcould elections. We wereawarethatwe had allseentheprecedentofAmerican we wereinatimewhereusjournalists "The projectworkedwellinFrancebecause CrossCheck - November 2017 34 . of journalism upontheof journalism tech industry. increasing technological andeconomic dependency backgroundwas animportant issue connected to the the platforms’ control of thetools andfinances theproject Nevertheless, asmentionedby several interviewees, their gatekeeping role without interference. external funding. Consequently, were journalists ableto play journalists, but onlyprovided access to software and Google andFacebook didnot interveneinthework of over content. Indeed, whenitcame to CrossCheck, advantages of platforms, whileretaining full autonomy could enjoy thetechnological andfinancial beneficial ground for wherecollaboration, newsrooms Thus, somesuggested theneedto agree onamutually dependency upononlinedistribution platforms. of both theirsector’s financialhardships andincreasing formation to spread. Theinterviewees were alsoaware ground for andmostfertile disin- to be the primary fully neutral actor – considered Facebook andGoogle the exception of oneinterviewee whosawGoogleasa The logicbehindthisrequest was that journalists – with that platforms could have provided even more resources. have existed without theplatforms’ input, but alsosaid The interviewees recognized that CrossCheck could not and worthy offurther, cautious development. input ofplatformsThe was deemedindispensable COLLABORATION WITHPLATFORMS

36 There’s noone." to? whomdotheyturn lists. Otherwise, - to,likereliablejourna someone toturn I thinkit’s essentialthatthepublichas as media,torecreatealinkoftrust, "Since we’reinthismomentofreflection, to sayit’s false." we consideritinteresting that has30000shares— ren’t suckers.Something understand thattheywe- But wetriedtomakethem ingeneral. journalists It’s symptomatic of very only suckerswillshare?’ nobody reads,orthat that do youdebunkstuff colleagues like,‘Butwhy "We hadreactionsfrom building,community restore trust, andfoster dialogue: alternatively justifiedby at the needto increase efforts process information. Yet, theaudience’s participation was sional journalists, they are better equipped to address and was thus questionedby thosewhobelieved that, asprofes oftion. Theimportance theaudience’s participation andusefulnessshould decideonthevalidity of thisparticipa- However, someinterviewees alsobelieved that journalists disinformation, thus overcoming challengesof filter bubbles. could provide better of onlinetrends visibility andsignal Their accounts denoted anacceptance that theaudience role, advantage ofthe whiletaking audience asaresource. need to findastrategy that maintained their gatekeeping The journalists that werethe interviewed generallyreferredto COLLABORATION WITHTHEAUDIENCE - PUBLIC SERVICE VS.COMPETITION Cambridge: Cambridge University Press. Wiley-Blackwell andO'Neill, O. (2002). Couldry,34 _See N., & Ward, S. Why mediaethics J. stillmatters. (2013). A. Review,162-180. 25 (2) 33 _Uscinski, J., & Butler, TheEpistemology R. of (2013). Fact Checking. Critical findings from socialscience. New America Foundation. 32 _Nyhan,B. & Reifler, Misinformation (2012). andfact-checking: ResearchJ.A. me onadailybasis." something thatmotivates scoops. It’s really medium, I’mlookingfor When Iworkformy I workedwithothers. competition intheway "I lostmysenseof scoops, itseemedabitcounter-intuitive." when media’s businessistoracefor "Having acollaborationbetweenmedia borative thing." is selfish.It’s notacolla- "By definitionjournalism understandably seenasasignificant collective achievement: scoop'. Ultimately, overcoming theobstacle of competition was considered devoid apublic service of theusual'race for the CrossCheck was fact-checking, whichthey essentially to work against each other. Second, thecore practice of by two factors. First, thestakes were too highfor journalists issue of competition between newsrooms was overcome of public service. Indeed, theinterviewsdenoted that the on fact-checking underacommon, overarching sense to temporarily overcome thisobstacle by focusing theirefforts competition. However, CrossCheck’smanaged participants An obvious obstacle to collaboration between newsrooms is (Reith lectures ; 2002). Aquestionof lectures ;2002). trust (Reith

37 out to be completely false. out tobecompletelyfalse. having workedonitlike us.Butitturned CrossCheck ontheirwebsite without their laurelsandpublishtheresultsof We feared thatotherswouldreston themost. identify whowasparticipating ning. Afteraweekofwork,wecould involvedfromthebegin- "AFP wasvery resources, according to thesituation: contribution dependedontheadequacy of theirskillsand participant’s seen asbeingbalanced by thefacteach that However,participants. thisuneven contribution was noticedgradually unequalcontribution from different viewees their mentioned fear of'freeloaders', asthey individual journalists’ choices. Inaddition, some inter- explained by hierarchical pressures andrigidities, rather than But thislack of participation by somenewsrooms was as someparticipants didnot seemto fully cooperate. Indeed, thesenseof competition was stillfelt by interviewees, as'coopetition'. ofsituation type greater societal benefits. Oneinterviewee qualified this of reachingthe possibility larger audiences, hence bringing organizations. Theusualrace for scoops was superseded by come withinasituation that was mutually beneficial to the Thus, by joiningforces, competition was temporarily over by falseinformation." and helpitifisassailed opinion inahealthyway, of democracy, the inform sure asoundfunctioning estate,toen- the fourth lism’s foundationstobe - linked tooneofjourna dimension ofpublicutility is notascoop.There’s a "Debunking fakenews CrossCheck - November 2017 - challenged accepted ofcompetition. notions but alsoreinforced the and value ofpublic service only gained prominence inthedailypractices of journalists, the provinces." concerned to contributewhenrumors rhythm ofnews.Localmediawereable It cameinwavesanditfollowedthe News andJournalism. London: Palgrave Macmillan. listic ethics inachangingmediascape. InMiekle, G.andRedden, G.(Eds), OnLine and Fenton, N.,Witschge, T. (2009). Newspeople Should Know andWhat thePublic Should Expect. New York: Crown Kovach,35 _See B. andRosenstiel, T. (2001). what moment." were speakingandat we knewwithwhom by nameandmedia, We werewellorganized inoneclick. everyone that wehadaccessto "The advantageswere instantaneous andmultimodal throughout theproject: in newsrooms across France were andtheUK.Discussions instantaneously, even though they were physically scattered app, Slack, which allowed alljournalists to discussanddebate Collaboration about CrossCheck took place onamessaging the mediathat translated into asense ofresponsibility. collaboration was asense ofinterdependency between newsrooms. What seemsto have initially enabledthe on collaboration between from different ahundred journalists One of theparticularities CrossCheck was that itrelied 2/ journalistic identity literature review andinterviewees asafounding valueof Fact-checking andverification have beendeemed by the COLLABORATION

A MODELOFHORIZONTAL 35 . Inthiscollaborative context,itnot Comment IsFree, Facts Are Sacred: journa- TheElementsof Journalism: What

38 expertise, whether that beatopic, alanguage orskill. ifithappened to be fromspecific story hisorherarea of advantage. Itbecame natural for aparticipantto debunk a and thediversity of profiles endedup turning into an were revealed at theindividualandorganizational level, Trust was established aseveryone’s different capacities of thegame." standards. Butthatwasalsothepurpose from mymedium,andwhohasdifferent someonewho’sused totrusting not "It’s beencomplicatedattimestoget they would not normallywork: other, andshare theirwork withindividualswhom contributors. Thus, journalists were obligedto trust each cipation was unequaldueto thedifferent profiles ofthe cipant’s resources andskills. Asexplained before,- parti dependedoneach- parti The decisionsto debunk astory same direction." wanted tomoveinthe to seethateverybody from thebeginningwas really mademeexcited "What Iperceivedthat that ofajournalist." waslessconsideredthanof anintern dynamics. Ididnotfeelthattheopinion few hierarchical "There werevery and aimedat reaching aconsensus: The discussionswere courteous, spontaneous, fluid, factual to experienced fact-checkers andjournalists. of hierarchy – even whenparticipantsrangedfrom interns wereDiscussions describedashorizontal – without asense array of disinformation: capable person, participantswere ableto cover avast joiningindividualforcesBy anddelegating to themost consensus about them: were establishedafterdiscussion,andonly ifthere was a accordingsituation to the resources available. Methods to adapt for to eachvaried, journalists itwas necessary the project. Since the form andimpact ofdisinformation would work, theworkflow was constantly adapted during verification skillandanunderstanding of how CrossCheck three-day training to ensure everyone similarlevels of the collaboration. Whileallparticipantshad received a The absence of strictly definedrules was also central to do itthebest." to thepersonwhocould a questionofdelegating butitwasmore method, rences intheworking - "There weresomediffe veracity ofinformation." maximum ofpossibilitiestoverifythe aggregating abunchofskills,wehad "What wasinterestingthat,by with ourglobalnetwork." We theAFPwereuseful deciphering information. by theirexperiencein deurs werelegitimatized in socialmedia.LesDéco- "BuzzFeed wasspecialized

39 participants relied onaworkflow that resembles theidea Instead of astandardized code of conduct, CrossCheck’s capable andknowledgeable, dependingonthesituation. reasonable debate andthedelegation of tasksto themost Thus, acollaborative modelemerged through horizontal, to sayitwasfalse." I wasright.Butforthemittoofast fast. Theproofis,acoupleofhourslater, toldme,you’regoingtoo journalists false, tothepointthatsometimesother energy toknowthatapieceofnewsis flair, clinically. Personally, itgivesme people workdifferently, somedowithout "That’s alsowhenyoudiscoverthat online debunking, whourged for caution: their andthose instinct with more experience with created divisionsbetween thosewhowanted to follow of disinformation. Different methods between journalists led to heated debates, dueto thediversity andcomplexity was apointof discord between journalists. Itsometimes principle of theCrossCheck project was prudence, but this But consensus was not easilyreached. Anoverarching sometimes." and evenusetricks time, wehadtoadapt methods andideaseach obviously. We hadtofind case offakenews, "It dependedoneach working." when werealizeditwasmoreorless processes thatwedefinedgradually ning. We byfumbling.Therewere learnt "It wasn’t establishedfromthebegin- CrossCheck - November 2017 with afocus onthepublic good common journalistic knowledge, without coercion, and were given to deliberate, equalopportunities from relatively conditions ofthe idealpublic sphere: equalparticipants practices intimeandspace. match the parameters These capacities andindifferent places, peopleand thus unifying present discussions, andto individualswithdifferent flexibility. In addition, Slack enabledaccess to pastand in theeffective conduct of work andallowed sufficient discussion andadaptability. Thisprinciplehad priority of another potential obstacle to ethical journalism forced theirautonomy versus that of theproject. Immediacy, content ontheirown mediaasapositive feature that rein- freedom to decidewhether ornot to publish CrossCheck’s were autonomy andimmediacy. Interviewees mentionedtheir Two otheringredients important of thecollaborative model mentioned by many astheultimate gauge of success: intheerror debunks. Indeed, theabsence of errors was because itwas rewarded by the absence ofasingle to working infast-paced news agencies orTV, particularly on verification was appreciated by somejournalists used ditional investigation. However, theamount of timetaken fact-checkers whooften recommended prudence andad- were alsoobligedto slow down andadapt for experienced quickly, because of thehierarchical pressures to publish, in addition to theirregular job. Journalists usedto working from other newsrooms, whowere working for CrossCheck to beslower whilewaiting for theapproval of journalists Fact-checking, whenconducted collaboratively, was noted Interviewees diverged regarding how to bestusetime. point 7. What kindof discourse didCrossCheck generate? debunk thesewas alsojustified by the public’s demand. For more detais onthat see also about rumours concerning Emmanuel Macron, whichmeansthat thechoice to 39 _At of timethebiggestpart theaudience’s thesame questionsto CrossCheck was The London of School Economics andPolitical Science. 38 _Tambini, Damian, & Fake Goodman,Emma. (2017). news: Public policy responses. Hampshire ;New York, NY: Palgrave Macmillan. Houndmills,A.,& Basingstoke, Ethics Madianou, of 39-56). media(pp. M.(Eds.), _Couldry,37 Living well withandthrough N.(2013). media.InCouldry, N.,Pinchevski, Mode of access: http://plato.stanford. edu/entries/habermas. 36 _Bohman,J., & Habermas, R. W. J. (2010). and the time-intensive offact-checking nature between seen asachallengeby theparticipants dueto thetension 'phronesis', apractical wisdombased onprudence, the to react quicklyagainst necessity rumours Stanford Encyclopedia of Philosophy. 36 . 37 , was also , was also 38 .

40 and between journalists andtheaudience. Transparency was demonstrated both between journalists, through the alliance. aforementioned cross-partisan and transparency. was meantto Credibility beachieved nature of disinformation, but itwas reinforced by credibility the of As apractice, itnecessarilyfaced challengesbecause Objectivity, asaclaim,was questionedby interviewees. isintrinsicallyObjectivity linked to credibility andtransparency. 3/ over immediacy made journalists more confident. pressurestime – as‘fake news’. Here, favouring quality – are not immune, from whichjournalists especially under They mentionedhow theaudience tends to consider mistakes This absence of mistakes reaffirmed journalists’ purpose. mess up." best isthatwedidn’t "To mewhatworkedthe intoday’sluxury journalism." until it’s fullyverified.It’s anincredible works inurgency. We wouldnotpublish while, comingfromamediathatalways able tosaywetakeourtimeonceina pleasanttobe "I thoughtitwasvery full-time onCrossCheck." wasn’t alwayssomeone had otherneeds,there hours. Butthenewsrooms would comeafterseveral the validationonSlack a bitslow. Sometimes "Maybe CrossCheckwas CREDIBILITY ANDTRANSPARENCY

THE QUESTIONOFOBJECTIVITY, journalists werejournalists happy to share their work with other peers. verification, soas to increase trust withthe audience. In addition, in theirdebunks, andto allow readers to follow thethread of Participants had agreed to includeasmany sources aspossible depth intheir work: They their justified choices by and onquality insisting partiality, but ignoringitwould goagainst theirdeontology reinforce theaudience’s assumption concerning themedia’s dilemma: debunking allof disinformation onMacron could at Emmanuel Macron. CrossCheck’s participantsfaced a by theoverwhelming amount of disinformation directed and credible, CrossCheck’s claimtowas objectivity challenged But even though participants had intentions to betransparent the readers." maximum ofsources,for "We insistedonputtinga to becredibleintheverification." sential me. Transparency it’s injournalism, es- other newsroomswasn’t aproblemfor "To sharemyverificationworkwith say that journalists defendhimanyway."say thatjournalists we reinforcethatcircleinwhichpeople whichdon’tof disinformation exist.Ergo is thatwe’renotgoingtoinventcases cessing impliesbalancing.Theproblem pro- complicated becauseallinformation it’sa trap.Obviouslyforusjournalists any othercandidate.Thatwaskindof targeted EmmanuelMacronmorethan "Factually, thebulkofdisinformation 39 .

41 believe inobjectivity." that Ipersonally don’t basically our job,knowing possible, whichis way tial things inthemostimpar things, inhowtosay in theinterpretationof All oftheproblemlay to defineathreshold. "We’ve nevermanaged processing." information that sometimestheycarriedabiasinthe Ihadthefeeling and politicalaffinities. wealso haveouropinions "As journalists unreachable, aprudence principlewas prioritized: biases and, asthey considered that absolute was objectivity discussions withpeers. Journalists seemedaware of their for participantstothrough question theirsubjectivity level ofvirality. Thesedisagreements were anopportunity constantly debated andits dependingonthe story threshold that were definedwhen to debunk astory thus of peoplepotentially get confused. methods The and one segment,orignore itandlet hundreds of thousands debunk thesatire andappearto state theobvious to people. Here thejournalists faced another dilemma: checked, itwas often shared by asignificant number of may seemobviously ‘fake’, which was adivisive issue. satirical, Some false information Another challengeto CrossCheck’s credibilitywas satire, cover quitevasttopics." it showedthatwetriedto we didqualitywork,and we fellinthetrapbecause "Honestly Idon’t think CrossCheck - November 2017 but whenthelevel was of virality - newsroom foryears." we accumulatebystaying inthesame yours andtohighlight reflexesthat other mediaitallows youtoreevaluate cation techniquesandsensibilitiesfrom "To beconfronted withothers’verifi- the content they mightencounter: tools inspired ahealthy scepticism intheparticipantsabout which facilitated theverification process. Mastering these NewsWhip, CrowdTangle, andGoogleReverse Image Search, also required theuseof technological tools suchas independently oftheir previous experience. CrossCheck Thus, their skills,flairandefficiency were increased, adapt to each other to finda consensus. platformsame obligedthemto find common ground and by themediumfor whichthey worked. Collaborating onthe working routines were shaped, andsometimes restrained, paring themethods of peers, journalists realized how their by learningfrom oneanother. Through witnessing andcom- levels. Most journalists mentionedgaining invaluableskills disinformation, at both the individual andorganizational CrossCheck beyond had brought benefits addressing After interviewingfifteen participants, itbecame clearthat 4/ phenomena: hoc attitude, by highlightingtwo academic debate to call for anad These accounts gobeyond the journalist, andfrom journalists to theaudience. to restore trust through transparency — from journalist to audience received thisapproach, they stillhighlightastrategy milieu. Whiletheseinterviewscannot demonstrate how the constitutes quitejournalism asociallyhomogenous industry dized fact-checking methods. NEWSROOMS, ANDTHEPUBLIC 2 1

Theunavoidability of biases, especiallysince themodern of maintainingcompletely Theimpossibility standar IMPACT FORJOURNALISTS, -

42 which isUGC." speed onaworkthatiscloselyrelated, and and itenabledmetogainefficiency in myprofession,fact-checking, "I strengthenedmyreflexes,Iprogressed with readers." that triestorebuild trust project likeCrossCheck, our brandtoa beautiful a priori,wastoassociate my boss’s motivation, "For themedia,Ithink probity constituted astrategic promotion: media. To bepublicly associated to CrossCheck andits reflected onnewsrooms, andontheimage of themainstream The positive aspects of thiscollaborative project also to government officials. online propaganda, andthe ofCrossCheck social utility debunked by CrossCheck to demonstrate the salience of things. Participants alsomentioned how theystory had useda journalism isaprofession where oneconstantly learnsnew they gained, but they denoted by humility highlightingthat journalists felt lessenthusiastic about theindividualbenefits collaborations. Understandably, themostexperienced web built lastingnetworks andare considering future pants internationally. Now acquainted with each other,- partici their own newsroom, withinthefield of fact-checking, and The professional of participants increased visibility within many moretopics." that I’vebecomemoreskepticalon really enlightensmywork.Ialsonoticed although they’rereallysimple.Ithinkit "These arereflexesIdidn’t havebefore,

43 and convince: verified information – are also the mostdifficult to reach segments media. Other – those ingreatest needof audience that isalready convinced andtrustful of mainstream Furthermore, fact-checking tends to beaddressed to an such asbelief perseverance andselective exposure posed. noted thesignificant barriers that psychological mechanisms increased theirefficiency andimmediacy, oneinterviewee Although additional timeandfinancial resources could have ofsome parts theaudience were out of reach. tation, CrossCheck’s material limitations, andthefact that aware that itmightbelimited dueto fact-checking’s repu remains CrossCheck’s purpose primary – journalists were With regard to impact ontheaudience – which,afterall, segments of theiraudiences. all of thesewere questioned andattacked by politiciansand responsibility, purpose, andprofessionalism, intimeswhere self-perceptionjournalists’ the by theirsenseof asserting and even regretted itwas over. Itpositively reinforced this collaborative experience, positive lauded energy, its interviewees,The unanimously, were enthusiastic about was enriched. media, whoseexperience andknowledge of disinformation was even useful for the fact-checking mostestablished intellectual to integrity newsrooms’ audiences. CrossCheck interfere withthenewsroom’s own visibility, but conveyed may have worked because CrossCheck’s image didnot traffic on CrossCheck-related content. Thus, the collaboration some partners reported gaining asignificant amount of They received positive feedback from theiraudience, and publishers, thisfeature increased theirbrand’s visibility. that they had helpeddebunk. Particularly appreciated by Newsrooms their could insert logoonCrossCheck stories CrossCheck - November 2017 will nevertouch." becausetherearepeoplewe further It’sof falseinformation. togo difficult restricted public,who’s alreadyaware "Fact-checking isaddressedtoa - having platforms provide more exposure andfunds. counter disinformation quantitatively, andthe ideaof the ideaofbuilding anetwork offaithful followers to They alsosuggested solutions to increase impact — like work andthepractical advantages of collaboration. limited reach was counterbalanced by of their thequality The interviewees’ disappointment regarding CrossCheck’s ourselves bescared." doesn’t meanwecan let always beadanger, that about media.Therewill a populationsceptical "There willalwaysbe

44

45 CrossCheck - November 2017 public received, used Audience: how the the ViewThe of the and interpreted CrossCheck 7

46 pleased andcurious about CrossCheck anditsimpact. semi-structured interview. Respondents were unanimously to theonlinesurvey, andanadditional seven tookina part on theproject website. Intotal, twenty-nine peopleresponded and emailswere sentto everyone whosubmitted aquestion requests were posted ontheCrossCheck Facebook page, In order to recruit members of theaudience for thisresearch, singly wideimpact for CrossCheck. and verification, there are elementswhichpoint to asurpri- of few users whowere already familiar withmisinformation While thisqualitative analysis isonlybased ontheaccounts more diverse thanexpected. interviewed suggesttherespondents’ backgrounds were far process for thisaudience research, theaccounts of those was asignificant element of self-selectionsampling inthe them to theirpeers. Whileit’s to note important that there by the andtook action journalists, by communicating debunks inalignmentto the way they were produced messages inacritical way by focusing onsources. had taught them skills,asthey felt they could now ‘read’ CrossCheck content. that of part thisknowledge was aresult of consuming motivations, targets, andorganization – andexplained knowledge about the ofdisinformation – agents their 3 2 1 Third, they interpreted CrossCheck’s approach and Second, the interviewees explained that the project they First, demonstrated asignificant amount of 1/ time, because and ofhateful comments While mostconsidered suchengagements to beawaste of comments of Facebook postsabout anews articles. sometimes engage inconversation withstrangers inthe French citizen. Two-thirds of respondents declared they and active ontheinternet andsocialmediathantheaverage We can thus assumethat therespondents are more connected use to sometimes postnews orcomment articles, onposts. allrespondentsAlmost have aFacebook account that they their mainsources of news. analyze this later), many mentionedFacebook asoneof and mediasuch international also mentioned asThe Guardian such asLe Monde, Libération andLe Figaro, but many information were French established mainstream media mobile phonesto consume news. mainsources Their of of respondentsming majority usemostlycomputers or Regarding their generalmediaconsumption, anoverwhel- out of civicduty. respondents they said are moderately interested inpolitics others didnot feel of at allengaged Amajority inpolitics. declared themselvesSome respondents while militant, In terms of political engagement, answers were alsodiverse. of participants whoidentifiedwiththe minority far right. Political orientations covered awidespectrum, withasmall and theyoungest amongthemwas 35years old. However, of respondents amajority had auniversity degree gender, occupation, area of residence andeducation level. Respondents’ profiles diversewere in terms very of age, on Facebook was more open andintelligent. their opinionandlearnfrom others, andwishedthedebate of convincing other individuals, they didlike to express

The New York Times. Interestingly (and weshallfurther PROFILES OFTHERESPONDENTS

the impossibility

47 audience: CrossCheck ofthe managed to reconcile two parts colours, but alongsideshort detailed andfactual articles, debunked. Overall, by usingaclearsystem of logosand a quickunderstandingof ofdisinformation thetype being pointed to thehelpfulness of visualicons ingivingaudiences high level of trust inthework of CrossCheck andalso headlines. Theirexplanations for thisbehaviour suggested a intheirentirety.the articles However, someonlyread the ofAn overwhelming respondents majority preferred to read they had encountered. week, whileathird didsodependingonwhat disinformation Most respondents checked thewebsite day every or appear systematically intheir newsfeed. of them followed CrossCheck, nordidCrossCheck posts Even had aFacebook whenrespondents profile, not all a project from URLapart thathaving of sites. thepartner visiting thewebsite, of whichhighlightstheimportance Most respondents preferred to follow CrossCheck by directly 2/ CrossCheck - November 2017 stuffed with hyperlinks' methods of verification. CROSSCHECK 2 1

Thosewho'are bored with reading long articles Thosewhoare eager to learnabout thedetails and THE RESPONDENTS’USEOF 40 been-an-explosion-of-international-fact-checkers-butthey-face-big-challenges/415468/ they face bigchallenges. Poynter. Available at: http://www.poynter.org/2016/theres- (2016) A. 40 _Mantzarlis, if youonlyread thetitle" saying. Ifnotit’s useless, in whatthedebunksare have gottobeinterested hoaxes aresaying,you interested inwhatthe entirely.articles Ifyou’re "Yes, I’vereadallthe enough forme" issignaled asfake,that’sinformation attention onascreen.Aslongasthe I’m havingahardtimefocusingmy In general,I’mnotabigreaderbecause false... in, thatisitwasfalse,partly and logosonthetopicsIwasinterested "Personally Iwascontentwiththesigns true" if theysayso,itmustbe and soIthought,well, know why, butIhadtrust enough forme.Idon’t thatwas false ortrue, tant, buttosaythatit’s the verificationisimpor opinion wasjust…Imean point. Thepointinmy because thatwasnotthe I readentirely]?Zero, did "[How manyarticles There’s beenanexplosion of international fact-checkers, but -

48 CrossCheck arrivedat thebestmoment" frommisinformation. information so Iwantedtobeable todistinguish media andI’mquitepolitical, activeonsocial CrossCheck. I’mvery "I waslookingforatool/websitelike mess" to tidythiselectronic and gossip...Itwastime ofstupidity supermarket " it’sof falseinformation, something" presidential electionviathediffusion even theworld.To influencea couldhaveonanation, disinformation "I thinkIrealizedthehugeimpactthat was reliable. a gap by helpingrespondents to determine what information Thus, CrossCheck camemoment,andfilled at anopportune potential impact ontheFrench presidential elections. CrossCheck was launched, andwere preoccupied withits had allheard about theproblem of disinformation before contacts. Having followed theUSelections, therespondents dis-information between theirfriends, family, orFacebook available andwitnessing thefast spread of mis- and include beingoverwhelmed by theamount of information The factors that ledtherespondents to follow CrossCheck CROSSCHECK MOTIVATIONS FORFOLLOWING Internet hasbecomethe Internet dents weredents overwhelmingly positive about the project. and purpose were not necessarily clearto them, respon- report asdubious. However, even ifCrossCheck’s identity Fact-Checking Initiative, that whichflagsusers articles also confused CrossCheck withFacebook’s Third Party alliance, andwhofunded CrossCheck. interviewees Some Respondents were unsure of whowas included inthemedia (which they alldid). know about CrossCheck untilthey informed themabout it Most respondents mentionedthat theirsocialcircles didnot project’s andpurpose. identity could have been,andthere were questionsraisedabout the ‘brand’ meantitsaudience was smallerthanitotherwise Certainly, thefact that CrossCheck was acompletely new project, mediacoverage andofexternal . from partners thewere during pushed particularly last weeks ofthe ofFacebook the highlights This importance ads, which networks to whichthey already belonged. about theproject through hoax-busting orverification discovering theproject through theirown research, orhearing via Facebook orotherreasons newsincluded media.Other The respondents first learnedabout CrossCheck mostly CROSSCHECK HOW THEYFOUNDOUTABOUT consume" I of theinformation need toverifythequality elections, I’vefeltthe rican, BritishandFrench tensified withtheAme- the phenomenonhasin- "Since manyyears,and

49 information" politically neutral toverify something that’s really at all,soIreallyprefer but ImeanLibérationnot editorialneutrality,certain Le Mondeasbeingofa "I stilldoconsider media alliance. the image of independence andcredibilityfostered by the option forcomplementary verification, precisely because of and Désintox. However, CrossCheck was seenasadifferent, verification organizations in France, Décodeurs suchasLes were aware of theexistence of other fact-checking and providing neutral andreliable information. respondents All useful intheway itresponded to theurgent needof one.as adurable They considered CrossCheck to bevery stopped, since they saw the problem ofdisinformation They were worried about the fact that the project had focused onlyonthepresidential elections. Most respondents didnot understand that theproject was work" a realjournalistic ofneutrality,there wasarealeffort targeted thanothers,butIthought I alsoassumethatsomeweremore Really allcandidateswerecovered. not justinfavorofonecandidate. itwas debunked. Itwasnotpartisan, neutral inthewayfakenewswere "Actually Ithoughtthatitwasquite As onerespondent explained: objectivity, neutrality, reliability, andefficiency. positive.was generallyseenasvery Itwas thought to reinforce The fact that CrossCheck constituted analliance of media HOW THEYUNDERSTOODCROSSCHECK CrossCheck - November 2017 and gained more than 1.2 millionviews. short explanatory videos, whichwere posted onFacebook confirmed by the success of CrossCheck’s further is cation appealoftheThe clearmethodology andsource- identifi and to bemore cautious before sharinginformation. misinformation, because ittaught themhow to verifysources made respondents feel better equippedagainst elements through whichchannelsrumours spread. Thesharingof these on verifyingit,CrossCheck helpedtheaudience comprehend thesourcesarticles of misinformation, andwhohad worked of news sources. clearlyandaccurately By inits identifying in effort researching andmotivations and theidentity verifying alternativesional media, andtheirlack andcitizen of media) differences betweentypes of publishers (e.g., profes- , ignore to tendency with young studentshighlighted their Facebook asasource of news. Onerespondent whoworks Facebook some respondents’ comments. Indeed, someconsidered American Educator. That’s BiggerThanFake News: Teaching Students to Engage OnlineReasoning. inCivic 4 trouble-judging-credibility-information-online Available at: https://ed.stanford.edu/news/stanford-researchers-findstudents-have- credibility of information online, Stanford News Center, November 22, 2016. 41 _Brooke, D. (2016), Stanford researchers findstudentshave trouble judgingthe internet hasbeenraisedby academics the sources ofidentifying The difficulty ofinformation onthe APPRAISAL FACT-CHECKING SKILLSANDCRITICAL beyond thesphere of fact-checking’s onlineaudience. typical consumers ofCrossCheck spread stories the debunks to critically appraisecontent. second The isthe fact that fact-checking skills,like sources to identify theability and success for CrossCheck. is first The typical the acquisition of by respondents that pointto long-term andwide-spread In terms of impact, two elementswere unanimously mentioned 3/ of the project. this concern didn’t translate to the public’s perception about inevitably having to cover more cases about Macron, While journalists, asexplained before, were concerned 2 _McGrew, S., T. Ortega, J. Breakstone & S. Wineburg, (Fall 2017) TheChallenge

IMPACT as amediumrather than aplatform, talkingabout 41 , and this echoes , andthisechoes

50 had gained any reflexes thanks to CrossCheck. of mis- anddis-information whenasked whether they these skills. characteristics identified Many respondents with CrossCheck consumers, itseemsthey alsodeveloped reflexes fact-checkers have, andjudging from theinterviews papers’ presentation" "Yes, theURLaddress andthenews- 3. Writing style" influence ifhehasany) media (andhispolitical 2. Lookatwhoownsthe "1. Lookatthesources As explained by arecent study towards, for example, emotionally appealingheadlines. other pages. Italsopushed therespondents to besceptical reading, whichrefers to assessingasource by checking of information seems to have translated into ‘lateral’ Interestingly, the fact that the debunks detailed sources skills they acquired through theCrossCheck initiative. pondents highlighted other potentially durablefact-checking In addition to learninghow to source information, theres relevant,andright." very verification. ThatIthoughtwasvery, andthelinksto misinformation, whyit’sexplaining misinformation, is allthesourcingandworkof "Really whatIlikedaboutCrossCheck, relevant" the titlethatwas sourced. Itwasnotjust because theywere were interestingexactly "I thoughtthatarticles 42 , theseare thetypical - follow blindly." the mediawhichIusedto a criticalmindtowards enabled metodevelop easily. CrossCheck "Hoaxes spreadvery CrossCheck." I’ve becomemoreawarefollowing strikingtoo,Ithink particularly CrossCheck. Orelseimagesthatare I thinkthatI’venoticedthanksto that generatestrongemotions, catchytitles, "Often, therearevery have?" does thisnewspaper paper?reputation What whatnews- of thearticle, in detail,andtheorigin and lookatthearticle title), Igetsuspicious shocking (ofteninthe tolookextremely trying displayedis formation "Yes. Assoonasthein-

51 important actorsimportant intheevaluation of information researched, aswell asthefact that peers andfamily represent when peoplemake sense ofinformation hasalso been asafactor ofendorsement ofcredibility importance The are not motivated enough to it,andthat peopledonot usefact-checking ifthey services checking islimited to asmallaudience that isalready inclined Academics have highlighted thefact that thereach of fact- disinformation. opposing views whohad beeninfluenced by cases of that critically appraiseonlinecontent, mostsignificant is evidence In addition to providing fact-checking skills, andtools to S omissions ordistortion" imprecisions, voluntary Point...) oftenshowingas (le Monde/l’Express/le sites ofyourcontributors ontheweb- sinformation "I’ve seenalotofmi- everything" "CrossCheck hastaughtmetodistrust reinforcing theirown beliefs further and research skills, may endupdistrusting themediaingeneral of information sources, anddonot have enough critical individuals whoare asked to determine thecredibility media literacy backfire effect. Ithasbeenshown that However, other accounts alsoseemto pointat apotential CrossCheck - November 2017 Furthermore, ithasbeenshown that theesteem of people brands inthe of‘fake’ brands category by includingmistakes involuntary from trusted media express ablindscepticism towards the‘media’ ingeneral of debunks that they considered asproblematic, somedid of respondents could not recall any examplesinparticular OCIAL REINFORCEMENT the convince project helpedrespondents with peers 45 . 44 : 43 . While a majority . Whileamajority 46 . is reduced whenthey share false information consequences. friends andfamily from doingthesame, dueto itssocial weren’t sharingfalse information, andwanted to prevent respondentsSome explained they wanted to make sure they shared itsdebunks withtheirsocialcircle. Many alsodescribedthat they talked about CrossCheck and formation, both inface to face conversations anddigitally. peers from theirown intimate socialcircles spreading misin- According to theiraccounts, therespondents had witnessed Respondents’ answers theseconclusions. support facts". nated as"alternative where liesaredissemi- inacontext importance CrossCheck worksandits to severalpeoplehow but I’vealreadyexplained [about the]CrossCheck, and familydon’t know the majorityofmyfriends "Unfortunately, And anotherexplained: thatisfake." by ditchinginformation useful formetonotlooklikealoser false info,moreinthatsense.Itwas own leveltorelayrealinfoandnot wasfalse,andthereforeatmy stuff "[I]t enabledmetorealizethatsome As onerespondentexplained: time,same peopletrust news shared by others they know 47 . But, at the 48 .

52 contacts. the debunksto allmy "[I]t enabledme to share admitted: As onerespondent beyond thisaudience. consume fact-checking, theproject seemsto have reached audience was limited to individualswhowe would expect to thought differently. Thus, even if CrossCheck’s direct online dience members withtheconfidence to confront peers who CrossCheck’s debunks appears to have provided someau intentions" and peddlethemwithdishonest and thusfightagainstthosewhocreate withconcretearguments debunk rumours is usefulbecauseit]enablespeopleto And anotherexplained:"[CrossCheck during aconversation". couple ofarguments debate, andtohavea enabled metoopena stated: "[CrossCheck] As anotherrespondent researched arguments andageneralsenseof credibility. convinced by disinformation, asitprovided themwithfactual, when debating face-to-face withpeers whohad been Overall, respondents highlighted theusefulness of CrossCheck - Available at: https://shorensteincenter.org/combatingfake-news-agenda-for-research/ C. 4 “objectivity”. Journal of Communication Inquiry, 246-262. 36(3), 4 Communication Research, 41(8), 1042-1063. media: Endorsements source trump partisan affiliation whenselecting news online. and Messing, S., & Westwood, S. J. (2014). Approaches to Credibility Evaluation Online. Journal of Communication, 413-439(3), 60 4 mittee/fake-news/written/48215.html dence/committeeevidence.svc/evidencedocument/culture-media-and-sport-com- onFaketo theUKParliamentary Inquiry News, http://data.parliament.uk/writtenevi- 4 believe”: Audience perspectives onfake news. Reuters Institute. 4 Points https://points.datasociety.net/didmedia-literacy-backfire-7418c084d88d 4 intention." to them.Thischangedtheirvoting beforeIdemonstratedtheopposite true was convinced that[misinformation] "Many peoplearoundmewere debates had convinced theirpeers to changetheirvote: formation, somerespondents went sofar asto saytheircasual or not CrossCheck effectively countered theimpact of disin- study are inadequate to provide definite insightsonwhether peers’ misconceptions. size andsample of this Whilethetype was usedby audience members asasource to debunk their Through itsimage of neutrality andaccuracy, CrossCheck election." has hadanimpactonthe I genuinelythinkthatit and it’s widelyshared, the websiteisrecognized, Since thecredibilityof 7 _Marchi,.WithFacebook, R. (2012) blogs, andfake news, teens reject journalistic 8 _Lazer, D., Baum.M, Grinberg, N.,Friedland, Joseph,K.,Hobbs, L., W. andMattsson, Metzger,6 _See M.J., Flanagin,J., A. Medders, R. B. (2010). 5 _Written evidence submitted by InformAll andtheCILIPInformation Literacy Group 4 _Thisisageneraltrend NielsenR.K., :See Graves (2017). L., 3 _Boyd, d. Media (Jan. Did Literacy 2017) 5, Backfire, DataSociety: and (2017) Combating Fake (2017) News: AnAgenda for Research andAction. Harvard. Selective exposure intheage of social

Social andHeuristicSocial “News you“News don’t

53 then let themdowhat theywant". without calling forhatetowardstheothers, coherent and respectful totheirideas the knowthewebsitesof I don’t particularly chosphère" assoonyou’reanationalist. ofthe"fa - understand thatyou’repart idea aboutthe"fachosphère" , wequickly "If welistentothemediaand thegeneral One respondentexplained: these sites are more nuanced that mightbeexpected. some produced answers that illustrate attitudes towards were asked to definetheaforementioned 'fachosphère', However, what isinteresting isthat, whentherespondents editors, andtheircontent was regularly debunked. monitoring processes usedby theCrossCheck project Soral.as Alain Thesewebsites emerged regularly inthe include websites suchasFdesouche andpersonalities such spread alarge amount of false news and/or propaganda and they definedasanetwork of far-right websites that tend to 'fachosphère' asthe mainagent ofdisinformation, which Nevertheless, identified the ofrespondents the majority by alack ofmedialiteracy andeducation ingeneral. of respondents were highlyaware blamed 'naïve'Others too moguls'. dependenton'media was caused whenjournalists were not objective enough and observed. Afew respondents thought that misinformation towards French politiciansandthemainstream mediawas surprising diversity of opinionsbut, still, ageneralscepticism the spread of disinformation, therespondents showed a When asked about whothey thought to beresponsible for RESPONSIBILITY CrossCheck - November 2017 and if what they say is "fachosphère", andifwhatthey sayis citizens anditshould benoted that the consequences caused

54 were stillwitnessing thespread of disinformation. tonecessity continue theproject, dueto thefact that they CrossCheck could improve. Thisincludesabove allthe The respondents provided someindications as to how 5/ and useful to, awideraudience. that the project’s were mainprinciples by, understood situated, theaudience’s accounts would seemto suggest regarding the'bubble' inwhichCrossCheck may have been alliance would beinterpreted, andalsoexpressed doubts while journalists were reasonably worried about how this who distrust themostlyParisian mainstream media.Thus, positive reception themediaalliance had even amongthose of having local mediaparticipating intheproject, andthe seen asafactor of credibility. Thisconfirms theusefulness ofthe importance CrossCheck’s independence, whichwas via alocal healready trusted. Several insisted on One respondent mentionedthat helearntabout theproject CrossCheck was followed by some people. Two are worth asto why elements mentioning the form, stating: 'Happy exists'. thatCrossCheck as far-right even at theendof added commentary anextra online survey. Onerespondent whoidentifiedthemselves took thetimeto thoroughly answer the 50 onan questions CrossCheck, read several debunks intheirentirety, and right-tendencies, stated that they knew andtrusted anti-establishment, andeven openlymentionedfar We thinkitisnoteworthy that peoplewhoidentifiedas and expressed distrust toward journalism. established They also blamed both extremes political for disinformation, and websites – more orlessrelated to Front National. the vastnetwork of looselyconnected, far-right onlinegroups some respondents toned down thedangerof the'fachosphère', checking initiative from For media'. the'mainstream example, ofseem toaudience bethetype you would expect for afact- the mainstream media.However, other respondents didnot and mostlyconvinced that there isaneedfor verification by expected. Most of therespondents were politically moderate CrossCheck’s audience was more diverse politically than AUDIENCE DIVERSITY

IMPROVEMENT ANDAPPRAISAL

innovative, asillustrated by thefollowing quotes: CrossCheck diditsjobwell, considering that itwas new and or abrowser However, extension. they allrecognized that navigable anduser-friendly by creating categories, anapp, The audience alsosuggested makingthewebsite more letters, andincreasing theproject’s onsocialmedia. visibility to thequestionsasked onthewebsite, sendingmore news forbility CrossCheck’s content. Suggestions includedreplying Furthermore, many highlighted aneedto provide better visi- mation inamore detailed, visual, andresearched way. of verification, andpossiblydebunking some cases of disinfor They of alsomentionedthenecessity diversifying thetopics groups ofinfluence" tand ofother countries or who don’t alwaysunders- tothose citizens necessary enrichingand very doing excellentwork, "Keep going,youare of democracy, keepgoing!" "Your workisvitalforthegoodhealth you wantdiscussagain." beautiful project!Iamavailableincase and viralside...Goodlucktoyouonthis other tracksbasedalsoontheemotional be enough...itmightneededtodig rational andobjectiveargumentsmaynot the fake newsquickly!Unfortunately we needtofindawaydeflateallthis "I findthatyourprojectissoimportant, societies" that issavingour "Thank youforyourwork - -

55 CrossCheck - November 2017 did CrossCheck generate What kindof discussions in the andfrom media on Facebook, Twitter, the audience? 8

56 CROSSCHECK’S WEBSITE REQUESTS SUBMITTEDBYUSERSON beginning of theproject, issubstantial. thisrhythm nature, andthechallengingtaskof at gainingthe visibility Given theproject’s this during period). day 12 per innovative andMaybetween 7(which April10 represents anaverage of were questions 621 posted onthe platform, including321 Indeed, 2017, between February 28 andJune16, The number of questionsshows thesuccess of theproject. worried about the during campaign. election ofFrenchdisinformation aportion web were users mostly Therefore thesequestionsprovide asense ofthe of types wanted verifying, asignthat they were gaining traction. of theclaimsand fabricated content that theaudience The questionssubmitted by theaudience provided anoverview monitoring of socialmediacarried out by project editors. as disinformation. These questions complemented the rumours that were spreading onlineandcould beconsidered lizing theaudience by having themreportelection-related One of CrossCheck’s originalfeatures consisted of mobi- 8.1/ of somefindings of thefirst two studies. associated withtheproject andallows to verifythevalidity audience members. Thus ithelpslocate themainthemes by andabout CrossCheck, intheFrench mediaandamong perspectivevides ageneral onthe discourses generated lexicometricThe processing ofthese four samples pro- press that mentionedCrossCheck. mentioned CrossCheck, and, finally, from articles the French that comments on CrossCheck’s tweets page, Facebook public viaCrossCheck’s the requests website, by submitted We looked at four different groups thequestions and of texts: interactions related to theproject. of it,we carried out adiscourse analysis of theonlinetextual work duringCrossCheck, andof theaudience’s interpretation After conducting thequalitative analyses of thejournalists’

ANALYSIS OFQUESTIONSAND indicates theauthors’ relatively highlevel of education. the comment section of news websites. Ontheonehand, this cases correct, at leastmore sothanwhat isusuallyfound in Finally, thegrammarandsyntax of thequestionsare inmost rumors. Very few appearto bespam orirrelevant. a largeare majority authentic requests for theverification of the qualitative analysis of thequestions addition, In eponymous responsibility for theirdemand. provided theirfirst andlast namesontheplatform, thus taking Lastly, alarge number of individualswhosubmitted aquestion the endeavor". assignedto importance which highlightsthe drafting ofquestions, was investedinthe and conscientiousness care it suggestsacertain "On theotherhand, shows that

57 CrossCheck - November 2017 faux bulletin vote présidentiel tour élection dreuz info vrai théo habitation police taxe nancier parquet logement source info24 propriétaire programme nouvelle gagner intox déclarer français fr lettre https wordpress loi lien SIMILARITY ANALYSIS supprimer www emmanuel exact islamiste aller passer retraité video média impôt the co-occurrence links Clusters materializing com travailler pari taxer http macron nance vraiment between words brigitte Figure n°15 politique marcher videos compte information scandale nancier parler site enchainé facebook canard déclaration souhaiter di user soutenir able payer milliard article bonjour page euro loyer circuler suppression patrimoine mr france a aire mohamed voir saou llon candidat françois marine vendre ministre brunet pen asselineau ctif bénécier premier jean violer mélenchon dealer luc emploi mettre examen

58 PAGE POSTED ONCROSSCHECK’SFACEBOOK and 347,800 comments), engagements shares, (likes, project’sThe official page gathered 180,598 followers, toin particular thepromotional budget offered by Facebook. project’s Facebook page, whichwas quite successful, thanks CrossCheck’s reportswere systematically published onthe 2/ This represents 32% of thequestions. Dreuz.info that are prominent members of the'fachosphère'. simply signalsuspicious websites suchasEurocalifat and one that mentionssources andaskfor verification, orthat Marine Le Pen. Anothergroup important of questionsisthe Comparatively, of questionsconnected to Emmanuel Macron (35.5%). Theanalysisterms. confirms there was an overrepresentation Macron compared to other candidates’ names,words or , 15) (Figure audience the words presentthe inquestionssubmitted by The above word analysis whichgraphically similarity depicts the election’s round. first campaign andMay between 6 April12 — the key of period 2017, including4,286 during themostintense periodof the on 94 different postsbetween 6andMay 25,February We were ableto collect andprocess comments 4,532 posted

1,207,642 videoviews ANALYSIS OF clearly showsclearly the predominance ofEmmanuel only 8.2% of thequestionswere about Facebook comments Figure n°16 Word cloud of

OF THECOMMENTS (See Appendices for details).(See Appendices

topics are the following: to refine theprevious results. four The mostcommented place inthecomments onCrossCheck’s Facebook page and ments enablesusto better understand thedebates that took A closerlookat thedifferent topics andthe number of com- quite present. and France. Marine Le Pen andtheFront National are also of thewords Macron (which with resonates the former corpus) clearlyshows theoverrepresentation 16) the comments (Figure The word cloud whichgraphically depicts thewords present in included inthesample. Facebook comments mentionedinthissection mightdifferdue to different periods 4 TV chiefMacron comments, to support 124TV (353 published). comments,(363 70 published). happened inFrance published). comments 166 (399 posted, after 299published moderation). posted, rumour of hisallegedoffshore account (456 comments initially 9 _For aglobal view onsocialmedia metrics seeAppendices. The total number of 4 3 2 1 Theallegedorders given to journalists by theFrench public TheFront National allegedlyfinanced by drugmoney AviolentincidentinaRussian hospitalthat supposedly Macron pressing charges against thosethat peddlethe 49

59 and print). CrossCheck - November 2017 Europresse. 5 to warmly thankhimfor it. 5 For thisanalysis, we have used asample project partners. and alsobenefited from alarge press coverage, from especially where itwas thesubject of asignificant number tweets,of Beyond Facebook, CrossCheck was alsopresent onTwitter, 3/ newsfeed viasponsored posts. Facebook thepresence of CrossCheck 'imposed' intheir bias inthiscross-media alliance andalsothefact that commentatorsSome denounced apolitical 'mainstream' debateproject. about the the andthe nature objectivesof about the content ofthe debunks there was also some even though mostofthe discussion onFacebook was disinformation, significantly expanding its reach. Nevertheless, became auseful tool for thoselookingfor arguments against This confirms earlierfindingsabout the fact that the project on Facebook. propaganda indiscussionsabout theelection that took place in adirect way against disinformation andespeciallyfar-right scope was limited. Indeed, itenableditsaudience to argue of CrossCheck assumedatruly political function, even ifits Further analysis of thecomments shows that theproduction commented topics. page. He was notably thefocus of two out of thefour most the onCrosscheck’s maintopic ofcomments Facebook analysisThe shows that Emmanuel Macron represents small number of topics. thereforea note asignificant concentrationon of comments out of thetotal of 4,532, whichrepresents 34%. We can These four posts, from atotal of 94, attracted 1,571 comments June 12,2017, by 83different French the term CrossCheck, 7and published between January different accounts mentioning sample andaofarticles 261 50 posted between February 1andJune27, 2017 by 1,277 ARTICLES ABOUTCROSSCHECK 0 _Thiscorpus was kindlyprovided by Teyssou Denis from AFP, andwe would like 1 _Thecorpus was of retrieved 261articles from thedatabases Factiva and

ANALYSIS OFTWEETSANDPRESS 50 of 2,063 tweets 51 (online

of the project. and alsothe cross-media alliance that was at the heart their responsibility inthespreading of disinformation, the project, whichcan belinked to thediscussionabout ofimportance thefact that GoogleandFacebook backed What inthiscoverage was salient particularly was the cases of debunked, viraldisinformation. anddiscussionanalysisparticipants?) of numerous areits objectives? areitslaunching Who period(What the presentation andexplanation of theproject duringthe It was essentiallyfocused topics: ontwo important important. ownCrossCheck space ofits hasbeenrelatively outside coverage of theproject denote that the presence of Globally, theanalysis of theTwitter corpus andthemedia

Figure n°17 Word cloud of tweets

60 Conclusion * is to bereplicated further. aswell asaudiences,industry ifit of projectsthese types onthe news to understand theimplications of Japan, andthere isaresponsibility similar collaborative project in CrossCheck hasalready inspired a analysis sowe can learnwhat works. and allelementsrequire systematic This project was afirst of itskind disinformation. ofa typology visual, election-related during theproject asaway of building visuals andmemesthat surfaced piece of researchthe isanalysing audience understanding? Another content. theicons Did helporhinder tanding of rumours andfabricated audiences gain agreater unders whether they were effective inhelping ofused aspart thestories, testing experiments around thevisualicons Fazio shortly. Sheisrunning project will bepublished by Lisa phase ofresearch around the consumed itsoutput. next The project andthepeoplewho journalists that participated inthe CrossCheck project onthe to understand theimpact of the This research isafirst attempt -

61 CrossCheck - November 2017 APPENDICES

62 Walid Salem Samuel Laurent Gallois Natalia Margaux Gatty Louis Pillot Juliette Mauban-Nivol Grégoire Lemarchand Gaël Favennec TeyssouDenis Bernard-Bruls Delphine David Dieudonné Clémence Lemaistre Anaïs Condomines Amandine Ambregni Alexandre Capron Adrien Sénécat 1/ Name

LIST OFINTERVIEWEES (JOURNALISTS) Walid SalemFounder andManaging Editor /Rue89 Bordeaux Lead /Les Décodeurs Reporter Director andDeputy /Explicite LSE Student /BBC CFJ Student/Rue89 Strasbourg CFJ Student /BuzzFeed EditorDeputy inChiefMedia Social /AFP Sports Journalist /AFP MediaLab Lead /AFP CFJ Student/Libération Lead /GoogleNews Lab Web Editor inChief /Les Echos Journalist /LCI Web Journalist /AFP Verification Specialist / Les Observateurs / France 24 Verification Specialist / DécodeursLes Media Journalist Journalist Journalist Project Editor Project Editor Project Editor Journalist Journalist Journalist Project Editor Organizer Journalist Journalist Journalist Journalist Journalist Role duringCrossCheck

63 CrossCheck - November 2017

64 Cambridge University Press. A questionof lectures;2002). trustCambridge: (Reith ethics stillmatters. Wiley-Blackwell andO'Neill, O. (2002). 10- 9- 8- 7- 6- 5- 4- 3- 2- 1- 2/ New York, NY: Palgrave Macmillan. Houndmills, Basingstoke,of 39-56). media(pp. Hampshire; In Couldry, N.,Pinchevski, A.,& Ethics Madianou, M.(Eds.), tion-online chers-find-students-have-trouble-judging-credibilityinforma- Available at: https://ed.stanford.edu/news/stanford-resear Stanford News Center, November 22, 2016. have trouble judgingthecredibilityof information online, dia-literacy-backfire-7418c084d88d Pointsand Society: https://points.datasociety.net/did-me- http://plato. stanford. edu/entries/habermas. Stanford Encyclopedia of Philosophy. Mode of access: happened to journalism. Polis, March 2017 Communication Research, 41 430-454. (3), news from afree speechperspective. pectives, 31 FakeNews inthe2016 Election. Journal of Economic Pers Abingdon: Routledge. companion to news andjournalism374-383). (pp. news /entertainment divide. S. InAllan, TheRoutledge (Ed.) Couldry, Living well withandthrough N.(2013). media. Brooke, Boyd, J., Bohman, & R. Habermas, W. J. (2010). Beckett, ‘Fake C. (2017). news’: Thebestthingthat’s Baym, G.(2010). Whenfake Balmas, M.(2014). news becomes real. November 25). mediaandfake Social (2016, 19 Article Allcott, H., & Gentzkow, Media Social and M.(2017).

Couldry, N., & Ward, S. J. Why media A.(2013). REFERENCES D. 211-236. (2), D. (2016), D. Did Media (Jan. Did Literacy 2017) 5, Backfire,Data Real News/Fake News :Beyond the Stanford researchers findstudents

- - Public Should Expect. New York: Crown Journalism: What Newspeople Should Know andWhat the publique net/IpsosFrance/la-confiance-des-franais-lgard-de-la-parole- Availableparole publique, at (2015). :https://fr.slideshare. Journalism. London: Palgrave Macmillan. cape. InMiekle, G.andRedden, G.(Eds), OnLine News and Facts Are Sacred: journalistic ethics inachangingmedias ting-fake-news-agenda-for-research/ Harvard. Available at:https://shorensteincenter.org/comba- bating Fake News: AnAgenda for Research andAction. Research Institute. How Youth Navigate theNews Landscape. Data & Society Joseph, K.,Hobbs, W. C. andMattsson, (2017) forculty Media & Communication, Bournemouth University. Sport Committee 'Fake news' presented by inquiry theFa- mass communication. Political Communication, 30(4), 521-547. quickly we forget: Theduration of persuasion effects from Communication Inquiry, 246-262. 36(3), news, teens reject journalistic Journal 'objectivity'. of 11- 18- 15- 14- 12- 17- 16- 13- 20- 19- News 38-49). andJournalism (pp. London: Routledge. nal identities.S. InAllan, TheRoutledge (Ed.), Companion to face-big-challenges/415468/ been-an-explosionof-international-fact-checkers-but-they- Poynter. Available at: http://www.poynter.org/2016/theres- international fact-checkers, but they face bigchallenges. Kovach, B. andRosenstiel, T. TheElementsof (2001). confiance Ipsos, La des àl’égardFrançais de la Fenton, N.,Witschge,T. (2009).Comment IsFree, W. Donsbach, (2010). Madden, M., Lenhart A.,andFontaine, Madden, M.,Lenhart C. (2017). Lazer, D., Baum. M,Grinberg, N.,Friedland, L., Lilleker, D. Evidence to theCulture, (2017). Media and Hill, S., Lo, J., Vavreck, L., & Zaller, J. How (2013). WithFacebook, Marchi, R. (2012). blogs, andfake A.(2016) Mantzarlis, There’s beenanexplosion of Journalists andtheirprofessio- Com - -

65 using theterm ‘fake news’ ? friends are warping your memory. Nature, 168-170. (7644), 543 Infomediation of News. Media Fields Journal, no 10. sphere andnetwork. First Monday, Volume 17, Number 11. New America Foundation. fact-checking: Research findings from socialscience. Journalism Practice 179-196. (2) (2016). 41 Communication Research, source affiliation whenselecting news online. partisan exposure intheage of socialmedia: Endorsements trump luation Online. Journal of Communication,, 413-439.(3) 60 (2010). ning. American Educator. News: Teaching Students to Engage OnlineReaso inCivic - burg, CrossCheck - November 2017 33- 32- 31- 30- 29- 28- 27- 26- 25- 24- 23- 22- 21- journalism. Journalism Practice, 373-382. 4(3), Political Science, 50 Political Science, the evaluation of political beliefs.in American Journal of 2002). Cambridge: Cambridge University Press. Tambini, Whobenefits from (2017). Damian (Ed.). Taber, C. S., & Lodge, M.(2006). Spinney, How Facebook, (2017). L. fake news and GoogleandtheAlgorithmic Smyrnaios, N.(2015). Therefraction RiederB., Twitter (2012). chamber: as Phillips, A.(2010). O'Neill, O. (2002). Nyhan,B. & Reifler, MisinformationJ.A. (2012). and Nip, J. Y. Thelastdays of civicjournalism. M.(2008). Newman, N.,Fletcher, R., Levy, D., & Nielsen, R. Messing, S., & Westwood, S. J. Selective (2014). Metzger, M.J., Flanagin, A.J., R. Medders, B. McGrew, S., T. J. Ortega, Breakstone & S. Wine- TheChallengeThat’s (Fall 2017) BiggerThanFake TheReuters Institute News Digital Report 2017. Social andHeuristic. Social Approaches to Credibility Eva- 755-769(3), Transparency andthenew ethics of Aquestionoflectures; trust (Reith (8), 1042-1063. Motivated skepticism Fact Checking. Critical Review,162-180. 25 (2) Political Science. policy responses. TheLondon of School Economics and 36- 35- 34- Disorder. Report for theCouncil of Europe. Wardle, C. & H.(2017) Derakhshan, Uscinski, J., & Butler, TheEpistemology R. of (2013). Tambini, D., & Fake Goodman,E.(2017). news: Public Information

66 Graphic design -Graphic Credits : Datagif.fr

67 CrossCheck - November 2017 firstdraftnews.com/crosscheck-findings

68