The main findings of this study were that collective attention to COVID-19 on YouTube peaked around March 2020 and flattened out until at least mid-August 2020. Early videos focused on prevention, but politics and blame became a more salient theme over time. Politics and blame received the most public attention, amassing the largest number of collective views as well as having high median comment and like counts.
In this study, we analyzed 536 highly viewed COVID-19-related videos posted on YouTube from December 1, 2019 to August 16, 2020. The earliest videos were from Singapore and Hong Kong, and reported a then-mysterious virus and the Chinese authorities’ claim of no evidence of human-to-human transmission. The first top-viewed video posted by WHO was “Q&A about coronavirus” (see Table 4 in the supplement) uploaded on January 16, after the organization announced that China had identified a novel coronavirus on January 7, 2020 [@WHO20a]. Straight news outlets posted the majority of the top-viewed videos, while public-health authorities posted the smallest number of top-viewed videos. The themes of Prevention and Politics received the largest number of collective views, while Treatment received the least (figure 3).
Based on our findings, we propose that, first, collective public attention to the COVID-19 pandemic on social media peaked early and flattened afterwards (see figure 4), likely due to information fatigue. Second, there was a clear shift of COVID-19 themes from prevention early on (22.8% of total videos coded) to politics and blame later (20.5% of total videos coded, which is clearly different from the early 22.8% prevention, based on a two-sided test of equal proportions using in R, \(pcript>\)). 27/225 (12.0%) videos published before April 8, 2020 had themes of politics and blame, compared to 83/331 (26.7%) videos published on or after April 8, 2020 (see figure 5). Videos oriented to politics and blame also received more reactions from the viewers as measured by comments and likes/dislikes (see Table 3 in supplementary material). That politics and blame became the most salient theme among the top-viewed videos after April could be due to the politicization of COVID-19, influenced especially by the upcoming presidential election in the U.S. during that time (see figure 5). Third, regarding prevention and control measures, masking received much less attention than lockdown and social distancing in our study period (figure 6).
Videos published before April 2020 received the most attention, and the majority of the videos during that time focused on basic information about the virus, the disease, and prevention guidelines (see figures 4 and 5). Although we are not able to confirm when views occurred, this pattern, together with similar patterns in Google search trends (see figure 4 and other studies [@EffeKron20; @KrisLore21; @MangSmit21]), likely point to attention fatigue. Despite spikes of COVID-19 cases, it appears that collective attention to the outbreak (at least on YouTube and Google searches) decreased after March 2020. This decline could indicate that the public was overwhelmed by COVID-19-related information, that they felt that they understood enough about the disease, or that they simply became less interested. Attention fatigue could have impacted health-related behaviors, increasing behaviors that went against regulations and worsened epidemic severity. It is also plausible that the initial panic that overwhelmed many as the virus emerged subsided as more information became available; and what followed was a shift to interest in political themes. Highly politicized messages tend to receive more engagement on YouTube [@MarcAu20], and distrust and competition among information sources can reduce dissemination of presumably helpful health information [@GozzTizz20; @Yilm20], as an “infodemic” of conflicting information means that valuable information cannot be communicated clearly [@WHO20].
The public’s perception of crises can be shaped by media content [@LiMa21; @MollHarm15; @TeslPham20; @GozzTizz20], as can behavioral responses [@ChanWinn18; @LiMa21]. Fatigue might contribute to behavior changes (e.g., becoming less alert and paying less attention to distancing measures) and influence epidemic dynamics, especially when messages about controlling spread become less clear and more political. When public attention switches to more controversial topics (e.g., political blaming games) and away from public health information (e.g., effectiveness of universal masking and non-physical COVID-19 impacts) during a public health crisis, what is missed or misread can be consequential.
For example, masking was not a major theme in top-viewed videos (figure 6) in the study time period. While search interest in how to make coronavirus masks spiked in May 2020 [@MangSmit21], universal masking in the United States was not proposed until the CDC called on Americans to wear masks to prevent the SARS-CoV-2 spread on July 14, 2020 [@BrooButl20; @CDC21]. From “Can masks protect against the new coronavirus infection?” (in Table 4 in the supplements) on February 5, 2020, “Trump Ignored Coronavirus Warnings; Pence Refused to Wear a Mask: A Closer Look” (in Table 4 in the supplements) on April 26 and “Wear a mask. Help slow the spread of Covid-19” (in Table 4 in the supplements) on July 26, the psychological and public-health effects of mask wearing remained culturally and politically controversial as observed elsewhere [@CzypGree21], despite the spike of infected cases in the United States and worldwide (see panel c in Figure 4). The politicization of masking, in addition to a public attention fatigue, served to undermine the control of SARS-CoV-2 transmission. Furthermore, given that the use of social media as a source of COVID-19 information is associated with fewer COVID-19 health-protective behaviors [@AlliDuff20], it is possible that the misuse of social media sites such as YouTube could have increased COVID-19 cases and deaths.
Politicization could extend also to vaccination and vaccine hesitancy due to ethno-cultural, religious, or political beliefs [@PuriCoom20], as observed internationally [@IslaKama21], and in the UK and USA in particular [@Loomde21]. Although there were few top-viewed videos about SARS-CoV-2 vaccines compared to other prevention methods, the number of vaccine-related videos increased while others decreased in August (Figure 6). The early videos about vaccination tended to focus on the timeline of availability with the assumption that everyone wanted to be vaccinated, but the later vaccination discussion became more skeptical (e.g., “The risky way to speed up a coronavirus vaccine” in Table 4 in the supplement), and more political (e.g., “Half of Britons would not get a coronavirus vaccination” in Table 4 in the supplement). The social media discussion regarding vaccines (and vaccine hesitancy) may have increased after the study period, due to the development and approval of numerous vaccines in late 2020. Social media could have enhanced the politicization of vaccination and other public health measures, therefore reducing their effectiveness as vaccine hesitancy was likely associated with political ideology [@KillCloo21].
A shift from science to politics at an early stage could represent an important missed opportunity to disseminate useful information about prevention, as suggested in a study of YouTube videos very early in the epidemic [@BascHill20]. This shift to politics contributed to the infodemic, and may have weakened mitigation measures during the first waves of the COVID-19 pandemic as government decisions are expected to be based on science [@FiscBroo20; @Slov12]. While political polarization may lead to an increase in the uptake of inaccurate information from individuals’ self-selected polarized news sources or echo chambers [@BaveBaic20], the subsequent reduction in media and public attention to COVID-19 may also have made it more difficult for public health authorities to disseminate key information. We hope that our findings can contribute to raising awareness of the importance of science communication in combatting polarization and the spread of misinformation.
An important study limitation is that we were not able to access the view dates of videos, only publication dates. Thus, our knowledge of when viewing patterns changed is limited, and our comparisons between older and newer videos are potentially biased, since the former had more time to accumulate views. However, we note that overall patterns in our YouTube viewership are consistent with those found in Google Trends – in particular, the pattern of a mid-March spike in attention with a decline over the subsequent months. Another limitation is that, since we could not determine the viewers’ ethnic, cultural and political backgrounds, controlling for or comparing based on these factors was not possible. Consequently, our inferences concerning the patterns of themes and collective attention should be generalized cautiously.