Public health organizations posted fewer tweets (1149) than news media organizations (14,562). On average, public health organizations received fewer replies but more retweets and likes per tweet (51 replies, 337 retweets, and 615 likes/tweet) compared with news media organizations (65 replies, 146 retweets, and 420 likes/tweet).

The machine-learning algorithm categorized 15,711 tweets into 37 clusters, achieving 75% accuracy and a 0.76 F1 score, indicating good and acceptable performance based on the previous research standards⁸².

Tweet topics

The 37 clusters were manually grouped into six topics: (A) policies, methods, and action, (B) case updates, (C) opinions and responses, (D) medical research and treatment information, (E) impacts and consequences, and (F) health instructions and suggestions. Table 2 provides descriptions and examples. Table 3 shows tweet counts and the number of replies, retweets, and likes per tweet.

1. a.

   (A) Policies, methods and action. This topic included organizational measures to curb virus spread, such as density restrictions, altered supermarket hours, service guarantees, travel bans, and hospital procedure changes. It also covered economic support policies like a $1 billion boost for the US Department of Veterans Affairs, payroll tax relief in Victoria, and initiatives to boost car sales in China. Actions by individuals or organizations to combat COVID-19, such as Ruth May’s call for retired nurses to return, Justin Bieber’s support video for Wuhan, and safety measures by retailers like Walmart, were also included. This was the most-tweeted topic, with 4231 tweets (26.9%).

2. b.

   (B) Case updates. This topic focused on real-time reports and statistics on COVID-19 outbreaks, transmissions, case numbers, and death tolls by region. It included notable events such as the Ruby Princess cruise ship outbreak in Australia and Prince Charles testing positive in the UK. This topic was covered by 3,647 tweets (23.2%).

3. c.

   (C) Opinions and responses. This topic captured reactions and viewpoints from politicians, health experts, influencers, and the public on COVID-19 and related events. Examples include debates on reducing social distancing in the US, criticism of President Trump’s suggested treatments, and New Jersey’s social distancing measures. This topic was covered by 3242 tweets (20.6%) and received the highest average engagement with 121 replies and 551 likes per tweet.

4. d.

   (D) Medical research and treatment information. This topic mainly included news and research on virology, pathology, transmission methods, medications, vaccines, and treatment standards. Examples include challenges in preventing COVID-19 compared to SARS, studies on transmission through prolonged contact, and claims about chloroquine and hydroxychloroquine as treatments. This topic was covered by 1819 tweets (11.6%).

5. e.

   (E) Impacts and consequences. This topic reported the health, economic, and social impact of COVID-19. It included COVID-19 casualties, economic downturns, business bankruptcies, school and workplace closures, and shifts to online education and work. It also covered major event disruptions like the postponement of the Tokyo Olympics, closures of California’s Universal Studios and LA Zoo, and the cancellation of BTS’s Seoul tour. This topic was covered by 1638 tweets (10.4%) and received the lowest engagement (24 replies, 74 retweets, and 213 likes/tweet).

6. f.

   (F) Health instructions and suggestions. This topic provided guidance on preventing virus spread, including mask-wearing, social distancing, staying home, and hand hygiene. It also addressed frequently asked questions, shared expert advice, and provided links to relevant resources. An example tweet: “Are facemasks useful? Where can I get the latest travel advice? Can I get coronavirus from food? Watch this video for answers.” Despite being the least-tweeted topic (n = 1,34, 7.2%), it received the highest average retweets (221 retweets/tweet).

A comparison of number of tweets and public engagement between public health and news media organizations on various topics

A significant difference was found in the number of tweets the two types of organizations posted on each of the six topics (Fig. 1). Public health organizations tweeted the most about “health instructions and suggestions” (n = 604, 52.6% of their total), while news media tweeted the fewest on this topic (n = 530, 3.6% of their total). Public health organizations tweeted least on “impacts and consequences” (n = 31, 2.7% of their total), while news media tweeted most about “policies, methods, and action” (n = 4102, 28.2% of their total).

Comparison of the percentage of tweets on each topic by organization type. Superscripts ‘a’ and ‘b’ indicate statistically significant differences between the two organization types in the percentage of tweets within each content category, based on a chi-square test (p < .005)..

Public engagement varied significantly between the two organization types across all six topics. Overall, “likes” were the most common response, followed by “retweets” and then “replies” (Fig. 2). Tweets on “medical research and treatment information” attracted the most retweets (public health: 533 retweets/tweet vs. news media: 179 retweets/tweet) and received the most replies and likes from public health organizations’ tweets (72 replies/tweet, 980 likes/tweet). For news media organizations, “opinions and responses” had the highest replies and likes (123 replies/tweet, 553 likes/tweet). “Impacts and consequences” received the lowest engagement for both types of organizations (public health: 31 replies/tweet, 182 retweets/tweet, 317 likes/tweet; news media: 24 replies/tweet, 72 retweets/tweet, 211 likes/tweet).

Comparison of replies, retweets and likes per tweet on each topic by organization type.

Communication phases

Five communication phases were identified by the trend, amplitude, and inflection point of the number of tweets, replies, retweets, and likes (I) inception phase, (II) awareness phase, (III) panic phase, (IV) spreading phase, and (V) cohabitation phase (Fig. 3).

Changes in the number of tweets and the public engagement they attracted over time.

Phase I: inception phase (Jan 1–21, 2020)

Started with the WHO’s first COVID-19 announcement (Jan 5) and ended after China confirmed human-to-human transmission (Jan 21). Tweets focused on identifying the virus and its potential threat. For example, a tweet posted by China Daily claimed, “An expert team found a new coronavirus on Tuesday at a lab after analyzing samples collected from the cases”, illustrating the early efforts to comprehend the emerging health crisis, marking this period the “inception phase”. Despite the fewest tweets posted per week (24 tweets/week) during this period, these tweets sparked an intense response from the public, attracting the highest average weekly replies and retweets (16 replies/tweet/week, 62 retweets/tweet/week). The public’s strong interest reflected the urgency to share early information.

Phase II: awareness phase (Jan 22–Feb 4, 2020)

The pandemic spread globally, with initial cases in Singapore (January 23), France (January 24), and Australia (January 25). Quarantine in Wuhan and US flight suspensions drove a surge in tweets (789/week), receiving the highest average likes (106 likes/tweet). This phase reflected a high level of public engagement of the virus crossing borders signifying the onset of the so-called “awareness phase”.

Phase III: panic phase (Feb 5–18, 2020)

Global deaths reached 1,000, and WHO named the virus COVID-19. Despite rising cases, tweets dropped to 280/week, and public engagement was low (7 replies/tweet/week). Such a brief period of stagnation was termed the “panic phase”, characterized by global panic and the severe disruption to everyday life caused by the pandemic. For example, the Australian tweeted “Rumours of toilet paper food shortages send Hong Kong residents into coronavirus panic buying frenzy”.

Phase IV: spreading phase (Feb 19–Mar 31, 2020)

After confirming more than 118,000 cases and 4,291 deaths in 114 countries, WHO declared COVID-19 a pandemic, pushing tweet activity to its peak (1,086/week), known as the “spreading phase”. Discussions covered virus mutations, vaccine research, and health recommendations, but engagement was lower (71 likes/tweet/week), indicating information overload amidst the crisis.

Phase V: cohabitation phase (Apr 1–May 21, 2020)

The number of tweets declined (998/week), and retweets hit their lowest (24/tweet/week). With fewer “hot topics,” public engagement waned as people adjusted to living with the virus, marking the normalization of the pandemic. Consequently, this period has been called the “cohabitation phase”.

A comparison of number of tweets and public engagement between the two types of organizations in different phases

News media organizations consistently posted more tweets than public health organizations across all five phases. In the inception phase, news media organizations posted 5.5 times more tweets, peaking at 24 times in the awareness phase, and then gradually declining to 19.7, 14.4, and 10.2 times in the panic, spreading, and co-habitant phases, respectively.

In terms of public engagement, news media organizations attracted 2.1, 1.1, and 2 times more replies per tweet than public health organizations in the inception, awareness, and cohabitation phases, respectively. However, during the panic and spreading phases, news media organizations received only 23.3% and 84.5% of the replies per tweet compared to public health organizations. News media organizations consistently attracted fewer retweets per tweet than public health organizations across all phases, ranging from 65.4 to 57.2% of the retweets per tweet that public health organizations received. Similarly, news media organizations attracted more likes per tweet in the inception (1.04 times) and awareness (1.4 times) phases, but fewer likes in the panic (47.3%), spreading (55.9%), and cohabitation (83.6%) phases.

Significant differences were found between the two organization types in the number of tweets, retweets, and likes per tweet across all five phases, and the number of replies per tweet in the inception, panic, and spreading phases (Table 4). The chi-square tests confirmed these differences with p-values < 0.05 for all highlighted metrics.

A comparison of number of tweets and public engagement between two organization types on various topics across five pandemic phases

During the inception phase, public health organizations posted 11 tweets over three weeks (Table 5). Three tweets on “case updates” were from the Chinese CDC (n = 2) and the Australian DHAC (n = 1), attracting minimal engagement (1 reply, 2 retweets, and 4 likes per tweet). The remaining eight tweets were from the US CDC, covering “health instructions and suggestions” (n = 3), “policies, methods and action” (n = 3), “impacts and consequences” (n = 1) and “medical research and treatment information” (n = 1). The tweets on “policies, methods and action” attracted the most retweets and likes (575 retweets and 516 likes/tweet), and the tweets on “impacts and consequences” attracted the most replies (54 replies/tweet). In contrast, news media organizations posted 60 tweets in this period, dominated by “case updates” (n = 30). Only two tweets each were on “health instructions and suggestions” (The Australian: n = 1; China Daily: n = 1) and “policies, methods and action” (China Daily: n = 1; CNN: n = 1). The 14 tweets on “medical research and treatment information” attracted the highest engagement (109 replies, 342 retweets, and 419 likes/tweet), whereas the seven tweets on “opinions and responses” saw the lowest (3 replies, 10 retweets, and 35 likes/tweet).

In the awareness phase, public health organizations increased the number of tweets, posting 63 tweets in 2 weeks. The most frequent topic was “health instructions and suggestions” (n = 22), with 17 from the Australian DHAC, four from the UK NHS and one from the Chinese CDC. The fewest tweets were on “opinions and responses” (n = 3), all from the US CDC, which attracted the highest engagement (68 replies, 477 retweets, and 563 likes/tweet). The 153 tweets on “impacts and consequences” attracted the lowest engagement (11 replies, 443 retweets, and 102 likes/tweet). In contrast, the four tweets on “impacts and consequences” (Chinese CDC: n = 2, Australian DHAC: n = 2) attracted the lowest engagement (5 replies, 16 retweets, and 13 likes/tweet). News media organizations also increased their activity, posting 1,515 tweets. Their most common topic was “policies, methods, and action” (n = 483), while “health instructions and suggestions” (n = 32) were the least posted. Tweets on “case updates” (n = 419) attracted the most engagement (35 replies, 145 retweets, and 270 likes/tweet).

In the panic phase, public health organizations posted 27 tweets over 2 weeks, a decrease from the previous phase. “Health instructions and suggestions” remained the most frequent topic (n = 10), with contributions from the Australian DHAC (n = 5), UK NHS (n = 3), and Chinese CDC (n = 2). These tweets attracted the highest average replies (103 replies/tweet). The single tweet on “impacts and consequences” from the Australian DHAC attracted the fewest average replies (2 replies/tweet). The three tweets on “medical research and treatment information” from the US CDC (n = 2) and the Chinese CDC (n = 1) attracted the highest average retweets (425 retweets/tweet). The three tweets on “policies, methods and action” from the US CDC (n = 2) and the UK NHS (n = 1) attracted the highest average likes (634 likes/tweet). Conversely, the three on “opinions and responses” from the Australian DHAC (n = 2) and the Chinese CDC (n = 1) saw the lowest average retweets and likes (7 retweets and 4 likes/tweet). Similarly, news media organizations reduced their posts to 532 tweets during this period. The most common topic was “policies, methods, and action” (155 tweets), which also received the highest average retweets (81 retweets/tweet) and likes (219 likes/tweet). Tweets on “case updates” (n = 131) had the highest average replies (19 replies/tweet). In contrast, “impacts and consequences” (78 tweets) saw the lowest engagement (5 replies, 15 retweets, and 48 likes/tweet).

In the spreading phase, public health organizations increased their posts with 423 tweets over seven weeks. “Health instructions and suggestions” dominated with 220 tweets, while “impacts and consequences” had the fewest (n = 13). The 83 tweets on “medical research and treatment information” received the highest engagement (120 replies, 860 retweets, 1,525 likes/tweet), whereas the 30 tweets on “case updates” saw the lowest (47 replies, 199 retweets, and 326 likes/tweet). News media organizations also saw a sharp rise, posting 6093 tweets. The most common topic was “policies, methods, and action” (n = 1807), while “health instructions and suggestions” remained the least covered (n = 244). “Case updates” attracted the most average retweets (189 retweets/tweet), while “opinions and responses” attracted the most average replies and likes (112 replies and 536 likes/tweet). In contrast, “impacts and consequences” (n = 746) saw the lowest engagement (23 replies, 91 retweets, and 240 likes/tweet).

In the cohabitation phase, public health organizations posted 631 tweets over seven weeks, with 349 on “health instructions and suggestions”. However, the 74 tweets on “case updates” attracted the highest engagement (64 replies, 385 retweets, and 1244 likes/tweet). In contrast, the 12 tweets on “impacts and consequences” (Australian DHAC: n = 8, UK NHS: n = 3, US CDC: n = 1) had the lowest engagement (22 replies, 81 retweets, and 160 likes/tweet). News media organizations posted 6435 tweets, still dominated by “policies, methods and action” (n = 1655), while “health instructions and suggestions” remained the least covered (n = 237). The 687 tweets on “medical research and treatment information” attracted the most average retweets (207 retweets/tweet), while the 1604 tweets on “opinions and responses” attracted the most average replies and likes (150 replies and 640 likes/tweet). “Impacts and consequences” (n = 626) continued to receive the lowest engagement (31 replies, 65 retweets, and 224 likes/tweet).

Differences in public engagement by topic uzing different communication strategies

Use of hashtags.

The correlation between the number of hashtags and public engagement varied for public health and news media organizations (Table 6). For news media organizations, a significant negative correlation was observed between hashtag usage and replies, retweets, and likes across all topics (all P < .0.01). In contrast, for public health organizations, the same negative correlation appeared only in the “case updates” topic (P < 0.01 for replies, retweets, and likes). Significant positive correlations were found for public health organizations between hashtag use and engagement in “health instructions and suggestions” and “policies, methods, and action” (P < .01 for replies, retweets, and likes), as well as for retweets and likes in “medical research and treatment information” (P < 0.01).

Length of the tweet

The correlation between tweet length and public engagement varied between public health and news media organizations (Table 7). For public health organizations, tweet length showed significant positive correlations with replies, retweets, and likes in “medical research and treatment information,” “health instructions and suggestions” (all P < 0.01), and “opinions and responses” (P < 0.05 for replies and likes; P < 0.01 for retweets). In contrast, for news media organizations, positive correlations were found only for “health instructions and suggestions” (P < 0.01 for replies) and “opinions and responses” (P < 0.01 for replies, retweets, and likes).

No significant correlation was observed for public health tweets in “case updates,” “impacts and consequences,” or “policies, methods, and action”. However, tweets from news media organizations showed significant negative correlations between tweet length and replies, retweets, and likes in “impacts and consequences” and “policies, methods, and action” (all P < 0.01), as well as for retweets and likes in “case updates” (P < 0.01).