Cross-platform spread: vaccine-related content, sources, and conspiracy theories in YouTube videos shared in early Twitter COVID-19 conversations.

High uptake of vaccinations is essential in fighting infectious diseases, including severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) that causes the ongoing coronavirus disease 2019 (COVID-19) pandemic. Social media play a crucial role in propagating misinformation about vaccination, including through conspiracy theories and can negatively impact trust in vaccination. Users typically engage with multiple social media platforms; however, little is known about the role and content of cross-platform use in spreading vaccination-related information. This study examined the content and dynamics of YouTube videos shared in vaccine-related tweets posted to COVID-19 conversations before the COVID-19 vaccine rollout. We screened approximately 144 million tweets posted to COVID-19 conversations and identified 930,539 unique tweets in English that discussed vaccinations posted between 1 February and 23 June 2020. We then identified links to 2,097 unique YouTube videos that were tweeted. Analysis of the video transcripts using Latent Dirichlet Allocation topic modeling and independent coders indicate the dominance of conspiracy theories. Following the World Health Organization's declaration of the COVID-19 outbreak as a public health emergency of international concern, anti-vaccination frames rapidly transitioned from claiming that vaccines cause autism to pandemic conspiracy theories, often featuring Bill Gates. Content analysis of the 20 most tweeted videos revealed that the majority (n = 15) opposed vaccination and included conspiracy theories. Their spread on Twitter was consistent with spamming and coordinated efforts. These findings show the role of cross-platform sharing of YouTube videos over Twitter as a strategy to propagate primarily anti-vaccination messages. Future policies and interventions should consider how to counteract misinformation spread via such cross-platform activities.

Introduction

The spread of vaccine misinformation over social media has been linked to increased global vaccine hesitancy in the past decade.[1-6] In a joint statement with other leading international health organizations, the World Health Organization (WHO) noted that, “the Coronavirus disease (COVID-19) is the first pandemic in history in which technology and social media are being used on a massive scale.”[7] The statement cautioned against the “infodemic,” which refers to an overabundance of information and misinformation that can cost lives.[7] To design effective communication efforts and policies, it is important to understand vaccination-related social media content and its dynamics. However, multiple factors, not the least of which are the quantity and constant changes in content complicate such understanding. Novel computational tools applied by interdisciplinary teams show promise in advancing knowledge of vaccination-related content on social media. In turn, this knowledge sheds light on public sentiments and sense-making.[8-11]

Twitter has documented importance in mining public discourse about vaccination.[10,12-28] Studies revealed that only a minority of tweets about vaccination opposed vaccination,[12,17,29] and these are generated by a relatively small number of sources and disseminated in close-knit social networks that create “echo chambers.”[5,30] Therefore, vaccine-opposing tweets are often not the reflection of authentic, individual users’ content but are instead spread through bot-like activity.[2] Spreaders of this misinformation include certain governments, politicians, celebrities, and other sources.[16,26,29] Other social media that were documented as advancing vaccine-opposing messages include Facebook,[31-35] Instagram,[36-38] Pinterest,[39-41] and YouTube.[42-46] Use of specific platforms is associated with different levels of trust in vaccination and intentions to get vaccinated,[47] and the vaccine-related content on these platforms differs. For example, in contrast to Twitter’s propagation of largely positive or neutral vaccination sentiments and medically substantiated information,[12,17,29] YouTube’s platform promotes mostly anti-vaccination videos with narratives that center on conspiracy theories and claims of civil liberty infringement.[42-45]

A limitation of previous research that examined vaccine-related social media content was the focus on analyzing discrete social media platforms. Social media users typically utilize multiple media sources[48,49] as part of media affordances.[50] Therefore, it is important to explore cross-platform use, which consists of utilizing one social media platform to disseminate information about the content available on other platforms. Social cybersecurity research documented the role of cross-platform use of social media – especially YouTube – in disinformation campaigns.[20,51] However, the role of cross-platform use in dissemination of vaccination-related content is largely unknown.

There is also a gap in the literature about content and spread of vaccination-related conspiracy theories over social media during the COVID-19 pandemic. Times of uncertainty and hardship, such as during pandemics, often lead to an increase in conspiracy theories as individuals attempt to make sense of rapidly unfolding events and fears.[52-54] Conspiracy theories are narratives that assign a strong potency of causation to evil forces and that have epistemologies that diverge from scientific methods of knowledge.[55] Conspiracy theories provide a powerful negative framing of vaccination,[32] and believing conspiracy theories is inversely correlated with the intention to get vaccinated for COVID-19.[56] Individuals who are exposed to anti-vaccine conspiracy theories are less likely to intent to vaccinate their children, and the effects of such exposure are long lasting.[57,58] Furthermore, beliefs in conspiracy theories are difficult to refute,[57] as they provide individuals with a sense of coherence and agency in the face of uncertainty.[59] Numerous studies documented the dissemination of COVID-19 related conspiracy tweets,[9,23,28,60-63] including a recent study that showed that their spread was consistent with bot-like propagation.[63] However, to our knowledge, no previous studies focused on vaccine-related conspiracy theories spread across platforms. Given the importance of YouTube in disseminating vaccination-opposing narratives, including conspiracy theories, it is likely that tweets shared during COVID-19-related conversations about vaccination would include links to related videos. Exploring the spread of vaccination-related information, including the content and propagation of conspiracy theories across platforms, is necessary for understanding this new media environment and designing evidence-based interventions and policies.

In this study, we analyzed tweets that were posted from 1 February 2020 (2 days after the WHO declared the outbreak of COVID-19 to be a public health emergency of international concern) through 23 June 2020. This time frame encompasses the early months of the pandemic prior to the rollout of COVID-19 vaccinations. These early conversations reveal vaccine-opposing strategies that were not influenced by the specific characteristics of the COVID-19 vaccines, including claims about their efficacy and safety. Instead, they have the potential to illuminate the discourse of vaccination sense-making early in the pandemic, including the success of public health sources in disseminating their messages on one side, and established vaccine-opposing forces leveraging the pandemic to advance their agenda on the other.

According to agenda-setting theory,[64,65] the salience of media content and its framing are important in shaping public perceptions.[66] In the last decade, social media has disrupted past models in which traditional media set the agenda. Both social and traditional media are now available online and interact in complex ways to shape agendas. Yet, it is largely unknown how social media is used to disseminate other media sources in the context of vaccination. Understanding the frequency in which links to YouTube videos are tweeted as part of vaccination-related tweets in COVID-19 related conversations, the temporal dynamics of these tweeted videos, their content, and the frames employed (including conspiracy theories) can reveal agenda-setting processes in this new media environment.[67,68] In addition, examining the propagation strategies of frequently tweeted YouTube videos can help identify potential bots and inauthentic propagation. Such inauthentic spread strategies constitute new processes that might impact agenda setting in digital environments. To understand these issues, we posed the following research questions.

To explore the magnitude of this phenomenon, the first question focused on the overall prevalence of cross-platform sharing of YouTube videos in vaccination-related tweets posted to COVID-19 conversations during the early months of the pandemic.

RQ1: What were the absolute and relative frequencies of YouTube videos tweeted as part of COVID-19 conversations during the early months of the pandemic?

YouTube videos that were highly circulated in our data set demonstrate agenda-setting processes. The content of the most tweeted videos therefore demonstrates both the content that certain users chose to share, the sources that were effective in dissemination over Twitter, and the content that users were exposed to. We chose the 20 most tweeted videos to allow an in-depth analysis, while also providing an insight about diverse videos.

We posed research question two:

RQ 2: What were the content and sources of the 20 most tweeted YouTube videos, including sentiments regarding vaccination and conspiracy theories?

As social media content often represents the efforts of inauthentic users, we also sought to identify the specific strategies of spreading the most tweeted YouTube videos. The third research question focused on these strategies.

RQ 3: What characterized the spread of the 20 most tweeted YouTube videos?

Finally, given the dynamic nature of social media, we sought to gain broader knowledge of the content and temporal distribution of videos tweeted in the data set. Research questions four and five were presented:

RQ 4: What were the topics and frames of YouTube videos that were tweeted as part of COVID-19 conversations during the early months of the pandemic?

RQ 5: What was the temporal distribution of YouTube videos that were tweeted as part of COVID-19 conversations during the early months of the pandemic?

Methods

Data

The analysis reported here is part of a larger study that examined vaccination-related tweets.[69] The analyzed data were originally collected in two different studies that examined COVID-19 Twitter conversations in the early months of the pandemic. The first data set was based on a collection of tweet IDs gathered using general COVID-19 related keywords, such as “coronavirus” and “Wuhancoronavirus.”[70] We used “hydration,” a process of gathering all the pertinent information about each tweet into the JavaScript Object Notation (JSON) format file[71] via the Twitter Search API on all of the tweet IDs. This process populated data from 140 million tweets that were still available on Twitter in August 2020. The second data set included approximately 4.5 million COVID-19-related tweets collected from 29 January 2020 to 23 June 2020[72] using Twitter’s Streaming API. To include tweets that referred to vaccination, we filtered the data sets using the sub-strings “vax” and “vaccin,” which resulted in 1.6 million English tweets that discussed vaccinations. Both data sets were then filtered to include only the dates of overlap (1 February to 23 June 2020) and to remove any duplicate tweets across the data sets. The final data set included 930,539 unique tweets. To identify tweets that contained URLs linking to YouTube videos, all URLs from the Tweet’s JSON “entities” object of each tweet were extracted. Figure 1 shows a graphical flowchart for the data selection and exclusion process.

Figure 1.

Data set combination, filtering, and exclusion process.

YouTube video topic modeling

All transcripts that were available for the YouTube videos were downloaded directly from YouTube.com; 61% (n = 1,280) of the YouTube videos in the data set had a transcript. Examples of videos lacking transcripts include those taken down by YouTube.com, videos without text (e.g., the opening scene of the 1995 film Outbreak), and videos in languages other than English that had the transcript feature disabled. To preprocess the transcripts, we removed common English stop words, and some commonly used phrases were mapped to their representative concept. Each of the terms was then lemmatized, a process that involves mapping each of the words to their base words, which facilitates their analysis as a single concept. Following this pre-processing, we used Latent Dirichlet Allocation (LDA),[73] a natural language processing, unsupervised machine learning technique to identify topics in the YouTube transcripts. LDA Topic modeling is based on commonly co-occurring words that are likely to belong to a topic. We created a number of different LDA models with varying hyperparameters and used a coherence metric[74] to select the final model and number of topics. Each transcript received a probability score that reflected the likelihood that the transcript is related to each topic. We then created word clouds for each topic to visually illustrate keyword frequencies in the transcripts.[75] We also analyzed the temporal breakdown of each topic. In addition, we examined the word counts of “Gates,” “autism,” and “5 G” as proxies of known vaccination/COVID-19 conspiracy theories[2] in each topic.

Content analysis

In addition to computational methods, content analysis methods[76] were applied in analyzing the 20 most tweeted YouTube videos and the topics generated by the LDA analysis of YouTube transcripts. The coders were pre-medical and medical students who were trained by the first author. The second author (IC) independently coded the content as a verification. First, the 20 most tweeted YouTube videos were coded to identify their source, number of comments, overall sentiment, presence of conspiracy theories, and source. We considered conspiracy theories as explanations to events that involve secretive and malevolent plotting by powerful and evil groups when more plausible explanations are available.[77] Intercoder reliability was 100%, and no conflicts needed to be resolved. In analyzing the YouTube topics, two independent coders watched and read 10 transcripts from each topic and independently assigned a descriptive label to each topic. These labels were compared and discussed with the first author, and a shared understanding was achieved.

Twitter spread analysis

In order to understand the particular Twitter spread strategies used to propagate the most tweeted YouTube videos, we applied social cybersecurity methods to identify coordinated link sharing and coordinated inauthentic behavior between different social media accounts.[78,79] To analyze all the tweets that shared one of the top 20 most tweeted YouTube videos, we first excluded retweets and removed from the remaining original tweet mentions, URLs, trailing whitespace, and formatting characters (i.e. ‘\n’). Then, we recorded the following: the numbers of tweets that included a URL to each of the videos, the unique users that tweeted the URL, the unique texts across all of the tweets that contained the URL, the tweets that contained a URL that were tweeted within an hour of the first tweet of the video, and non-reply mentions. Finally, we documented the range of days between the first and last tweets that included these URLs.

Results

The results demonstrated the magnitude, temporal distribution, and content of YouTube videos shared within vaccine-related tweets as part of early COVID-19 Twitter conversations. Specifically, this study explored the content and spread of the 20 most tweeted YouTube videos, and the content and frames of all tweeted YouTube videos, including the presence of conspiracy theories.

The first research question examined the absolute and relative frequencies of YouTube videos tweeted as part of COVID-19 Twitter conversations during the early months of the pandemic. A total of 2,097 unique YouTube video URLs were identified. Two percent (n = 3,722) of all tweets that shared a URL had a URL of a YouTube video. YouTube.com was the most persistently tweeted domain. On most days (134 of 143 days, or 93.7%) in our sample, at least one tweet with at least one YouTube URL was tweeted. YouTube was the third most tweeted domain across all of the domains with 4,106 tweets featuring a YouTube URL. Almost 90% of the tweeted videos were only tweeted within a 1-week period, and 75% of videos were only tweeted once.

The second research question focused on the content and sources of the 20 most tweeted YouTube videos, including their sentiments regarding vaccination and conspiracy theories. These results are presented in Table 1, with videos ranked-order according to the number of tweets that included their URL.

Table 1.

Content and sources of the 20 most tweeted YouTube videos

Rankby number of tweets	Source	Type of source	Title	Length in minutes	Number of views	Date posted	Number of comments	Opposed	Conspiracy theory
1	RT America	Russian government international TV network	The CDC is actually a vaccine company – Robert F. Kennedy Jr	12:46	218,620	1/312020	1,740	Y	Y
2	Peak Prosperity	US blog, financial information	Coronavirus, the potential for a vaccine, and the cytokine storm!	22:57	279,443	5/112020	3,209	Y	Y
3	Brodel Nation	US blog pro-Trump	Dr Judy Mikovits, PhD interview	3:18	1,450	4/262020	11	Y	Y
4	All Express News	NTD, anti-Chinese government US-based organization	Coronavirus Covid19 China First. The US and the World Next	10:46	1,727	2/102020	2	N	N
5	Corbett report	Vlog, Japan-based Canadian	Bill Gates’ Plan to Vaccinate the World	28:01	514,319	5/82020	9,684	Y	Y
6	Guardian News	UK newspaper	Donald Trump says coronavirus will ‘go away without a vaccine’	0:58	318,573	5/92020	1,801	Y	N
7	Corbett report	Vlog, Japan-based Canadian	Bill Gates and the Population Control Grid	36:47	544,538	5/172020	8,050	Y	Y
8	Corbett report	Vlog, Japan-based Canadian	Corona World Order	59:16	227,906	4/102020	4,169	Y	Y
9	Psych Hunter	Independent/Unknown	Psychiatry: Science or Fraud	4:52	33,144	9/8/2009	0	N	Y
10	CorbettReport	Vlog, Japan-based Canadian	Meet Bill Gates	38:45	782,374	5/242020	9,640	Y	Y
11	CNBC	US TV news network	UK scientist makes coronavirus vaccine breakthrough, reports Sky News	6:52	413,246	2/52020	1,080	Y	N
12	Eschatology Explorer	Independent/Unknown	Agenda 21: Global Depopulation, Global Enslavement, One World Totalitarian Government (Hindi)	16:12	256	2/212020	5	Y	Y
13	Informed Consent Action Network	Anti-vaccination advocacy organization	Did CDC approve a vaccine that causes heart attacks?	2:33	9,108	3/192018	54	Y	Y
14	Autism Media Channel	Anti-vaccination advocacy organization	CDC Whistleblower Revealed	9:30	207,647	8/262014	693	Y	Y
15	Corbett report	Vlog, Japan-based Canadian	How Bill Gates Monopolized Global Health	23:57	1,136,774	6/102020	7,584	Y	Y
16	CNBC	US TV	Trump visits NIH vaccine research center amid coronavirus spread in the US – 3/3/2020	23:49	68,449	3/32020	323	N	N
17	Peggy Hall-The Health American	Vlog of Peggy Hall, an anti-masks and anti vaccination advocate	[Removed]	[Removed]	[Removed]		[Removed]	Y	N
18	The White House	White House	03/03/20: President Trump Delivers Remarks Upon Arrival	6:37	35,925	3/32020	[Disabled]	N	N
19	Construuct	Conspiracy theories YouTube channel	PENTAGON BRIEFING ON REMOVING THE GOD GENE – “FUNVAX”	4:15	236,792	4/2/2012	8	Y	Y
20	Ben Swann	Anti-vaccine advocate	FOLLOW THE MONEY Because Here Comes the MANDATORY Coronavirus Vaccine	7:57	50,656	4/192020	654	Y	Y

As Table 1 shows, one video (ranked 17th) was removed by the time of the analysis. The video transcript was available and analyzed by the team. It indicated that it was produced by Peggy Hall, who has a vlog named “The Healthy American.” The video focused on her objection to mask mandates and mentioned vaccines as unhealthy. Additional information on this video is unavailable. Of the remaining 19 videos, 15 were classified as opposing vaccination, and 15 presented conspiracy theories. Most videos that were classified as opposing vaccination were also classified as featuring a conspiracy theory. The exception included one video that did not focus on vaccines but rather on debunking psychiatry as a conspiracy theory (“Psychiatry: Science or Fraud” ranked 9th). Two other videos were classified as opposing vaccination, as they included messages that doubted the need for vaccination with no conspiracy theory. These consisted of a video (ranked 6th) that featured President Trump saying: “I feel about vaccines like I feel about tests, this is all going to go away without the need for a vaccine,” and a CNBC video (ranked 11th) reporting that the vaccine is unlikely to be produced or helpful in combating the pandemic. Moreover, 4 of these videos (ranked 9th, 13th, 14th and 19th) were originally produced and circulated years before the pandemic. In addition, the most tweeted video and the third most tweeted video did not relate to the pandemic, for a total of 7 videos that contained vaccine-opposed content or other conspiracy theories unrelated to the current pandemic.

The sources of the videos were diverse. Traditional media presented only three of the most tweeted YouTube videos. These included two CNBC videos (ranked 11th and 16th) and one by the Guardian (ranked 6th). The most tweeted video was produced by RT (Russian TV) America, a media outlet funded and produced in English by the Russian government. This video presented Robert F. Kennedy, Jr., the chairman of an anti-vaccine advocacy group. In it, he alleged corruption in the Centers for Disease Control and Prevention (CDC), including that the CDC has allegedly engaged in silencing whistleblowers who proved that vaccines caused autism and other epidemics, along with additional conspiratorial claims. The vlog “The Corbett Report” created 5 of the 20 most tweeted videos (ranked 5th,7 th,8 th, and 10th). In his vlogs, Mr. Corbett presented himself as a Canadian English teacher living in Japan. Other sources included vlogs based in different countries. All 15 videos that presented conspiracy theories were posted by non-official sources, such as vlogs and YouTube channels, with RT as the only source overtly created by a government.

The analysis further revealed that the conspiracies depicted in these videos were explicitly stated. They included narratives about Bill Gates planning the pandemic and benefiting from vaccinations, and the CDC as an organization aiming to hurt the public by knowingly administering harmful vaccinations. RT, the Corbett Report, and the Informed Consent Action Network presented a news-like set up that resembled mainstream newsrooms or scientific broadcasts. Other videos documented real or manufactured communication between officials. These included a video titled “Funvax,” an old hoax claiming to show a lecture to the Pentagon about a vaccine that would treat religious extremists by targeting their brain.[80] Another was an edited video of the CDC deliberation and subsequent approval of a hepatitis B vaccine that increased the risk for acute myocardial infarction. One video’s title claimed it was in Hindi, but in fact had content in Urdu (as verified by two Hindi/Urdu speakers) following 5 minutes of English content.

The third research question centered on the spread of the 20 most tweeted YouTube videos in the data set, as evident in the tweets sharing these URLs. The results are presented in Table 2.

Table 2.

Tweets spreading the 20 most tweeted YouTube videos

Rank	Original YouTube video title	Number of Unique tweets with video	Number of Unique texts of tweets	Number of Unique users that tweet video	Average number of non-reply mentions in tweets with video	Number of Retweets of tweets with video	Number of days between first and last tweet of video	Number of Tweets within first hour of first tweet
1	‘The CDC is actually a vaccine company’ Robert F. Kennedy Jr	89	32	84	1.1 ± 1.6	40	134	6
2	Coronavirus, the potential for a vaccine, and the cytokine storm!	64	27	59	0.88 ± 0.45	6	2	25
3	Dr Judy Mikovits, PhD interview	30	11	18	1.57 ± 0.76	40	12	1
4	Coronavirus Covid19 China First. The US and the World Next	54	18	1	0.0 ± 0.0	0	10	6
5	Bill Gates’ Plan to Vaccinate the World	59	12	8	1.58 ± 1.14	3	35	1
6	Donald Trump says coronavirus will ‘go away without a vaccine’	3	3	3	2.0 ± 2.16	63	0	2
7	Bill Gates and the Population Control Grid	52	9	7	1.5 ± 0.93	0	27	1
8	Corona World Order	42	2	1	1.62 ± 0.69	0	20	1
9	Psychiatry: Science or Fraud	23	5	1	2.13 ± 1.51	0	91	1
10	Meet Bill Gates	34	3	3	1.624 ± 0.77	0	20	1
11	UK scientist makes coronavirus vaccine breakthrough, reports Sky News	18	8	18	0.72 ± 0.45	4	28	2
12	Agenda 21: Global Depopulation, Global Enslavement, One World Totalitarian Government (Hindi)	22	1	1	4.27 ± 9.03	0	0	22
13	Did CDC approve a vaccine that causes heart attacks?	16	2	2	0.25 ± 0.97	4	4	1
14	CDC Whistleblower Revealed	20	15	2	1.15 ± 2.78	2	104	4
15	How Bill Gates Monopolized Global Health	28	11	7	1.68 ± 1.28	0	19	7
16	Trump visits NIH vaccine research center amid coronavirus spread in the US – 3/3/2020	18	10	15	1.39 ± 1.29	1	0	8
17	[Removed]	29	1	1	2.76 ± 8.93	0	0	29
18	03/03/20: President Trump Delivers Remarks Upon Arrival	8	8	7	0.87 ± 1.27	12	1	1
19	PENTAGON BRIEFING ON REMOVING THE GOD GENE. -’FUNVAX’	24	1	1	1.29 ± 1.21	0	0	24
20	FOLLOW THE MONEY Because here comes the MANDATORY Coronavirus Vaccine	1	1	1	0.0 ± 0.0	17	0	1

As the table reveals, three videos (#17: [Removed], #19 “Pentagon briefing on removing the God Gene -Funvax,” and #12 “Agenda 21: Global Depopulation, Global Enslavement, One World Totalitarian Government (Hindi)”) were propagated over tweets that originated from one account and were typically tweeted within less than 1 hour. For each video, the texts of these tweets were identical. The only variations included the number of mentions and users referred to in those mentions. For example, two of the tweets that shared the second-most tweeted video (“Coronavirus, the potential for a vaccine, and the cytokine storm”) are detailed below:

@davidicke Remember this Video?

“Unexpected” results were: Disabling other Free-will Brain-functions to create Slaves (Closed Minded, Obeying, Limiting Imagination, etc).

Why some can’t Awaken to Reality.

Why “They” desire to push COVID vaccination ASAP …

(Not Bill)

@Ujalacygnus @shuchinbajaj Remember this Video?

“Unexpected” results were: Disabling other Free-will Brain-functions to create Slaves (Closed Minded, Obeying, Limiting Imagination, etc).

Why some can’t Awaken to Reality.

Why “They” desire to push COVID vaccination ASAP …

(Not Bill)

This spread strategy indicates that individual accounts were spreading the videos by tweeting the same message over multiple tweets and using mentions to draw other users’ attention to their tweets. This is consistent with spamming strategies (also referred to as flooding), which is a common tactic employed by Twitter bots and in spreading disinformation.[81,82] Moreover, the second and tenth most tweeted videos had more unique users tweeting the videos than unique texts of tweets that had videos, indicating likely coordination between different accounts.

The fourth research question examined the topics and frames of YouTube videos that were tweeted in the data set, including their frequency over time. The topic modeling analysis of the transcripts provided a final LDA model with 12 topics and had a coherence of 0.40. The Appendix displays the twenty most common words for each of the topics.

Table 3 presents the summary of the content analysis and the number of YouTube videos included in each topic in descending order based on the number of videos in each topic. As the table shows, only topics 3 and 4 were consistently positive or neutral in depicting vaccination. The other topics consisted of videos that were either opposed to vaccinations (topics 1, 5, 8, and 12) or included a mixture of videos with different sentiments toward vaccination. Some topics included both pro- and anti-vaccination videos (topics 2, 6, 7, and 9). The three topics with the lowest number of videos were largely opposed to vaccination but also included some videos that supported vaccinations (topics 10 and 12) or that were neutral (topic 11).

Table 3.

Topics of YouTube transcripts by order of magnitude

Topic	Topic descriptions	Number of videos	Vaccine sentiment	5 G	Autism	Gates
1	Vaccines infringe on individual freedom, big government	310	Opposed	36, 12.6%	16, 5.1%	86, 27.7%
2	Coronavirus vaccine medical trials	235	Mixed	4, 1.7%	1, 0.4%	7, 2.9%
3	Coronavirus news by world leaders	232	Neutral	7, 3.0%	0, 0%	12, 5.1%
4	Coronavirus outbreak news, epidemiological information	187	Neutral/pro	4, 2.1%	2, 1.0%	15, 8.0%
5	Vaccines cause autism conspiracies	93	Opposed	9, 9.7%	52, 55.9%	14, 15.0%
6	General health and conspiracies	61	Mixed	11, 18.0%	5, 8.2%	12, 19.6%
7	Information on viruses including conspiracies	59	Mixed	4, 6.7%	10, 16.9%	5, 8.4%
8	Conspiracy theories, Bill Gates, 5 G	47	Opposed	8, 17.0%	3, 6.4%	41, 87.2%
9	News on economic markets	19	Mixed	3, 15.7%	0, 0%	3, 15.8%
10	Flu vaccine and mask wearing conspiracies	15	Opposed/mixed	1, 6.7%	1, 6.6%	0, 0%
11	Christian content & conspiracies	14	Opposed/neutral	3, 21%	0, 0%	3, 21.4%
12	Coronavirus research, including conspiracies	8	Opposed/mixed	0, 0%	0, 0%	2, 25.0%
Total		1,280		90	90	200

The analysis also documented the frequency across the topics of “Gates,” “Autism,” and “5 G” as proxies of conspiracy theories. The qualitative analysis demonstrated that these topics were consistent with conspiracy theories depicting Bill Gates as planning the pandemic in order to control the population and benefit from the vaccine, rehashing the myth that vaccinations cause autism spectrum disorder, and claims that fifth generation (5 G) cellphone towers caused the pandemic[2,62] (presented in Table 3). The quantitative analysis depicted in the table shows that one out of six transcripts referred to Bill Gates (15.6%, n = 200), whereas “5 G” and “autism” were mentioned in 90 (7%) transcripts each. These references dominated topic 8 and were also prevalent in topic 1.

Word clouds with the top words in the data set and in 5 sample topics are presented in Figure 2.

Figure 2.

Word clouds of the most common terms from the YouTube video transcript topics. A word’s relative size indicates its relative usage for a given topic of transcripts (larger = more usage).

Finally, the fifth research question examined the temporal distribution of the videos. The temporal dynamics of the above topics are displayed in Figure 3, which shows the proportion of each topic per each week in the data set.

Figure 3.

Temporal heatmap of the YouTube transcript topics used in each week of the data. A darker cell indicates a greater proportion of YouTube videos from that topic being used that week.

Figure 3 shows the temporal dynamics by weeks of the different topics, indicating notable changes in the relative frequencies of certain topics over time. Specifically, videos that advanced the claim that vaccines cause autism dominated the first week but were quickly replaced as a dominant frame. While this frame decreased in prevalence over the following weeks, it did not disappear, and was the fifth most common frame overall.

The most tweeted topic over time framed vaccinations as conspiracies that aimed to infringe on individual freedom and civil rights. This topic was detected from the first week and increased rather steadily with a spike in the last week of April and in the first three weeks of June. Medical trials, the second-most common framing was also tweeted throughout the data set but spiked in March and April 2020. Similarly, the third- and fourth-most tweeted topics of government news and news about the outbreak were prominent in the sample until mid-May 2020. In May to June 2020, videos featuring Bill Gates and infringement on freedom – two topics that advanced conspiracy theories and resistance of public health messages – became the most tweeted topics.

Discussion

This study examined the prevalence, propagation, dynamics, and content of YouTube videos shared in vaccination-related tweets that were posted in early COVID-19 conversations. The findings indicate that sharing links to YouTube videos was a common practice in the months following the declaration of the pandemic and that these videos typically opposed vaccination. The findings also captured a “bursty” pattern of sharing these videos, that is, links to a rather small number of YouTube videos were tweeted for short durations of time. This cross-platform sharing revealed both the enduring importance of legacy media and the emergence of non-traditional media outlets in the production of vaccine-related content. It therefore exemplifies the disruption brought by digital media and online publics. Whereas mass media dictated the public agenda in the past,[64] today social media both “rehash” mass media content and allow for new topics and non-elite, fringe voices to emerge.[68] These processes were previously documented in other contexts,[67,68] and this study extends understanding of agenda setting over Twitter to a public health context. Whereas political communication scientists hailed the potential of social media to democratize civic engagement,[68] the picture that emerged from this analysis is concerning from a public health perspective. Anti-vaccination sentiments and conspiracy theories were evident in the YouTube videos shared in the early months of the COVID-19 pandemic. This finding supports and extends previous studies that reported a strong anti-vaccination sentiment expressed on this platform.[42-45]

These results reveal that the anti-vaccination movement is using a hybrid strategy of tweeting links of YouTube videos to advance their messages and engage users. It is possible that tweeting vaccine-opposing YouTube videos is a tactic to overcome changes to the YouTube recommendation algorithms that attempt to decrease the propagation of vaccine-opposing messages.[45] Bot-like activities were previously documented in tweeting vaccine-opposing messages[16,17] and more recently in COVID-19-related tweets.[63] This study contributes to current research by showing that anti-vaccination advocates are using various tweeting strategies to boost the spread of these videos, including spamming and use of mentions. These strategies were recently identified in global disinformation campaigns,[51] and they could indicate the influence of global forces in disseminating anti-vaccine disinformation.[26,83] The findings also point to the salience of international vaccine-opposing content in videos spread over tweets in English. The prominence of RT reveals the unabashed intervention of the Russian government in promoting anti-vaccination misinformation and conspiracy theories.[16,26]

These findings are important in underscoring the relentless efforts and nimble strategies of the anti-vaccination movement. All of the tweets included in this study were parts of COVID-19 conversations posted many months before any concrete information about COVID-19 vaccines was available to the public. Our data clearly show that vaccine-opposing entities utilized COVID-19 conversations to promote their messages from the first week when the pandemic was announced. Therefore, they expose efforts of anti-vaccination advocates to “hijack” conversations about the pandemic to advance their agenda rather than to raise sincere concerns. These efforts were carried out as a tragic global crisis was unfolding and lives were lost.

Our study is also important for identifying specific frames and their persistence over time. Whereas the long-standing hoax of framing vaccinations as causing autism[84] was the first main frame of these videos, they quickly switched to framing vaccinations as infringing on individual freedom and the creation of new frames featuring Bill Gates as the major villain behind the pandemic. While another study examined depiction of vaccination on YouTube,[85] this is the first that examined framing in a large number of YouTube videos over time during the pandemic.

Unfortunately, health organizations were not the leading information sources driving the agenda in our data. While they were depicted in news that were shared over tweets, they did not directly communicate with the public via cross-platform sharing of YouTube videos over Twitter. The absence of evidence-based information communicated by public health figures is alarming, especially in view of the prominence of videos produced by alternative media sources that oppose vaccination and spread misinformation and conspiracy theories.

The strengths of this study stem from its analysis of a large data set that was collected during a historically important period. Moreover, we employed cutting-edge, interdisciplinary computational methods and human coding to investigate a previously unexplored communication strategy within vaccine discourse on Twitter. However, this work is not without limitations. First, our methods allow for exploring online content, and do not establish relationship between this content and online publics’ vaccine hesitancy. Moreover, tweets were collected by searching common vaccine-related keywords and hashtags. While these keywords and hashtags were identified following an extensive literature review and analysis of tweets by multiple research teams, it is possible that some emerging keywords and hashtags were not included. Additional limitations are grounded in our focus on tweets in English during a specific time frame. Future studies should expand research to include additional languages and time frames, particularly during and following the rollout of the COVID-19 vaccines. In addition, many vaccine-related tweets and videos – particularly those advancing vaccine-opposing messages – were deleted by Twitter and YouTube by the time of analysis. Hence, the actual number of anti-vaccination tweets and videos shared is likely higher than what we were able to report, and their content might be different from what was collected. Similarly, we might have under-reported the number of unique texts across all tweets, as we used computational methods to search for exact text matches. Notably, the time lapses between these tweets were significantly longer than reported for other coordinated inauthentic behaviors of link sharing.[78] While there is some evidence of coordination in sharing the videos between user accounts, we cannot determine the strength and degree of inauthenticity of this strategy in spreading the videos.

Finally, we focused on vaccine-related tweets that were part of COVID-19 Twitter conversations. Although our data set is unique in including all related tweets rather than a sample, our findings do not apply to vaccine-related discourse on Twitter that was not part of the pandemic discourse.

Conclusions

Although YouTube claimed to have reduced the availability of misinformation including anti-vaccination videos on its platform years before the data collection period,[43] this study documented that it continued to feature vaccine-opposing content, misinformation, and conspiracy theories early in the pandemic. It is therefore paramount that more attention be given to censoring anti-vaccination content, as well as to creating and propagating scientific evidence and clear messages about vaccine efficacy and safety. Such messages should include narratives and information that accurately contextualize the risks and benefits of vaccinations. The absence of such messages and policies early in the pandemic likely helped conspiracy theories to flourish. Accurate and transparent information that refutes claims made in vaccination-opposing videos are necessary in preventing individuals who are vaccine hesitant from joining those who actively oppose vaccinations.

The findings of this work pave the way for future studies that would answer additional questions. First, researchers should expand the scope of this study by examining URLs shared after June 2020, especially as new COVID-19 vaccines were approved and disseminated. Given the global importance of the pandemic and vaccinations, studies should also increase the scope of analysis to include languages other than English. Research should also examine relationships between video visuals, content, and dissemination. Finally, we call for hypothesis-driven communication interventions to measure how and why anti-vaccination messages propagate over social media and that attempt to prevent propagation and instead advance scientifically accurate content. Such future interventions should not focus on one social media platform; rather, they should consider and integrate cross-platform use for message sharing.

Click here for additional data file.

LDA	Latent Dirichlet Allocation
URL	Uniform Resource Locator

Topic Number	20 Most Probable Topic Words
1	Right, want, come, time, mean, look, tell, let, happen, way, day, world, good, year, United States, try, need, trump, guy, start
2	Virus, work, antibody, coronavirus, test, trial, cell, year, phase, different, develop, time, right, protein, human, company, need, start, look, mean
3	Country, United States, work, world, China, need, health, government, want, good, time, question, today, come, year, coronavirus, global, state, great, look
4	Virus, coronavirus, case, number, test, time, disease, death, spread, look, come, mean, infection, flu, day, need, right, infect, way, rate
5	Child, year, study, polio, cause, look, time, come, vaccination, mean, want, vaccinate, disease, kid, CDC, doctor, autism, tell, right, start
6	Right, look, kind, way, want, work, different, system, need, come, time, point, good, try, mean, vitamin, food, technology, level, basically
7	Virus, cell, human, aids, disease, paper, research, look, body, cause, come, find, lab, year, mean, HIV, Wuhan, time, animal, protein
8	Bill gates, world, gates foundation, health, control, gate, global, year, million, billion, government, population, dollar, coronavirus, new, foundation, fund, gates, report, work
9	Look, right, market, guy, let, want, short, come, big, money, little, break, time, today, dollar, price, percent, stock, long, way
10	Flu, mask, swine, shoot, year, United States, shot, good, government, million, kill, claim, doctor, tell, die, time, program, face, case, disease
11	God, come, Jesus, bible, lord, man, time, verse, day, word, Christian, want, pay, shall, Israel, chapter, mark, world, church, beast
12	Patient, drug, chloroquine, doctor, hospital, coronavirus, oxygen, test, treatment, need, day, treat, immune system, India, work, blood, Africa, die, zinc, medicine