During the 4-month test period, the Terra Explore account gained over 12 000 followers, and the 48 videos published garnered 2 106 504 views, 3579 comments, and 5144 shares. The videos had a combined 22 297 h of watch time. Individual video views ranged between 522 views and 684 100 views, with a median of approximately 2400 views. Seven videos received over 90 000 views, and 30 videos received less than 3000 views. Videos were uploaded at various times throughout the day, and the videos that received the most views (> 90 000) were uploaded either in the morning or early afternoon Mountain Standard Time (MST, UTC-7) (Fig. 2). However, there was no statistically significant correlation between video upload time and the number of video views (p > 0.05). We did not spread video publication evenly throughout the week, and the greatest number of videos were posted on Tuesdays. Videos that were published on Wednesdays and Thursdays received fewer views than other days of the week, with videos published on Mondays receiving the most cumulative views on average (Fig. 3).

Tables 2 and 3 record the topic and metrics of our 10 most-viewed and 10 least-viewed videos. Our highest viewed video described an offset curb in Hayward, California that was famous among geologists for recording the continual slow motion along the Hayward Fault (part of the San Andreas fault system). Three of our 10 most-viewed videos were GMVs, one was a demonstration, and the rest were lecture-style videos. The lecture-style videos were either related to discussing a geologic topic related to a specific location or a recent newsworthy event or anniversary. Our 10 least viewed videos included five lecture-style videos, two demonstration videos, one in-person video, one miscellaneous video, and one meme-style video. These lecture-style videos were all explainer videos of topics related to seismology or geodesy. Our least-viewed videos were all published in either October or November 2021, when we first began our account. However, our second and third most-viewed videos were also published during this timeframe. Our top 10 most-viewed videos averaged 67 s in length and had an average video view duration of 58 %, while our 10 least-viewed videos averaged 44 s in length and had an average video view duration of 37 %. The top 10 most-viewed videos had an average engagement rate of 11 % for likes, while the 10 least-viewed videos had an average engagement rate of 8 % for likes.

Of the account's over 2.1 million video views, 95 % of those views came from videos shown on the For You page. Nearly twice as many videos were viewed on the Terra Explore profile page than on the Following feed. Videos were rarely discovered by a user directly searching for a specific hashtag (Table 4). Our most-viewed videos received nearly all (∼ 97 %) of their views from the For You page. Videos with fewer views were shown to fewer users on the For You page and received a higher percentage of views from the Terra Explore profile page or the Following feed (Fig. 4).

A total of 85 % of the account's followers are located within the United States, followed by 3 % from Canada; 84 % of the total number of video views came from users located within the United States, with each individual video averaging 78 % of its views from users in the United States. Canada was the second highest video viewer region on 22 videos, Australia on 17 videos, the United Kingdom on 5, and the Philippines on 4, with the secondary country yielding, on average, 5 % of a video's views. As of February 2022, the Terra Explore account had 36.5 % female and 63.5 % male followers. This percentage remained relatively unchanged over the 1-month duration of video posts.

The average percentage of time a video was viewed was 44 % of its duration, and on average, 19 % of users watched the full duration of a video. We evaluated the relationship between the average view duration of a video (%) and the number of video views based on the length of the video (< 30 s, 30 s to 1 min, 1 to 2 min, and > 2 min in duration) to assess whether longer watch times resulted in greater promotion to other users, resulting in higher views (Fig. 5). Amongst all videos, there is a weak positive correlation that is statistically significant between the average video view duration and number of video views (r = 0.37, p = 0.008). Amongst the individual video length populations, only videos between 30 s to 1 min long showed a statistically significant correlation between average video view duration and number of video views (r = 0.57, p = 0.002). Videos < 30 s averaged the highest average view duration (63 %). However, there is a statistically significant negative correlation between the average video view duration and the length of a video (r = -0.41, p = 0.004) (Fig. S6 in the Supplement), indicating that the average view duration is, in part, tied to the length of the video itself.

Videos had average engagement rates of 10 %, 0.38 %, and 0.21 % for likes, comments, and shares, respectively. These engagement rates are similar to the engagement rates of other analyzed educational science content on TikTok (Habibi and Salim, 2021). We found a statistically significant strong positive correlation between the number of video views and the engagement rate of likes for videos with over 90 000 views (r = 0.90, p = 0.006). However, there is no statistically significant correlation between the number of video views and the engagement rate of likes for videos with fewer than 90 000 views (p > 0.05). We also observe no statistically significant correlation between the number of video views and the engagement rate of comments or shares (p > 0.05) (Fig. 6). Therefore, videos that received a large number of views (> 90 000) did not necessarily have a higher engagement rate than videos with far fewer views (∼ 2000).

We observed no statistically significant correlation between the average video view duration (%) and the engagement rate of likes, comments, or shares (p > 0.05) (Fig. 7), two factors that are hypothesized to strongly impact the reach (views) a video can receive. Just because a user watches more of a video does not indicate that they are more likely to engage with the video. However, our analyses indicated that the average video view duration is, in part, tied to the length of the video. In looking at video length, we found a weak positive correlation that is statistically significant between the length of a video and the engagement rate of likes (r = 0.34, p = 0.017). There was no statistically significant correlation between video length and the engagement rate of comments (p > 0.05), but there was a weak to moderate positive relationship that is statistically significant between video length and the engagement rate of shares (r = 0.46, p = 0.001). Thus, longer videos tended to have higher engagement rates (Fig. 8).

Across our 48 published videos, we used a total of 81 unique hashtags. Our videos most commonly used the hashtags “geophysics”, “geology”, and “earthquake” (Table 4). Even though the hashtags “geophysics” and “geology” were used on the most videos, videos that used the hashtag “earthquake” generated the most cumulative views. Hashtags like “SanAndreas” and “California” that were only used in four videos were also associated with a large number of views.

Only one video had higher than 1 % of its views come directly from searching for hashtags or clicking on a hashtag to view videos (video ranked number 11 on Fig. 5). That video, explaining the Hunga Tonga-Hunga Ha`apai volcanic eruption in January 2022, received 3 % of its views from direct hashtag searches. This post included the hashtags “HungaTonga” and “Tonga”, from which the views were likely derived.

Between October 2021 and February 2022, the cumulative views of all videos using the “geology” hashtag across TikTok increased by 167.5 million views, with Terra Explore videos contributing 9 % of those views, and the “geophysics” hashtag increased by 1.2 million views, with Terra Explore videos contributing 89 % of those views.

Nine of the 48 videos produced featured ground motion visualizations (GMVs) of recent notable earthquakes. All nine of the GMV videos highlighted earthquakes that were of magnitude 6 or larger. GMV videos were uploaded the day of or the day after an earthquake occurred and were posted approximately twice a month during this 4-month duration. Views for these videos ranged between 2106 and 372 300 views. Eight videos were under 60 s in length, and one was 100 s long. The videos with the most views had a duration between 30–60 s, but there is no statistically significant correlation between video length and the number of views (p > 0.05) (Fig. 9a). We found that there is a strong positive statistically significant correlation between the average video view duration and the number of video views (r = 0.847, p = 0.004) (Fig. 9b). There was no statistically significant correlation between the number of video views and the engagement rate of likes, comments, or shares (p > 0.05) (Fig. 9c). Videos were posted at different times during the day, with the videos that received the most views being posted in the morning Mountain Standard Time (MST, UTC-7). However, there is no statistically significant correlation between time posted and number of views (p > 0.05) (Fig. 9d). Regardless of the global location where the earthquake occurred, on average, 84 % of each videos' views came from users located within the United States. Therefore, these GMV videos were largely not being shown to users in countries where the earthquake occurred but rather shown to users in North America (where the visualization is focused).

The vast number of video views received from the For You page rather than the Following feed demonstrates that content on TikTok is primarily consumed from the For You page and that the success of a video largely hinges on whether it will continue to be shown to more users on this algorithmic feed. Our findings are in accordance with TikTok's internal data that users spend the majority of time (nearly 70 %) on the For You page on the app (Stokel-Walker, 2020). With users spending less time on the Following feed, this feed provides a lower overall source of video views. Even when a person is following an account, those videos can still be shown on the For You page. The view, although coming from a follower, would still be marked as a view from the For You page. Our findings support other work showing that video views on TikTok are less dependent on the number of followers or subscribers as compared to other platforms like YouTube (Guinaudeau et al., 2022). While the number of followers can support the credibility of an account, a large number of followers is not required to reach a large viewership on TikTok, and having a high number of followers on TikTok does not guarantee equally high video views.

Short videos (∼ 20 s or less) on TikTok may have a higher potential to accumulate views, as they can easily be played in a “loop” like a GIF, but our data suggest that users are more likely to engage and interact with longer (> 60 s) educational science videos (Fig. 8). Even though some of these longer videos did not perform particularly well in terms of view counts (Fig. 6), they still tended to have relatively high engagement rates. Videos < 20 s had an average engagement rate of 6.5 % for likes, whereas videos between 40 s and 2 minutes had an average engagement rate of 10 % for likes. While it is suspected that higher engagement rates on TikTok videos lead to higher views (Klug et al., 2021), we did not observe any clear relationship between the number of video views and the engagement rate (Fig. 6). We found that there was a strong positive correlation between the number of views and the engagement rate of likes only for our most highly viewed videos (videos with > 90 000 views). High engagement rates on videos with lower view counts may come from people who are following the account and have an expressed interest in the subject.

The lack of any discernible relationship between the average view duration of a video and the engagement rate (Fig. 7) demonstrates that these two factors are not coupled and suggests that a video likely needs both a high engagement rate and a high view duration to be promoted within algorithmic feeds. Our top 10 most-viewed videos had an average view duration of 58 % and an average engagement rate of 11 % for likes (Table 2), which is greater than the average view duration of 44 % and 10 % likes across all of our videos. Unlike platforms like YouTube, where users select which video to view, users on TikTok choose whether to continue viewing a video they were automatically shown. While some highly interested users may watch the majority of the video and engage with it, if too many users quickly scroll away, it will negatively impact the average view duration of the video and its reach. If the intent of the science content is to maximize the engagement level from the viewer, then producing longer (> 60 s) lecture-style videos is preferable. However, the reach of the video may be sacrificed with longer videos if too many users scroll away from the video too quickly. Thus, the communication objectives should dictate what video type and duration is used.

Unlike platforms such as Twitter, where users may search for a specific hashtag to view content related to a topic (Chang and Iyer, 2012), viewer behavior in regard to educational science content on TikTok largely does not entail searching for specific hashtags to view content, unless related to a specific newsworthy event. On average, only 0.02 % of our video views came directly from hashtag searches. Thus, hashtags appear most useful in categorizing content, which may aid in its algorithmic discovery on the For You page – although the factors used in TikTok's algorithm have not been publicly disclosed. The use of a hashtag itself will not necessarily determine the popularity of a video, but rather, the hashtag is reflective of the content and subject matter of the video.

We also found that there was little relationship between the time a video was posted and how well it performed (Figs. 2, 9d). Post time optimization is the concept of posting social media content at a time when most users or followers are active and online and likely to see the content, thus increasing its potential reach. As the majority of our account's followers are located within the United States, we focused our posting during daytime hours within the US before followers would log off for the night. However, it would be interesting, especially if wanting to attract an international audience, to see how videos perform if they are posted during nighttime hours in the US. We anticipate that videos should be posted at a time local to when the desired or majority audience is primarily online.

While intrinsic features such as the length of the video may impact engagement metrics, the content of the video itself is an equally important consideration when attempting to reach a wide audience and generate high engagement. Based on our most-viewed videos (Table 2), we found that videos with the highest reach focused on recent newsworthy events (e.g., an earthquake) or discussed specific location-based geology. These findings support pedagogical research that suggests that recent, real-world incident examples should be included when teaching geoscience or environmental science topics to increase student engagement and interest (Singh et al., 2022).

Given our organizational missions, a high percentage of our video content focused on earthquakes and seismology, with all notable recent earthquakes being covered in GMV videos. These videos generated high interest by providing real-time information about earthquakes in the news, shown in a data visualization format that was likely novel to the public. User comments on GMV videos typically expressed how it was a unique, interesting visualization that they had not seen before. Additionally, by showing the motion recorded on seismometers across North America, where the vast majority of our followers are located, users in those areas were able to view how the ground moved near them, adding a level of personal, location-based interest. User comments also often expressed interest in the distribution of seismic instruments in North America.

We also found high reach and engagement with videos covering other relevant, newsworthy geology topics. For example, we produced videos covering the January 2022 Hunga Tonga-Hunga Ha`apai volcanic eruption and the anniversary of the January 1700 Cascadia earthquake and tsunami. These videos were our eleventh and eighth most-viewed videos, respectively, and were presented in a “lecture style” that primarily used image overlays. Comments left on these videos typically expressed that the users enjoyed the overview of information and how it was conveyed. The Hunga Tonga video was also shown to a wider international audience, with only 66 % of views coming from users located in the United States. Thus, if the focus of the video is on a specific international location or event, the video may be served to a broader audience. Research from educational geoscience videos on YouTube also suggests that short, timely videos about natural hazards are useful for engaging the public, especially those who live near where the hazards occur (Wang et al., 2022).

Videos describing geologic processes or features were most successful when they were tied to a specific, recognizable location. These videos were done in a “lecture style”, typically using “green screen” backgrounds and image overlays. Based on user comments, videos that were about a geologic feature in a specific location that were tagged with location indicators (e.g., #Hayward #California; #Hollister #California) were shown to many people who live in those cities. Although these users may not have previously interacted with or shown interest in geoscience content, we hypothesize that location-based information was used to deliver these videos to their For You page. Based on these video comments, the high engagement was generated in part by users from those geographic locations. Many commented that they were from the locations being covered; some recognized features in the town they were familiar with, and others learned something new and interesting about the geology of where they live.

In spite of these data, there is still complexity in understanding why certain videos perform better than others. For example, we produced two demonstration videos using food as analogs for geologic concepts. A video using spaghetti strands to demonstrate earthquake magnitude received 305 500 views and was our third most-viewed video (Table 2), while a video using Halloween candy bars to demonstrate types of faults received only 840 views and was our fourth least-viewed video (Table 3). Both of these videos presented concepts that were made easy for the viewer to understand by using common food items, and they were both posted on the same day of the week (Friday) and before noon MST (19:00 UTC). It is possible that for demonstration videos like these – which are more analogous to chemistry or physics content showing experiments – that a shorter video duration is better for achieving high reach and engagement. The earthquake spaghetti video was 34 s long, while the candy bar faults video was 59 s long. A viewer may be more likely to watch a longer lecture-style video but may prefer shorter, dynamic demonstration videos.

Although we used the hashtags #WomenInSTEM and #WomenInScience on a number of videos, the majority of the Terra Explore account's followers remained male (∼ 64 % male). Given the fact that 61 % of users on TikTok are female (Statista Research Department, 2021), the gender breakdown of our followers is not representative of the demographics of TikTok users. The large percentage (25 %) of TikTok users that are under the age of 19 (Statista Research Department, 2022a) represents a critical demographic for maintaining interest in STEM. Research has shown that STEM interest begins to decline in middle school (ages 11–13), especially for girls (Archer et al., 2013; Riegle-Crumb and Morton, 2017; Sadler et al., 2012). Additionally, many students have little exposure to the geosciences during their primary and secondary education (grades K–12 in the US) (Dodick and Orion, 2003; Lewis and Baker, 2010; Ridky, 2002), which means that young girls are often left without geoscience role models. Through TikTok, we can engage with this critical demographic and hope to inspire young women to continue an interest in pursuing STEM and to see themselves as future STEM professionals.

However, the presentation of material by female scientists and the use of hashtags like #WomenInSTEM or #WomenInScience on TikTok may not be enough to gain female followers and viewership; the content itself may also have to focus on gender to be shown on women's For You pages. Experience from the lead author's personal TikTok account suggests that women may be more likely to receive videos about gender discrimination and gender disparity on their For You page than educational science content, even if they are interested in science. The lead author made a video on their personal TikTok account discussing how their follower base had become overwhelmingly male (75 % male followers) and used the #WomenInSTEM hashtag. The video gained wide female viewership, leading to a large increase in followers and flipping the percentage of followers to 86.1 % female. The second author also had a similar experience on their personal TikTok after making a series of posts about the IF/THEN organization and International Women's Day, with male following decreasing from 65 % to 17 %.

As TikTok did not provide information about the age of followers or viewers for videos posted prior to August 2022, we are unable to evaluate the full demographics of who our videos reach. Future work may benefit from creating and analyzing new videos where the gender percentage and age of viewers are reported or from using the paid “promote” feature that allows accounts to target specific demographics.

When conducting research on social media platforms, it is necessary to address the ethical issues related to privacy, anonymity, and consent of human subjects (Vitak et al., 2016). Kanthawala et al. (2022) discuss the specific ethical considerations related to research on TikTok, given its heavier dependence on algorithmic curation and its younger user base. More so than other social media platforms, TikTok has faced scrutiny over the data that it collects from its users and user security. TikTok has previously paid USD 92 million to settle dozens of lawsuits that allege that users' personal data was tracked and sold to advertisers in violation of state and federal law (Allyn, 2021). Additionally, there has been scrutiny over the biometric data, including “faceprints” and “voiceprints”, that TikTok can now collect from users (McCluskey, 2021).

According to TikTok's privacy policy, TikTok collects information that the user provides when creating an account, such as their name, age, and email, as well as information from other sources, such as linked social media accounts and third-party services, and device information, including the user's approximate location (TikTok, 2021b). By creating a user account and agreeing to the terms and conditions, TikTok users provide their consent to have their information collected. However, many users are unlikely to read the privacy policy when creating an account and will not be aware of what they are consenting to.

Although TikTok provides analytical information about the gender breakdown of an account's followers, users do not provide information on their gender when registering for an account. We hypothesize that TikTok either acquires gender information through connected social media accounts or assumes the gender of an account based on a user's activity on the app. We are therefore unable to determine how accurate the information provided on the gender of followers is and whether users are aware that this information is assumed or collected on their behalf. However, we have no reason to believe that this information would be inaccurate due to TikTok's heavy data mining and the fact that this same demographic information is also provided to million-dollar corporations for targeted advertising campaigns.

The information and data we analyze in this study are restricted to video view duration and engagement metrics and are not related to in-depth personal user information that TikTok may collect. We do not evaluate or analyze the individual followers of the Terra Explore account or include specific comments from any users, providing full anonymity. The data on video views, likes, comments, and shares are fully publicly available to all users, and only total cumulative play time, average duration the video was watched, percentage of viewers who watched the full video, and the percentage of video views by section and region is private to the account user. Therefore, the data used are largely public, and there is minimal risk to the individual users from this research. While TikTok as a platform may have larger ethical concerns that are beyond the scope of this study, we argue that the data used in this research are ethical.

Data collected over this 4-month time span demonstrates that TikTok provides significant potential for the reach and growth of science communication. As a brand-new account, the second video we posted gained over 165 000 views within the first few days and now has over 370 000 views. The algorithm-driven (rather than follower-driven) nature of TikTok creates a low barrier of entry for new creators. An account does not have to focus on creating a large following in order to achieve a wide reach. Additionally, there are no materials required to produce content other than a mobile smartphone with a camera. However, given that views from the Following feed constitute a low overall percentage of views, there may be pressure to constantly produce “viral” videos that are shown widely on the For You page. Having a large number of followers does not guarantee that a video will be seen by a large number of people.

Despite potential shortcomings, TikTok presents an opportunity to deliver educational science videos to wide audiences with relatively little effort. All six Terra Explore videos related to lidar were cross-posted on OpenTopography's YouTube channel as “Shorts”, which is YouTube's TikTok-like short-form video category. The OpenTopography YouTube channel had 1900 subscribers at this time. These videos received an average of only 35 views on YouTube, while the same videos received between 2422 and 133 500 views on TikTok. Starting in February 2022, we began cross-posting GMV videos for recent earthquakes on IRIS's Instagram account as “Reels”, Instagram's TikTok-like video category. The IRIS Instagram account has approximately 2300 followers as of fall 2022. The exact same videos received, on average, 5 times more views on TikTok than they did on Instagram. The vast discrepancy in views of the exact same videos posted on different platforms highlights the benefits of TikTok's recommendation algorithm in reaching a wide audience.

A potential challenge is that TikTok videos require an engaging, charismatic video presenter: two people could create the same video, but one video might perform significantly better if one host has a more natural, charismatic presentation style. We have not evaluated how this factor impacts video performance. In addition to the content of the video itself, practice in the presentation style of videos (which is beyond the scope of this study) is also an important component for science communicators to consider. Additionally, we acknowledge that TikTok's recommendation algorithm may change over time, although we do not anticipate such significant changes would occur in such a short amount of time that it would invalidate our results. Future studies will benefit from continued analysis of video content as well as a more qualitative evaluation of the impact and effectiveness of the science communication.

We thus recommend the following for the best chances of high reach and engagement with science communication on TikTok.

Produce lecture-style videos using green screen and image overlay effects that are approximately 60 s or longer in duration.