Introduction

‘Carve a whale on your hand with a razor, send a photo to curator’, ‘Do something painful to yourself, make yourself sick’, ‘Jump off a high building. Take your life.’ These examples present what vulnerable adolescents are challenged with when playing the suicide-inducing Blue Whale Game (Mukhra et al. 2017). Programmed in 2013 by the Russian psychologist Philipp Budeikin, media reports reports state that the Blue Whale Challenge has caused the deaths of more than one hundred adolescents worldwide (Evon 2017; Mukhra et al. 2017). Via mobile applications, the game connects players with anonymous curators, who mentor the vulnerable youngsters over a period of 50 days. During this time, the curators dare their mentees to perform escalating tasks and, ultimately, push them to commit suicide (Lupariello et al. 2019). While the Blue Whale Game most likely originated as a sensational hoax by a Russian newspaper (Evon 2017), cyberbullying researchers believe that copycats have since made use of its contagious effect (Adeane 2019; Timm-Garcia and Hartung 2017). Evidence is provided by Sumner et al. (2019), who show that social media content in support of the Blue Whale Game spread rapidly to 127 countries from 2013 to 2017. In 2018, Balhara, Bhargava, Pakhre, and Bhati presented the first medical report showing self-inflicted injuries of a young boy, who followed instructions presented by a mobile app resembling the Blue Whale Game. In 2019, five medical reports from Italian and Romanian girls were released that confirm a “‘contagious’ quality of the Internet and the importance of epidemiological, psychological, psychiatric, social, and cultural risk factors” for participating in the game (Lupariello et al. 2019, p. 641).

In addition to the Blue Whale Challenge, youngsters turned to another suicide game in 2018/2019. Reportedly, the Momo Challenge shared many similarities with its predecessor. To this day, media reports have tried to link the deaths of five teenagers to the Momo Challenge (batimes.com.ar 2018; Davidson 2018; Ferber 2018; Kitching 2018; Schneider 2018). Again, it remains unclear which incidents were hoaxes or wrongly attributed to the game. Also, little is known about the challenge from a scientific point-of-view. As part of the challenge, adolescents communicate with a WhatsApp account called Momo, which shows a picture of a grotesque sculpture as an avatar. When written to, the Momo account shares horrifying pictures and escalating assignments instigating self-harm and suicide. Cases have been reported of players refusing to perform these tasks, resulting in the account threatening to harm the participant's families (e.g., Webb 2018b) or to leak private information (e.g., newindianexpress.com 2018). Therefore, law enforcement agencies warned that the Momo Challenge might be linked to data theft (e.g., Chanda 2018; Webb 2018a).

Contrary to the Blue Whale Game, Momo accounts use snowball effects to gain reach. Journalists have therefore criticized the large amount of YouTube videos in which YouTubers have promoted the challenge by passing on the phone numbers to their viewers (theweek.co.uk 2018). Momo even became an internet meme: The horror figure made appearances in Let’s Play videos of the game Fortnite, in storylines of popular Minecraft series, and in screen recordings of the preschool television show Peppa Pig (Cooper 2019). This prompted celebrities to express their concern about the impact on young children and call on YouTube to take action (Needham 2019; Waterson 2019). Finally, YouTube published an official statement: “We want to clear something up regarding the Momo Challenge: We’ve seen no recent evidence of videos promoting the Momo Challenge on YouTube. Videos encouraging harmful and dangerous challenges are against our policies.” (YouTube 2019)

However, videos promoting the challenge did in fact exist, even though YouTube has demonetized, flagged and deleted most of these video since March 2019 (Alexander 2019), shortly after the data collection for this study was completed (turn 2018/2019). The present research raises questions about the extent to which these videos have portrayed harmful and helpful depictions of the suicide game. Additionally, we seek to understand what consequences these depictions have for the engagement rates of the videos and for community formation and what variables drive this relationship. We proceed in three steps. First, we derive an analytical scheme for assessing the extent to which these videos feature harmful and helpful depictions. Second, we combine these depictions to assess the individual danger level of each video, using this index in multivariate regression analyses to predict engagement rates, i.e., views, comments, likes, and dislikes. Third, we use methods from network community analysis to understand what media diet and content preferences specific audiences expose themselves to. Our findings suggest that more than two thirds of the videos include at least some form of harmful depiction and that the danger level is a predictor of the engagement rates that a video will receive, except for views. In accordance with this finding, the most dominant community on YouTube is also the one that prefers to engage with the most harmful depictions of the challenge. This study expands the literature on suicide games by providing an analytical scheme to assess media depictions and by drawing assumptions about the consequences of these depictions based on the literature on Werther and Papageno effects. Most importantly, it advances our understanding of the spreading mechanism of suicide games in social media by investigating the relationship between harmful depictions of suicide games and engagement rates.

But why should we even care about YouTube videos that endorse or oppose the Momo Challenge? As we will argue more extensively in the following theory section, media play a decisive role in suicide and non-suicidal self-injury prevention for adolescents; especially social media like YouTube. Drawing from the guidelines on responsible reporting on suicide (World Health Organization [W.H.O.] 2008), we argue that some characteristics of the Momo Challenge videos will be more likely to elicit Werther effects (i.e., suicide and non-suicidal self-injury imitation), while others will be more likely to elicit Papageno effects (i.e., suicide prevention). Of course, the Momo Challenge is neither the first suicide game nor is it likely to be the last. Therefore, “it is urgent to monitor social media posts related to... self-harm challenges (e.g., the Momo Challenge)” (Khasawneh et al. 2019, p. 888).

In our view, the viral spread of Momo-related content shows that suicide games will continue to be a difficult-to-handle cyber threat. While the challenge itself was probably nothing more than a hoax, institutions issued warnings that related media content, such as YouTube videos and news reports, was nonetheless dangerous to young children (Waterson 2019). New challenges, such as the Skull Breaker Challenge and the Jonathan Galindo Challenge, have already been launched as successors to the Momo Challenge. This material confronts youngsters with the concept of self-harm, sometimes for the first time in their life. Therefore, “posting about the challenges in unsafe manners can contribute to contagion regardless of the challenges’ true nature” (Khasawneh et al. 2019, p. 888). On that note, the Momo Challenge evolves into a case study for the Thomas theorem: if adolescents and parents are made to believe that the Momo Challenge is real, it has real consequences.

Theoretical background

Adolescents, YouTube use, and the role of YouTube for the Momo Challenge

“Adolescence is a period of unique vulnerability” (Miller and Prinstein 2019, p. 425), a period of biological restructuring. Acute stress is less properly processed during adolescence, often causing emotional overreactions. This makes adolescents very susceptible to harmful media depictions in social media: “Because of their limited capacity for self-regulation and susceptibility to peer pressure, children and adolescents are at some risk as they navigate and experiment with social media” (O'Keeffe and Clarke-Pearson 2011, p. 800). Nevertheless, adolescents tend to spend more time on SNS than adults (Yoon et al. 2019) and video-sharing has become “one of the fastest growing online activities among youth” (Lewis et al. 2011, p. e553). In 2018, during the peak of the Momo Challenge (July 2018 until March 2019), YouTube was the most important SNS for teenagers aged 13 to 17 (Anderson and Jiang 2018; Farokhmanesh 2018; Smith et al. 2018). This position is now increasingly contested by competitors like TikTok (Qustodio 2020). Moreover, YouTube was the platform that got under crossfire during the peak of the Momo Challenge July 2018 until March 2019 since journalists and celebrities gave hints that YouTubers did not only inform about the challenge, but also distributed alleged Momo phone numbers and dares (Needham 2019; Tarlton 2019; theweek.co.uk 2018; Waterson 2019). Finally, in February 2019, pictures of Momo, together with suicide and self-harm advice, were spliced into children’s cartoons on YouTube Kids. This was not the first time that YouTube Kids had to act against violent and disturbing content on its platform (Maheshwari 2017), but the Momo Challenge offered a new vehicle. YouTube also took on outstanding significance for the Momo Challenge because other video-sharing platforms such as TikTok were still new on the market and did not enjoy similar popularity back then. TikTok, for example, did not become a competitor until August 2018 (when TikTok merged with musical.ly), and it took until May 2020 for TikTok to catch up with the average minutes children aged between 4 and 15 spend each day on YouTube (Qustodio 2020). Even Snapchat reached approximately 15% less of the teenagers than YouTube during the peak of the Momo Challenge (Anderson and Jiang 2018) and also accounts for less time spent per day, namely 65 min compared to 97 min (Qustodio 2020). This, combined with studies indicating that YouTube videos have described tasks of the Blue Whale Challenge in the past (Khasawneh et al. 2019) and promoted NSSI videos (Bastone 2019), sparked our interest and the need to investigate the depiction of the Momo Challenge on YouTube.

Possible consequences of adolescents’ engagement with the Momo Challenge on YouTube

Among adolescents aged between 15 and 29, suicide is the second leading cause of death (W.H.O. 2017). Suicide poses a global health problem with youths as a particularly vulnerable population group at its center (Wasserman and Wasserman 2009). The media play a key role, both, in suicide promotion as well as suicide prevention (Mann et al. 2005; Niederkrotenthaler et al. 2010; Phillips 1974), mainly because “the characters’ [displayed in the media] experience suggests ways for the viewer to deal with their own problems” (Hoffner and Buchanan 2005, p. 329) either positively (Papageno effect) or negatively (Werther effect). The occurrence of both effects depends on how the media display such sensitive content (Mann et al. 2005).

Precisely, the Werther effect describes the phenomenon of an increased suicide rate after a suicide has been depicted in the media (Phillips 1974). Strikingly, even details, such as method or place of the suicide may be imitated (Stack 2000). To date, the Werther effect as a possible media effect is widely considered confirmed in research (Frei et al. 2003; John et al. 2017; Schäfer and Quiring 2015). Another self-harm inducing factor—albeit with non-fatal intentions—is non-suicidal self-injury (NSSI; Arendt 2019). NSSI is common among adolescents with an estimated rate of 5.6 to 6.7% of adolescents engaging in self-harming, non-fatal behavior (Buelens et al. 2019). In the context of suicide and suicidal behavior, NSSI is seen to be a critical factor as it may mark “the transition from suicidal ideation to actions” (Ernst et al. 2019, p. 2).

Following Bandura’s social-cognitive theory of learning, the occurrence of both, NSSI and suicide can be explained by a weakening of social norms and a reduction of inhibitions (Scherr 2016). Behavior depicted in the media can serve as a dry practice for the viewer (Gould et al. 2003; Valkenburg and Peter 2013). Importantly, it has been acknowledged that these principles do not only apply for traditional media, but also for the internet and social media (Luxton et al. 2012; Térrasse et al. 2019). For youngsters who are on the lookout for role models worth imitating, social media may become a platform of involvement and identification—two factors well-known for enlarging the chances of (imitation) suicides (Arendt et al. 2016; Hoffner and Buchanan 2005; John et al. 2017; Niederkrotenthaler et al. 2015; Till et al. 2010). Consequently, Twenge et al. (2018) were able to show that the use of social media by U.S. adolescents is associated with suicide-related outcomes.

However, depictions of suicide and NSSI may pose potential for positive outcomes, i.e., prevention, as well. Media depictions can help vulnerable recipients to seek support by depicting favorable coping strategies for suicidal ideation or behavior, stories of individuals overcoming a crisis or by providing information on counseling services (Niederkrotenthaler et al. 2010). Just as with the Werther effect, the Papageno effects are not limited to traditional media. In social media, the process of involvement and identification with role models may not only increase Werther effects, but may also enhance positive outcomes, e.g., help-seeking (Niederkrotenthaler et al. 2010). Overall, whether social media have positive or negative effects on their recipients depends on how the sensitive content is presented and framed (Markiewitz et al. 2020b; Schäfer and Quiring 2015). This may be the reason why research on the relationship of social media use and self-harming behavior is ambiguous and has produced mixed findings in the past (Markiewitz et al. 2020a).

Thus, when adolescents watch YouTube videos dealing with suicide games like the Momo Challenge, they expose themselves to the chance of encountering these effects, even if they do not participate in the challenge themselves (Uvais 2019). One the one hand, watching YouTubers play it cool with suicide games or even promote the Momo Challenge could lead to the weakening of inhibitions and social norms, thus normalizing self-harming behavior (Jonas and Brömer 2002). On the other hand, trigger warnings and references to counseling could relieve adolescents from acute stress and encourage help-seeking behavior. Given these likely to occur effects, this study examines whether and to what extent Momo Challenge videos feature depictions that are suspected of triggering these effects:

RQ1

To what extent do Momo Challenge videos on YouTube portray harmful and helpful depictions of the suicide game?

The relationship between harmful or helpful depictions of the Momo Challenge and engagement rates

Even if YouTube videos show harmful and helpful depictions of the suicide game, these depictions will have no effect on adolescents if no one actually engages with them because of them being unpopular. To measure the popularity of a YouTube video, engagement rates are used, such as views, comments, likes, and dislikes. The higher the popularity of a video, the more likely the YouTube recommendation system will recommend it to new viewers. This recommendation in turn increases the popularity of the video, resulting in a spiral process called rich-get-richer effect (Borghol et al. 2013; Welbourne and Grant 2016). This effect would be particularly problematic if it leads to the recommendation system recommending videos with harmful depictions more frequently to new viewers. For this spiral process to start, however, viewers would first have to show a preference for harmful depictions.

Do we have any reason to believe that viewers show a preference for harmful depictions? Research on the Uses and Gratifications (U&G) framework shows that entertainment motives are the strongest predictors for video viewing, liking, and disliking, and that social interaction motives best explain commenting behavior on YouTube (Khan 2017). Ladhari et al. (2020) investigated the role of emotions as a driver of popularity on YouTube and were able to find a significant positive relationship between the emotional attachment of the viewers to a YouTuber and the YouTuber’s popularity. In addition, Alhabash et al. (2015) used an experimental design to show that highly arousing video content increases the chances of behavioral reactions, such as engagement. This finding is not only consistent with other studies on arousal levels of videos and viewing / sharing behavior (Hagerstrom et al. 2014; Tellis et al. 2019), but also with arousal and news value theory. Unpleasant stimuli and negativity, especially in the form of damage, violence, and death, create high arousal and are therefore strong predictors of media engagement (Bednarek 2016; Greer 2007; Schimmack and Derryberry 2005; Takeuchi et al. 2005).

The extent to which this arousal is used in YouTube videos, i.e., the extent of harmful depictions, might vary depending on the video’s genre. For example, some genres may increase the chance of seeing interactions with alleged Momo accounts, while others may increase the chance of encountering information on counseling. It seems reasonable to assume that videos of YouTubers performing live demonstrations of the dare will portray more harmful depictions than, for example, information videos, which serve a fact-based, informative purpose. Since information videos on YouTube often snippets from news broadcasts or talk shows, chances are higher that the news anchors already adhere to guidelines on responsible reporting on suicides compared to YouTubers who perform live demonstrations of the dare. At the same time, the genre of a video could also be a predictor of its engagement rates, as certain genres could attract more viewers and encourage more users to leave a comment. It is plausible, for example, that emotionally charged art productions inspire more comments than fact-based, unemotional information videos. In addition, live demonstrations could promise a higher thrill than information videos and therefore convince more viewers to click and view. Because of these plausible associations, we need to control for video genre when analyzing the relationship between the extent of harmful or helpful depictions and engagement rates.

Following these arguments, we assume that if YouTube videos contain harmful depictions of the Momo Challenge, these videos will not only be watched but even more frequently than videos with helpful depictions. In fact, a higher extent of harmful depictions should promote engagement rates, such as views and comments, even when controlling for other content characteristics such as genre. Supporting this assumption, Lewis et al. (2011) found that NSSI videos attract more views and likes when they depict more NSSI images and multiple NSSI methods. We assume a similar trend for Momo Challenge videos and therefore hypothesize:

H1

More harmful depictions of the Momo Challenge are positively related to engagement rates.

The Momo Challenge and community formation

Engagement rates do not only show how popular a video is, but also enable community detection on YouTube. Since engagement rates can almost directly be monetized in the age of social media, YouTube’s recommendation system and autoplay feature will do their best to keep the users engaged. The lion’s share of YouTube users (81%) state that they at least occasionally watch suggestions presented by the recommendation system (Smith et al. 2018). Therefore, researchers can use engagement rates as a proxy that indicates which two videos are likely to be recommended to each other’s audience. For example, based on the previously commented videos, YouTube recommends videos that have been commented on by viewers with similar tastes. The aggregation of shared commenters not only shows how similar the audience’s preferences of two videos are but also how likely they are to be recommended to and viewed by each other’s audience. Mapping these connections leads to the identification of communities. Each community represents a group of intertwined videos that share audience preferences, similar content, and are more likely to be recommended to each other’s respective viewers.

It is very likely that such distinct communities have also developed around Momo Challenge videos. As this community formation is driven by content preferences, some communities might show an affinity for content that shows a greater extent of harmful depictions, while others might show a preference for more helpful depictions. The advantages of identifying these communities are twofold: First, it gives deeper insight into the spreading mechanism of suicide games. Second, it is a starting point for educators, parents, and adolescents to learn how to identify and avoid harmful communities and to instead look out for helpful content. Therefore, we propose a second research question:

RQ2

What communities have formed around the Momo Challenge videos and what content preferences do they show?

Toward an analytical scheme to assess the extent to which social media portray harmful and helpful depictions of a suicide game

In order to answer the research questions—and, thus, capture the extent to which videos on YouTube portray harmful and helpful depictions of the Momo Challenge—this section combines the above-discussed strands of theory and draws from practical experience with guidelines on responsible reporting (RRS) to provide an analytical scheme to capture the extent of harmful and helpful depictions. This analytical scheme is later applied to YouTube videos, but could also be used to analyze various social media content.

RSS have originally been developed for non-fictional, journalistic media content in order to improve NSSI and suicide depictions in favor of prevention, i.e., to counteract negative effects and promote positive effects. In fact, these guidelines have largely been found to obtain the desired beneficial effects (Beam et al. 2018; Pirkis et al. 2006; Schäfer et al. 2006; W.H.O. 2008; Yaqub et al. 2017). Similarly, guidelines for suicide-related content on the internet have been published by The Suicide Prevention Resource Center (SPRC 2014). In an initial, two page long presentation, Khasawneh et al. (2019) analyzed whether the portrayal of the suicide game Blue Whale Challenge on Twitter and YouTube adheres to these guidelines. The results show that 87% of the videos adhere to half or less of the guidelines. Following this fruitful approach, we adapted the media guidelines on RRS and combined them with media effect theory of Werther and Papageno effects to measure both harmful and helpful depictions.

First, for media depictions to elicit helpful media effects, guidelines suggest to include information on help and assistance (e.g., telephone counseling). Research on Papageno effects shows that indicating places to go or call when in need can have positive effects in terms of suicide prevention because this information makes it easier to overcome fears and seek help (Arendt et al. 2016; Arendt and Scherr 2017). Therefore, Momo Challenge videos should provide information on help and assistance. Second, guidelines ask for sensitivity when dealing with the subject of suicide/suicidality (Mueller 2017), e.g., YouTube videos should provide trigger warnings and should avoid portraying the issue in a highly emotionalized manner (Frei et al. 2003; Niederkrotenthaler et al. 2010). Similarly, guidelines suggest that emotionalization and evoking fear should be avoided (e.g., when YouTubers verbalize fear or depict Momo as a real entity that could pose actual harm) since this might motivate adolescents to test their courage (Johnston and Warkentin 2010). Finally, YouTubers should avoid presenting any depictions that could serve as a dry practice or tutorial for viewers on how to perform self-harm. This may include providing contact details of Momo accounts or showing how to establish contact with an alleged Momo account. Similarly, YouTubers can cause harm by showing live interactions with alleged Momo accounts (e.g., text messaging, calling, or even the completion of tasks related to the challenge).

This analytical scheme is our basis for assessing the extent to which Momo Challenge videos on YouTube portray harmful and helpful depictions of the suicide game. Although we cannot assess the actual effects, this analytical scheme at least allows us to investigate how much content circulates the platform that is suspected of triggering these effects. In the method section, we will further explicate how we put this scheme into practice by using it for building content analytical categories and combining the results of this content analysis with network analysis.

Method

A study that tries to assess the extent of harmful and helpful depictions of social media content requires a large data set. To further analyze a video’s content, its engagement rates, and community formation, it is also necessary to trace the relationships between users and media content. For this purpose, we used a common snowball sampling with multiple starting points, which crawl from the anchor points to new network members by following ties iteratively (Hansen et al. 2011; Paolillo 2008). Following this sampling technique, we collected a data set of interrelated YouTube videos dealing with the Momo Challenge, which we then subjected to content and network analyses.

Sample and data evaluation

We performed our data crawl with NodeXL, an add-in for Microsoft Excel that uses the YouTube API to collect videos automatically. NodeXL can handle medium-sized data sets of a few thousand videos and less than 200,000 edges before overstraining (Smith et al. 2010). Using the keywords ‘Momo Challenge English’, we ran a search of 2000 videos that contain these keywords in their titles, descriptions, or tags. The YouTube API was able to identify 487 videos that met our specifications, making NodeXL’s data size limitations irrelevant. We extracted the descriptive statistics of the videos, i.e., the videos’ creators, views, comments, likes, and dislikes. For each video, we downloaded 1000 top-level user comments as well as the first 100 replies to these top-level comments. Then, NodeXL created an edge between each two YouTube videos that were commented on by the same user. Overall, we identified 137,955 edges.

In order to enhance data quality, we conducted an editing process in which we controlled for English language (including subtitles) and thematic proximity (e.g., videos presenting pictures, songs, or movies about Momo, showing WhatsApp interactions with Momo accounts, or YouTubers talking about suicide games). In a time-consuming but thorough process, we watched every single video, checked its content and deleted all videos (as well as their related ties) that did not fulfill the above-mentioned criteria (e.g., videos about an eating contest of Indian noodles called ‘Momo’). This process led to the deletion of 376 videos, leaving only 111 videos in the data set.

Although it has been speculated that samples drawn from the YouTube API should be a rather accurate representation of the YouTube search results (Rieder et al. 2018), APIs never reproduce the exact same selection and order of videos found in a user’s search results. For the Twitter API, González-Bailón et al. (2012) showed that the API misses 2.5% of all tweets, 1% of the authors, and 1.3% of the hashtags found by the search query. We therefore assumed a similar limitation for the YouTube API. To address this limitation, we complemented the automated data crawl with a manual search by using the keyword ‘Momo Challenge’ and setting the order parameter to ‘relevance’, which is the default option for users. We then scrolled down the search list until it showed “No more results” and added every single video (and all related ties to videos in our existing data set) that the NodeXL crawl had missed. Of course, these manual additions are also dependent on the YouTube search algorithm and we cannot be sure that the search really captures all videos related to the Momo Challenge that YouTube has to offer. However, a manual complement to the automated data crawl is the most extensive data acquisition method to date. The procedure proved to be worthwhile because we could identify 98 videos that the crawl had originally missed and added them and their edges to the existing data. Just as suggested by González-Bailón et al. (2012), the videos detected by the search were more densely connected and central to the network than those of the API crawl, thus adding more edges to the network than were deleted beforehand. In the end, the revised data set included 209 videos and 169,519 ties.

Although it is highly likely that the collected videos were mainly watched by adolescents, as a substantial proportion of the related YouTube channels produce content for children, we needed a more reliable indicator. We therefore randomly sampled 1000 YouTube comments, either top-level or reply. To find out if it is really adolescents who engaged with the video content, we measured an age proxy: we recorded the date on which the authors of these comments joined YouTube. We assumed that more recent dates of joining, e.g., after 2014, are more likely to be associated with a younger audience, while less recent join dates definitely point to an audience of advanced age. Since the average commentator did not join YouTube until 2015 (M = 2015, SD = 2.6), we feel reassured that we are actually investigating a young audience.

Data analysis

Content analysis

After data collection, we randomly sampled 50% of the videos for coding (n = 105). One coder conducted all of the manual coding by following our codebook. The codebook was pre-tested on 10 videos, discussed, and then tested on 5 more videos, which were finally excluded from the analysis, i.e., they were treated as uncoded. Whenever uncertainties arose, we discussed the coding guidelines until consensus was found. In the end, the codebook contained a nominal measurement of the video’s genre (live demonstration, information video, art production, reference video, or other). In addition, we used the analytical scheme presented in the theory section of this paper to build a content analytical category of characteristics that indicate harmful or helpful depictions of the Momo Challenge. First, we dummy-coded all harmful depictions. This included whether or not the video showed interactions with Momo accounts (text-based and voice-based interactions were coded separately), verbalized that Momo is an entity to be feared, and/or provided contact details of Momo accounts. Second, we coded all danger-related characteristics that counteract these possible harmful effects, namely the prompt of trigger warnings (automated or by YouTuber) and the provision of counseling contacts. Finally, we calculated the intra-rater reliability by reassessing 10 videos after two months had passed. Intra-rater reliability (Holsti) was more than 0.9 on all variables—except for automated trigger warnings (0.72). However, this exception is easily explained by the fact that YouTube added additional trigger warnings to their videos after celebrities demanded the platform to take action.

After the coding procedure, we combined the count of these depictions to form a danger index. To this end, we added up all helpful depictions that may reduce the likelihood of suffering from negative effects (trigger warnings, counseling) and subtracted this number from the sum of harmful depictions (text- and voice-based interactions, fear, contact details). The asymmetric index scales from − 2 (very unproblematic) to 4 (very problematic). By subtracting the count of helpful depictions we assume that harmful content is compensated by helpful content, which reduces the overall extent of harmful depictions. This is especially important with regard to educational content that might portray harmful depictions of NSSI and suicidal ideation but also provides trigger warnings, information on counseling, and encourages adolescents to talk to their parents about what they have just watched. For example, an educational video that portrays a young adult engaging in text interactions with “Momo” would be rated 1 (unproblematic) if it also displays trigger warnings and counseling information on whom to turn to in case of involvement in the challenge or NSSI. We believe that with an asymmetric sum index the assessment of the extent of harmful depictions becomes more intuitive and accurate, which prevents alarmism.

Network analysis

For the network analysis, we imported our data files to Gephi, an open-source network visualization software. Gephi offers the application of a modularity algorithm that outperforms its fellow clustering algorithms in terms of simplicity, speed, and precision. Blondel et al. (2008) developed this algorithm to optimize community identification within large social networks. By employing the modularity algorithm to our data (for a complete documentation, please refer to the supplementary material), we were able to distinguish seven communities within the Momo Challenge network. For network visualization, we unfolded the Momo Challenge network using the common Force Atlas 2 algorithm, while filtering isolated network components consisting of three or less nodes. Filtering these marginal components has proven to be a vital technique among network researchers because otherwise the network would clutter (Hansen et al. 2011).

Results

We found 209 Momo Challenge videos, with an average reach of 524,671 views and 3,242 comments (568,108 views and 3851 comments in the sample of our content analysis, n = 105). The most watched video attracts more than 11 million users. Even for YouTube standards, these are impressive engagement rates (Lewis et al. 2011). Of the videos, 34% are information videos (news clips or talk shows), which serve a fact-based, informative purpose. Another 31% of the videos present art productions (e.g., short films, animated storylines, Let’s Plays, and make-up tutorials). Live demonstrations of how YouTubers try to establish contact with Momo accounts also make up an essential part of the content (24%). Additionally, we identified reference videos, which focus on YouTube celebrities reacting to other channels’ Momo-related content (8%) and three videos that did not fit into any of these categories (other).

The distribution of views and comments is highly skewed within the Momo Challenge network, with one striking extreme value (#1 in Fig. 1). This extreme value #1 represents a video in which a YouTuber presents a set of distinct ‘cursed’ phone numbers that are supposed to cause the death of the caller. Because the video deals with a wide range of cursed numbers in general, it reaches a wider audience than those that focus exclusively on the Momo Challenge. For comparison, the second and the third placed video attract approximately 3 to 4 million views. Since the video #1 is an extreme outlier in terms of the number of views and comments, it was excluded from subsequent calculations. After filtering, the average number of views and comments for the content analysis sample decreased to 471,899 views and 3076 comments.

Fig. 1
figure 1

View and comment count of videos related to the Momo Challenge. The extreme value #1 was excluded from all further calculations that related to viewer and comment count

Full size image

Figure 1 also shows that few videos receive the lion's share of the views and comments, while most videos struggle to attract any audience at all. This trend is reinforced by a strong, positive correlation between engagement rates (see Table 1) such as views and comments (ρ = 0.619, p < 0.001), views and likes (ρ = 0.666, p < 0.001), and views and dislikes (ρ = 0.728, p < 0.001), suggesting that Momo Challenge videos that attract a high number of viewers are also more likely to be engaged with. User engagement in turn makes the video more likely to be viewed, leading to a rich-get-richer effect (Borghol et al. 2013; Welbourne and Grant 2016).

Table 1 Bivariate correlation for danger predicting engagement (n = 104)
Full size table

The extent of harmful and helpful depictions of the Momo Challenge (RQ1)

In order to answer the first research question of this study, namely the extent to which the YouTube videos studied feature harmful and helpful depictions of the Momo Challenge, we have to consider descriptive statistics. Table 2 presents the proportion of videos that feature harmful or helpful depictions, broken down by video genre.

Table 2 Proportion of videos featuring harmful and helpful depictions of the Momo Challenge, by video genre (n = 104)
Full size table

Let’s first look at harmful depictions. 47% of YouTubers are exchanging text messages with alleged Momo accounts. Additionally, phone or video calls are shown in 15% of the videos. In sum, about 53% of the videos under study demonstrate at least some kind of interaction and 57% verbalize that Momo is an entity to be feared. Yet, this does not stop the videos’ creators from further promoting the challenge by passing on alleged Momo phone numbers (25%).

Helpful depictions, on the other hand, are rare. Only 11% of the videos prompt an automated trigger warning that is aimed at preventing these media depictions from harming viewers. In addition, less than a quarter of YouTubers include a personal (oral or written) trigger warning that the following content could cause distress to vulnerable people and/or that viewers should not try the challenge themselves. Taken together, about 29% of all videos present some form of warning to their viewers. Besides taking a video down or flagging it with a trigger warning, the provision of counseling contacts (e.g., lifelines, specialized websites) can also help to deal with potentially harmful media depictions. Nonetheless, this kind of support is rather unpopular within the community of YouTubers, as only about 5% of the Momo Challenge videos refer to further counseling.

Not surprisingly, live demonstrations tend to show the greatest extent of harmful depictions and therefore can be considered as the most problematic of all genres. Almost every live demonstration shows interactions with Momo accounts (96%) and about half of them also feature contact details of these accounts (50%). Importantly, not a single live demonstration offers counseling numbers for its viewers, which can only be stumbled upon in information (8%) and reference videos (25%). As a consequence, live demonstrations are most likely to be flagged as inappropriate by users (24%). The creators of live demonstrations seem to be aware of this development, causing about a quarter of them to give an (additional) oral/written trigger warning (28%). However, live demonstrations are not the only place for YouTubers to verbalize that Momo is an entity to be feared (50%); in fact, it is more likely done in art productions (79%). In art productions, the protagonists’ inner monologs are often presented to the viewer, giving direct insights into their emotions, i.e., fears.

If we combine these harmful and helpful depictions, we can see that Momo Challenge videos score an average of 1.14 points on the danger index (SD = 1.09, scale ranges from − 2 to 4, Table 3). As Fig. 2 shows, videos that exclusively show helpful depictions of the Momo Challenge (danger index of − 2 or − 1) are sparse (4.8%), while more than two thirds of the videos include at least some form of harmful depiction (danger index of 1 to 4). This result suggests that young people are at some risk when they navigate Momo Challenge content on YouTube. However, a quarter of these videos are safe (danger index of 0), some even helpful. From a policy perspective, it would therefore be unjustified to place publishers of self-harm or suicide game content under general suspicion.

Table 3 Average danger and engagement rates, by video genre (n = 104)
Full size table
Fig. 2
figure 2

Number of videos related to the Momo Challenge, sorted by danger level ranging from -2 (very unproblematic) to 4 (very problematic)

Full size image

Danger level and engagement rates (H1)

We assumed that more harmful depictions of the Momo Challenge will be positively related to engagement rates, even when controlling for other content characteristics such as genre (H1). From Table 1, we can already deduce that all engagement rates show a positive correlation with the danger level of a video. View count and danger index share a rather weak relationship (ρ = 0.232, p < 0.01), but comments (ρ = 0.373, p < 0.001), dislikes (ρ = 0.363, p < 0.001), and likes (ρ = 0.397, p < 0.001) reach a weak to moderate association. To test H1, we ran Ordinary Least Square (OLS) regressions that treated each engagement rate as separate dependent variable and the danger index as the independent variable. As a control, the video genre was introduced into the regression in the form of dummy variables. However, since we manually coded the danger index based on the video content, we have to ensure that our prediction is not limited by a high correlation between video genre and danger index, i.e., multicollinearity. The tolerance of all variables in the regression is greater than 0.10 and the variance inflation factor (VIF) values are well below 10.0 (see Table 4), so the collinearity diagnostics indicate no cause for concern (Pituch and Stevens 2016, p. 77; Tabachnick and Fidell 2014, p. 124). The results of all regressions are shown in Table 5.

Table 4 Multicollinearity analysis of independent variables (n = 104)
Full size table
Table 5 Regressions for danger predicting engagement (n = 101)
Full size table

With an adjusted R2 of 0.085, a statistically significant regression equation was found for the model comprising video views (F(4,101) = 3.317, p = 0.014). Similarly, the regression equation for video comments was statistically significant (F(4,101) = 5.260, p = 0.001) with an adjusted R2 of 0.146. A statistically significant regression equation was also found for likes (F(4,101) = 5.403, p = 0.001) with an adjusted R2 of 0.150, and for dislikes (F(4,101) = 4.604, p = 0.002) with an adjusted R2 of 0.126. For all of these models, the danger level proved to be a significant positive predictor with weak to moderate strength (p < 0.01 and p < 0.001), even when controlling for video genre. The only exception is the model comprising video views, where the danger index merely becomes significant at a 0.10 significance level. Although the danger index is significantly correlated with views in the bivariate analysis (p < 0.01, see Table 1), this association decreases when controlling for other video characteristics like genre. Why are views the only engagement rates that can no longer be predicted at a 0.05 significance level when introducing controls to the analysis? Our best guess is that the differences between the engagement rates can be explained by the fact that the decision to watch a video is made before the content of the video—and thus its danger level—is even known. Most likely, viewers are only to a limited extent able to anticipate the danger level of a video when they make the decision whether to watch it or not. As the danger level of the video becomes less important for the decision-making process, more manifest content characteristics, such as the video genre, become influential and, thus, disturb the relationship between danger index and views. In contrast, comments, likes, and dislikes can only be published after the video has been clicked and—at least in parts—viewed. Therefore, a verdict about the video’s danger level has already been reached when a comment is posted or the like/dislike button is pressed. From this perspective, it seems logical that the danger index is better suited as a predictor for comments, likes, and dislikes than for views.

Despite this exception for video views, a higher extent of harmful depictions of the Momo Challenge predicts higher engagement rates, thus confirming H1. As YouTube’s recommendation system rewards high engagement rates, there is an even greater likelihood that such harmful depictions will later be offered to new viewers. Consolidating this argument, Smith et al. (2018) were also able to find that about 60% of the YouTube users come across content promoting dangerous behavior through the recommendation system at least sometimes.

Momo Challenge network visualization and community identification (RQ2)

Up to this point, we analyzed the extent of harmful and helpful depictions of the Momo Challenge that can be encountered when watching a single video. However, this approach does not really capture the reality of YouTube use. In reality, users move from video to video. Regarding the Momo Challenge, community analysis helps to improve the understanding of what media diet and content preferences specific audiences expose themselves to. If these media diets show a great extent of harmful depictions, their respective communities should be avoided. We asked what communities have formed around the Momo Challenge videos and what content preferences they show (RQ2).

In Figs. 3 and 4, we visualized the Momo Challenge network with regard to shared commenters, respective communities, and danger levels. Each tie between two videos indicates content similarity and a greater likelihood for a video being recommended after having watched the other. The graph shows that the Momo Challenge network comes with seven communities, visualized by their corresponding color.

Fig. 3
figure 3

Visualization of the Momo Challenge on YouTube. Each node represents a video. Each edge represents a shared commenter between two videos. Node size represents the video’s number of views. Edge size represents the number of shared commenters. Color indicates community membership. Videos without a shared commenter were excluded from the visualization. Created in Gephi using Force Atlas 2

Full size image
Fig. 4
figure 4

Second visualization of the Momo Challenge on YouTube. Color indicates danger level from -2 (very unproblematic) to 4 (very problematic)

Full size image
  1. 1.

    The yellow community is the most dominant within the Momo Challenge network, which means that the lion’s share of video content about the Momo Challenge comes from this community. It predominantly consists of live demonstration content with high engagement rates (as indicated by the nodes’ size) and dense connections. As we have stated before, live demonstrations show a greater extent of harmful depictions. Thus, users who engage with videos from the yellow community increase the possibility of being (repeatedly) confronted with harmful media depictions.

  2. 2.

    In contrast, viewers moving within the green community expose themselves to more helpful depictions. This community is the second most dominant and consists primarily of information videos (and some educational art productions). They have a small audience and score low to medium on the danger index. Yet, the green community provides strikingly many boundary-spanning connections to other communities. On the one hand, this may indicate that users who watch Momo-related content also want to inform themselves further about the challenge and therefore seek out information videos. On the other hand, it may also be the case that users who hear about the Momo Challenge first through information videos are prone to checking out the original content.

  3. 3.

    The red community, which consists of video game material, is somewhat separated from the rest of the network. Some of these videos show how gamers try to establish contact with Momo accounts. A reference video of a gaming vlogger reacting to Momo videos connects this community to the rest of the network. Overall, the red community depicts Momo as a pop cultural character and not as a potentially harmful challenge. Nevertheless, adolescents could get curious about where the horror figurine originated from and search for more information.

  4. 4.

    The other four communities, orange, cyan, blue, and gray are small and not very dominant within the network:

    • The orange community consists of Indian short films that are often flagged with trigger warnings. Overall, they serve an educational purpose, but do include troublesome media depictions of self-inflicted injuries that should only be watched under supervision of an educator.

    • The cyan community is based on videos uploaded by a single YouTube channel, a gaming vlog. In the course of a month, the YouTuber had been playing a Momo horror game, then went on to present several Momo chatbots and apps, and finally uploaded an information video explaining that the Momo Challenge has caused the death of a child, thereby warning his viewers to stay away from the challenge. Generally, this video series is an accurate example of how many YouTubers handled the Momo Challenge, treating it as a funny dare at first and, after the release of media reports on teen deaths linked to the challenge, warning their viewers.

    • The blue community consists of the most problematic videos within the Momo Challenge network. They claim to show real WhatsApp chats with Momo, often split into several sequels, which include disturbing pictures of horror figures and self-harm. The titles of these videos indicate that they are 18 + content, an information voluntarily provided by the author. These videos were the first to be removed by YouTube when the social networking service took action against Momo-related content.

    • Finally, the gray community consists of isolates that are neither connected to the network nor receive a lot of views.

Obviously, the two most dominant communities on YouTube could hardly be more incongruent. While the yellow community prefers harmful depictions and live demonstrations of the Momo Challenge, the second most dominant community (green) is interested in fact-based, less engaging, and also more helpful depictions. This finding reinforces the impression that it would be unjustified to place publishers of self-harm or suicide game content under general suspicion, but that a more differentiated assessment is necessary before demonetizing or deleting content.

Discussion

As with many social media challenges before, the Momo Challenge was just a temporary phenomenon. However, suicide games still expose adolescents to inappropriate video content and self-harming ideation, scare youngsters that horror figures like Momo could turn up at night and hurt them, and keep parents worried. Copycats may spread telephone numbers and fake conversations to scare youngsters or steal private information. In a conirmed case, a 14-year-old Romanian girl even faked conversations with a Blue Whale curator while engaging in self-harming behavior. She wanted to show the conversations to her classmates to attract attention (Lupariello et al. 2019). If adolescents and parents are made to believe that suicide games are real, they have real consequences. Thus, Khasawneh et al. (2019) have argued “that it is urgent to monitor social media posts related to BWC [Blue Whale Challenge] and similar self-harm challenges (e.g., the Momo Challenge)” (p. 888).

Of course, the Momo Challenge is neither the first suicide game nor is it likely to be the last on YouTube (or similar short video apps), “where challenge videos have become a weird trope of doing the craziest thing to get views” (Mazhari 2018). This trend is problematic because many of these YouTubers appeal to a young audience that does not understand that these challenges are made for views and likes only (Mazhari 2018). In 2020, only two years after the peak of the Momo Challenge, another self-harm and another suicide game have already been launched: The Skull Breaker Challenge and the Jonathan Galindo Challenge. While the Skull Breaker Challenge can be considered the heir of the Tide Pod Challenge, the Jonathan Galindo Challenge is the direct successor of the Momo Challenge. Both, the Skull Breaker and the Tide Pod Challenge are self-harm challenges where dangerous stunts are posted to look cool or funny, in this case when two children kick the legs out from under a third to make him fall over, or when children eat toxic detergent pods. The Jonathan Galindo Challenge is a suicide game that mimics all the characteristics of the Momo Challenge, except that the name and profile pictures of the predatory accounts have been exchanged. Now the Jonathan Galindo accounts show an image of a man wearing a dog mask reminiscent of a twisted version of the Disney character Goofy.

Comparing the Momo Challenge with its predecessor, the Blue Whale Challenge, and its successor, the Jonathan Galindo Challenge, the Momo Challenge can be described as a turning point in how suicide games are organized. While the Blue Whale Game targeted introverts, who had already developed depression or suicidal tendencies (Lupariello et al. 2019; Mukhra et al. 2017), the Momo Challenge aimed at no homogenous audience. The participants of the Blue Whale Challenge did not aim at a homogenous were recruited and encouraged to perform self-harming tasks through social media messages from strangers (Lupariello et al. 2019; Mukhra et al. 2017). As confirmed cases show, some children even posted blue whales on their social media accounts to attract the attention of unknown curators (Lupariello et al. 2019). A confirmed case of an Indian boy also involved a download link to a mobile phone app that provided him with self-harming tasks and, thus, replaced the curator for future interactions (Balhara et al. 2018). As hospital reports show, curiosity and the desire to attract attention were the main drivers for young people to participate in the Blue Whale Challenge, in addition to prevalence of depression (Balhara et al. 2018; Lupariello et al. 2019). Building on these motivations, the Momo Challenge optimized the dissemination mechanism of suicide games. This new challenge used snowball effects to gain reach, i.e., the adolescents no longer came into contact with the challenge through strangers but through people they trusted, e.g., peers, close friends, or influencers on YouTube. Through this type of distribution, even adolescents who have never suffered from psychiatric disorders (e.g., depression) in the past are confronted with self-harming ideation. This mechanism also allows suicide games to spread virally, which makes them much more difficult to handle. After all, reactions from social networking sites, educators, and parents only occur with delay. For the same reason, the successor to the Momo Challenge, the Jonathan Galindo Challenge, was able to spread virally again, although YouTube had already taken action against its predecessor.

Implications for practice

If these challenges do not disappear and only tend to reappear under a new name and become viral again, how should we react in the future? When YouTube became aware of the problem of the Momo Challenge in March 2019, the online video-sharing platform decided to demonetarize all videos on the subject (Alexander 2019). About two years after the data collection, 41% of the videos in our sample were deleted and another 13% became access-restricted, which means that viewers must give their consent to watch the possibly disturbing following content before they can access the video. Live demonstrations were most affected. While 72% of all live demonstration videos have disappeared from the platform and another 16% got access-restricted, 70% of all information videos are still available without any restriction. However, these videos were also demonetized, even information videos from well-known news companies (Alexander 2019).

Overall, YouTube’s approach to restricting content is very much in line with our findings. Remember that in our content analysis (n = 105) we found that about 30% of the videos related to the Momo Challenge can be viewed without safety concerns, and 5% of them even offer exclusively helpful depictions of the challenge that are likely to trigger positive Papageno effects. By preferentially banning live demonstrations from the platform, YouTube has focused its efforts on content that often presents harmful depictions of the challenge, we welcome the fact that YouTube has taken action against the Momo Challenge. Only two points of criticism remain. The first one is that YouTube needed half a year before becoming aware of the challenge and taking action. In order to prevent future suicide games from becoming viral in the first place, faster action is required. Second, the decision to demonetize all videos related to the Momo Challenge seems disproportionate. If the content of suicide games is demonetarized as a matter of principle, then opportunities to educate young people about suicide games and show them ways they can seek help are missed, because the creation of such content is no longer financially viable. YouTube’s machine learning algorithms, which are currently taking over the task of automatically identifying and demonetarizing problematic content, are prone to ambiguous decisions and mistakes, a problem that has already been discussed by Kain (2017). For example, YouTubers have found creative loopholes to outsmart the machine learning algorithms in Jonathan Galindo Challenge videos. By paraphrasing suicide games, e.g., “the challenge for the whales who just so happen to be blue”, YouTube’s speech recognition cannot automatically demonetarize the videos or even ban the entire channel.

We do not have the ideal solution to this problem. However, videos could be checked for specific features that indicate more harmful depictions of a suicide game. In addition to depictions of violence or verbal cues like challenge names, it could be examined whether these videos show or read out telephone numbers, provide chat transcripts, or whether they show facial expressions that suggest fear. The source/creator of the content and the provision of counseling contacts could also be considered (though the latter should not be confused with telephone numbers of predatory accounts), so that information videos are not demonetized just because they feature the challenge. In the end, it is also in the interest of an online video-sharing platform like YouTube to keep as many videos online as possible and to monetize them. Similarly, parents should not panic when they discover that their children have been exposed to suicide game content. Instead, they should engage with their children and try to fathom what they saw and how it was presented. The knowledge that not every social media content about suicide games has to be harmful enables parents to have open and informed discussions with their children and provide appropriate support based on the specific content.

Implications for literature and theory

On YouTube, over 200 videos dealt with the Momo Challenge during the turn of 2018/2019. Typical for the platform, few of these videos received the lion’s share of the views and comments, thus being further promoted. As we have shown, the videos under study show depictions that are under suspicion of eliciting harmful effects. In fact, more harmful depictions of the suicide game are significantly correlated with the videos’ engagement rates. In addition, most of the videos demonstrate interactions with alleged Momo accounts and even pass on contact details. By not only providing contact data, but also showing how role models establish contact, these videos might dare teens to imitate their role models and test their own courage. It is also regrettable that only 29% of all videos under survey have implemented some kind of trigger warning and less than 5% refer to counseling contacts, two characteristics often theorized to be associated with helpful Papageno effects. Thus, viewers may not only be unprepared for what content will follow but are also not provided with references on where to get support when feeling distressed.

Besides these empirical findings, this paper also provides a theoretical contribution to the research on suicide games. The W.H.O. has already released guidelines on how to report on suicide both for non-fictitious media depictions (W.H.O. 2008) and for fictional media depictions (W.H.O. 2019). Similar guidelines for social media environments were released by SPRC (2014). By combining the literature on Werther and Papageno effects with these guidelines (RRS), we derive an analytical scheme to assess the extent to which social media portray harmful and helpful depictions of sensitive topics like suicidal ideation and suicide games. In this study, we used this scheme to analyze YouTube content, but it could also be used to analyze various social media content. To date, it still poses a challenge to identify content on social media dealing with self-harm and/or suicide (George 2019). This is also evident in the context of suicide games, where short, explorative reports have only just begun to assess the dissemination of related social media content (Khasawneh et al. 2019; Sumner et al. 2019). We underpin these analyses theoretically by drawing assumptions about the consequences of these depictions based on the literature on Werther and Papageno effects. We are aware that our scheme merely allows identifying content characteristics that are under suspicion to cause harmful or helpful effects; we have no proof that these effects will actually be evoked. Nevertheless, all of our assumptions rest on well-established theory and empirical evidence.

Limitations

In this study, we had to rely on the YouTube API. With few studies analyzing the inner workings of such APIs (e.g., Diakopoulos 2014; Driscoll and Walker 2014; González-Bailón et al. 2012; Rieder et al. 2018), little is known about the bias with which the YouTube algorithm produces search results. At the very least, Rieder et al. (2018) showed that the ordering of search results is quite stable over time. It is assumed that these natural search results are fairly well reflected by crawls with the API, with only small variations (Driscoll and Walker 2014; Rieder et al. 2018). These variations will most likely manifest themselves in a skew toward central, densely connected network hubs for search results, while data crawls will include peripheral users more accurately (González-Bailón et al. 2012). In the light of these uncertainties, we decided to complement our crawl data with data from a manual search. This proved to be not only worthwhile, but necessary, as we were able to identify 98 videos that the crawl missed, which is a much greater deviation than anticipated. Through this two-step procedure, we created a data set that reflects the video selection viewed by users on YouTube at the time of data collection (turn of 2018/2019) as adequately as possible. During the course of our analysis, numerous videos have been provided with trigger warnings, deleted, or newly uploaded. As a result, this analysis represents a snapshot of how the network was laid out at the turn of 2018/2019.

Another limitation of our study is that while we can demonstrate the effect of a video’s danger level on its engagement rates, we cannot provide any insight into how these effects come about. Adolescents may process the depiction of harmful content in very different ways. While some may express their thoughts in comments, others may dislike the video to express themselves. And a third group might do nothing or even like the video. Our study cannot shed light on the emotional and cognitive processes that lead to active engagement with content, nor can it explain why adolescents turn to a particular form of engagement. As is so often the case in exploratory research, this study merely contributes to the literature by providing evidence of an association between harmful depictions of a suicide game and engagement rates. Future research might address how and when this association occurs, leading to a deeper understanding of the mechanisms by which the effect operates.

As a final limitation, it should be noted that we cannot provide definitive proof that the audience of Momo Challenge videos is young. However, our proxy measure and a look at the platform’s overall demographic structure and media usage trends of teenagers (e.g., Anderson and Jiang 2018) has given us reason to believe that we are indeed investigating a young audience.

Future prospects

The present study relies on basic media effects approaches (Werther effect and Papageno effect). Actual effect sizes are hard to measure in this context since they are usually traced by intra-extra media analyses. Therefore, future research could focus on surveys, experiments, or qualitative interviews asking for how participants would assess the impact of harmful media content and how they feel when engaging with it. What consequences does (prolonged) exposure have? What coping strategies do young viewers develop? It is important that ethical aspects are taken into account when conducting these studies and that the approval of an ethics committee is obtained in advance.

Future studies might also address the responsibility of social media influencers. What is their self-image and their perception of professional responsibility? How can they be effectively incentivized for not further promoting and popularizing highly problematic media depictions of sensitive topics? Researchers might also take on the challenge to seek out alternative solutions, such as creating awareness material and guidelines for the content creators of online video-sharing platforms on how to produce more favorable media depictions of sensitive content.