Internet users engage more with phatic posts than with health misinformation on Facebook

Concerns about fake news have intensified since the 2016 American presidential election and Brexit vote, with social media platforms like Facebook facing criticism for spreading misleading content. While research on political fake news is prevalent, health and scientific misinformation have received less attention, despite posing significant societal risks. The study distinguishes between disinformation (false information intentionally causing harm) and misinformation (false or inaccurate information without malicious intent), focusing on the broader term 'misinformation' in this research. The study defines health misinformation as information contradicting the scientific community's consensus. A literature review reveals that health misinformation on social media is highly prevalent and often more popular than accurate information, potentially leading to severe public health consequences. However, empirically documented cases of harm from consuming health misinformation are limited. In France, the Facebook page *Santé + Mag* exemplifies concerns about health misinformation. Despite regularly posting potentially dangerous health misinformation (*e.g.*, “Use frozen lemons to fight diabetes”), it boasts a massive audience and significantly more interactions than leading French media outlets combined. However, this high interaction rate doesn't automatically imply popularity due to misinformation; a content analysis is needed to determine (1) the proportion of health misinformation versus other content and (2) the success rate of health misinformation specifically. The study analyzes 500 posts from *Santé + Mag*, examining 6,483,888 interactions (comments, shares, reactions) over 21 days. The goal is to identify post characteristics that correlate with higher or lower interaction rates, using an interdisciplinary approach drawing upon cultural evolution theory and media studies. Cultural attraction theory posits that cultural items' spread depends on psychological or ecological factors of attraction—properties making them more appealing and memorable. Research suggests that false information thrives not because of low quality but because of its high psychological attractiveness. This study investigates whether this is applicable to *Santé + Mag*. Media studies, focusing on the function of information consumption, highlight the role of social interaction in content sharing on social media. The study combines these perspectives to explore how both cognitive attractiveness and social function predict a post's success.

The study draws on two main bodies of literature: cultural attraction theory and media studies. Cultural attraction theory, a cornerstone of cultural evolution research, posits that the propagation and persistence of cultural elements (beliefs, narratives, rumors) are influenced by psychological and ecological factors of attraction. These factors enhance cognitive processing, making information more appealing and memorable. Previous research using this framework in the digital context has shown promising results, for instance, revealing that false information's success isn't solely due to its falsity but to its inherent psychological attractiveness. Cognitive factors like threat, disgust, social relevance, and negative emotions are suggested as key drivers of information spread. Contradictory findings exist regarding the virality of negative versus positive content; while some studies suggest a negativity bias in information transmission, others show positive or morally good content being more viral. The study acknowledges this discrepancy but leans towards hypotheses predicted by the majority of studies showing negativity bias. Media studies literature focuses on the functional aspects of information consumption, demonstrating that information isn't consumed solely for its content, but for its utility in satisfying individual needs, including social interaction. Studies show that Facebook use is primarily for maintaining offline relationships, interpersonal conversations with friends and family. It highlights the tendency to use bonding strategies to reinforce close relationships, not for meeting strangers. The study mentions the concept of 'phatic communication', coined by Malinowski, referring to statements devoid of factual information yet used for engaging in or maintaining social interactions (e.g., greetings). The study notes the increasing use of informal, relatable language in online media to foster audience engagement.

Data for this study were collected using the CrowdTangle API, provided by the Social Science Research Council (SSRC), gathering 500 posts from the *Santé + Mag* Facebook page and their associated 6,483,888 interactions over 21 days (January 10–31, 2019). The sample included a mix of articles, photos, and videos, mirroring the page's overall content distribution. Each post was coded manually by the first author, with inter-rater reliability checked by the second author and three additional coders. Cohen’s Kappa coefficients showed substantial to almost perfect agreement among coders (85.55–90.91%). The coding process involved multiple steps: First, posts were classified as health misinformation or not, further categorized as potentially harmful or not based on content accuracy and potential impact on readers. Second, the emotional valence (positive, negative, neutral) of each post was assessed, along with the presence of cognitive factors of attraction (negative emotions, threat, disgust, sexuality, social relations). Social relation mentions were further divided into subcategories (family, love, friends, society, antisocial interactions, pregnancy, mourning, infidelity). Finally, posts were classified as phatic (primarily social interaction) or non-phatic (primarily informational). To account for posts with multiple factors, up to nine categories could be assigned per post. Linear regression analyses were conducted to examine the relationship between post characteristics (coded categories) and the total number of interactions. The dependent variable (total interactions) was log-transformed to achieve normality. Initially, regressions focused on specific hypotheses (e.g., the relationship between health misinformation containing attraction factors and interaction levels). A final multiple regression model included all coded categories as predictors to determine the most significant factors influencing interactions.

The analysis revealed that only 28% of *Santé + Mag* posts were classified as health misinformation, and only 21% as potentially harmful. Despite the presence of cognitive factors of attraction (negative emotions, threat, disgust) in many health misinformation posts (79.3%), health misinformation was a negative predictor of interactions. Posts containing health misinformation accounted for less than 15% of total interactions, and that dropped even further (10.7%) when considering only potentially harmful misinformation. An analysis of comments on the top five most-commented harmful health misinformation posts indicated that only a small percentage of comments (3%) expressed approval. Many comments (48%) were jokes or seemed primarily socially motivated (sharing with friends, tagging). More than half (50.4%) of the posts dealt with social interactions, primarily family (34.9%) and love (31.8%). The multiple regression analysis indicated that the model explained 52.8% of the variance in interactions. Phatic posts were the strongest predictor of interactions, followed by positive emotional valence. Sexual content negatively predicted interactions, while threat, disgust, social information, negative posts, health misinformation, potentially harmful information, and posts lacking attraction factors showed no significant effect.

The findings challenge the assumption that *Santé + Mag*'s popularity stems primarily from the dissemination of health misinformation. Health misinformation, while present, was not the main driver of interactions. The significant negative prediction of interactions by health misinformation aligns with research indicating that fake news consumption and sharing are limited to a small minority. The study's results challenge the widely held belief in a negativity bias in information transmission, finding that positive content generated more interactions than negative or neutral content. The lack of a strong relationship between cognitive attraction factors (except for the negative effect of sexual content) and interactions suggests that factors beyond content may be at play. The overwhelming impact of phatic posts strongly suggests that the social function of posts—fostering social relationships—was the dominant driver of interactions on *Santé + Mag*. The results highlight the importance of conducting in-depth content analyses and the need for an interdisciplinary approach. The study points to potential biases in the literature on cultural evolution, possibly overlooking the role of positive emotions and social context in information transmission. While focusing on a French Facebook page, the study suggests the broader applicability of the findings, evidenced by similar trends observed on the English-language *Bright Side* page.

This study demonstrates that high interaction rates on social media do not automatically translate to significant engagement with misinformation. In the case of *Santé + Mag*, phatic posts and positive content were far more successful than health misinformation. The findings underscore the importance of nuanced analyses of online content and highlight the need for interdisciplinary approaches that consider both cognitive and social factors influencing information spread. Future research could focus on comparing phatic communication across various social media platforms and investigate how the 'public-private' nature of information sharing affects content success. Overall, the results emphasize Facebook's primary role as a social network, prioritizing social interaction over the spread of misinformation.

The study's focus on a single French-language Facebook page limits the generalizability of the findings to other social media platforms, languages, or cultural contexts. The manual coding, while ensuring accuracy, is labor-intensive and may be subject to coder bias, although inter-rater reliability checks mitigated this. The limited analysis of the comments on posts might not fully capture the depth of user engagement with health misinformation.