The propagation of fake news on Twitter

Synthesis produced by the Fondation Descartes of:

CSA (Conseil Supérieur de l'Audiovisuel). "La propagation des fausses informations sur les réseaux sociaux : étude de la plateforme Twitter". 25 novembre 2020.

This CSA report (of which this summary covers only a few aspects) analyzes the mechanisms by which false information spreads on Twitter. The report defines “false information” as “information that has been analyzed by journalists, particularly those specialized in fact-checking, and found to be false.” (p.13) The authors note that this report is limited to the social media platform Twitter and is therefore concerned with individuals who are not representative of the French population as a whole.

The first part of this report analyzes the activity of various types of Twitter accounts: reliable accounts, unreliable accounts, and fact-checking accounts. The report uses a classification system created by Décodex in order to categorize sources of information (category 1: parodical; category 2: spreading false information; category 3: questionable; category 4: reliable). Here are some interesting findings stemming from this report:

• Reliable accounts have significantly more followers that other types of accounts.

• An account’s audience on Twitter is not only represented by its number of followers, but also by its propensity to be shared via ReTweets by third-party accounts. However, the report shows that unreliable accounts possess a significant number of ReTweets, particularly when compared to their number of followers.¹ Followers of unreliable accounts therefore spread the information shared by these accounts more actively that followers of reliable accounts.

• The authors of the report sought to evaluate the extent to which followers of accounts spreading false information are likely to also be following reliable accounts. The report states that 80% of individuals following accounts spreading false information (category 2) follow at least one account deemed reliable (category 4), and that on average these individuals predominantly follow reliable accounts. These findings therefore do not reveal the existence of an echo chamber phenomenon on Twitter.

• The report shows that there is significant overlap in the subscriber base of numerous category 2 accounts, and that many of these accounts follow one another on Twitter.

• Journalistic accounts specialized in fact-checking (such as Checknews, Les Décodeurs, or AFP Factuel) possess more followers on average that accounts spreading false information (category 2) but generate less involvement from third-party accounts. The report further notes that fact-checking accounts are particularly interested in content that pertains to their work, such as news on media outlets or media literacy (19% of their content)

The second part of the report analyzes tweets spreading false information (identified by fact-checking websites) in order to better understand the accounts that are sharing this information, and those that are correcting it. The authors show that, contrary to what we may have expected (or hoped for), truthful information does not drive away false information. Indeed, Twitter conversations surrounding topics prone to the spread of false information rapidly die down without the truthful corrective information ever reaching the same amount of visibility.