Body image issues prevalent during pre- and early adolescence negatively impact the mental and physical health of young people [1-4]. For example, prospective studies have established that body dissatisfaction during later childhood and early adolescence predicts dieting behaviors and reduced physical activity [5], depression [6], poor self-esteem [7], and eating disorders [8]. Although most body image interventions aimed at young people focus on adolescents [9-11], up to half of children aged 6 to 12 years are unhappy with their appearance [2], suggesting that more interventions aimed at younger cohorts to prevent or reduce poor body image and promote positive body image are vital [5,11]. In addition, the prospective evaluation by Lacroix et al [12] of body esteem development in girls and boys aged 11 to 15 years found that, among those with low body esteem, dissatisfaction was entrenched by the age of 11 years and remained stable throughout adolescence, which further supports the need for earlier intervention before poor body image becomes established.

Given the widespread nature of body image issues among young people, traditional approaches tend to be used, namely, interventions based in schools [10], the community, or clinical environments [13]. However, substantial barriers to the dissemination and acceptability of face-to-face interventions exist, including the global lack of human resources required to deliver interventions [14], the costs involved [15], and the stigma attached to mental health issues [15,16]. As a result, other approaches need to be explored to reach young people at scale. In particular, a focus on prevention or low-level concerns (eg, mild to moderate discontent related to one’s appearance that does not impact day-to-day living, such as opting out of activities or engaging in body-changing strategies) requires fewer resources and could help reduce the strain on the mental health care system through intervention in digital or community settings [17,18].

Much evidence has shown that interventions targeting universal, nonclinical samples of young people have been successful in reducing body image concerns [19-22]. One such approach is through the use of microinterventions. Microinterventions offer a lighter touch in comparison to traditional interventions. They are designed to be self-guided and of short duration to elicit immediate improvements in the targeted symptoms [23]. Furthermore, microinterventions may be delivered only once [24,25] or repeatedly [23,26]. Given the brevity and self-guided nature of microinterventions, they are well suited to digital environments [27], such as via videos [23,28], social media [29], and online games [30]. To date, we are aware of only 2 rigorously tested body image microinterventions aimed at children—a 60-second psychoeducational cartoon and a 120-second playable (ie, in-app) interactive game for adolescents aged 13 and 14 years—both of which demonstrated improvements in state body satisfaction [30,31]. Similarly, a body image chatbot for adolescents aged 13 to 18 years containing various microinterventions proved effective at improving both state and trait body image [32].

Given the encouraging results from Matheson et al [30] and the wide scope and popularity of video games among young people [33], the online gaming world holds great promise for housing body image microinterventions. Researchers have been tapping into the use of online games to deliver mental health interventions for several years [34-36]. However, other than the playable intervention by Matheson et al [30], to our knowledge, no other online game designed to improve body image has attempted to enter the gaming landscape. As such, this paper focuses on Super U Story, an innovative stand-alone video game microintervention embedded within the Roblox gaming platform intended to reduce body dissatisfaction and promote positive body image among girls and boys aged 9 to 13 years. The Roblox gaming platform houses a huge array of online games aimed at young people, most of which are free to play, created by both gaming developers and users alike for the purposes of entertainment [37] (ie, it is generally not used for educational or intervention content). Although no academic literature exists regarding the impact of Roblox games on body image and related outcomes, this platform was chosen because of its potential to reach millions of children; as of 2024, Roblox had >216 million active monthly users worldwide, with almost an equal number of female and male players, and 42% were aged <13 years [38]. As is typical of Roblox games, Super U Story combines both mandatory and nonmandatory content to improve body image in its players embedded in a dramatic narrative, a key element in video games promoting health-related behavior change [39].

Super U Story is grounded in the tripartite influence model of body dissatisfaction [40] and the tenets of positive body image [41]. The Tripartite influence model addresses well-established risk factors for negative body image, namely, internalization of appearance ideals (ie, personal belief in society’s standards of beauty and taking action to try to achieve the ideal appearance) [40] and appearance-based social comparisons (ie, making appearance-based comparisons with others) via the sociocultural influences of the media, peers, and family. Super U Story incorporates 2 of these influences—media and peers—with a focus on media, which encompasses both traditional and social media [42,43] and is considered to be the most pervasive [44]. These risk factors are targeted through psychoeducation that shares strategies to manage appearance-related teasing and bullying, elucidates the harms of engaging in appearance-based comparisons, and promotes social media literacy (ie, skills related to critical thinking and skepticism regarding social media content and the motivations for creating such content) [45,46]. Given social media’s focus on appearance and its prolific content and messaging regarding achieving an “ideal” body [47], targeting these risk factors was especially pertinent. Furthermore, evidence exists suggesting that facilitating social media literacy skills in young people plays a protective role against body dissatisfaction [45,46]. Although body image interventions have historically focused on reducing established risk factors [10], there is a growing trend to address and reinforce body image–related strengths and protective factors [48] in interventions. Specifically, tapping into positive body image—a construct involving various facets independent of negative body image, namely, an appreciation of and respect for one’s body and its functions regardless of whether it conforms with society’s beauty standards—shows promise [49]. Positive body image has been associated with psychological well-being (eg, positive affect, high self-esteem, and optimism) as well as healthy behaviors such as increased engagement with exercise, positive self-care behaviors (eg, protecting the skin from sun damage and engaging in regular oral health habits), and intuitive eating [49]. For this reason, Super U Story also applies tenets of positive body image by focusing on bolstering protective factors known to promote positive body image, including body functionality (ie, valuing one’s body for what it can do instead of for how it looks), body appreciation (ie, respecting and caring for the body), and an appreciation of diverse body shapes and features [50,51]. Targeting protective factors during childhood could reduce the likelihood of body dissatisfaction developing during early adolescence [4,52]. Similar to the approach taken with risk factors, psychoeducation is used in Super U Story to promote the 3 aforementioned protective factors.

This 3-arm, online randomized controlled trial (RCT) evaluated the effectiveness of Super U Story in eliciting immediate and short-term improvements in children’s and adolescents’ state- and trait-based body image and related outcomes.

This study aimed to test four hypotheses:

This study was a 3-arm, online, parallel RCT conducted in 2023 in the United States to evaluate the effectiveness of Super U Story using an intervention group, an active control group, and an attention control group. Participants were randomized to a group using the least filled quota to ensure an even distribution of age, gender, socioeconomic status, and ethnicity across the 3 groups. The feasibility of the study was first tested in a pilot study.

This study received ethics approval from the Health and Applied Sciences Research Ethics Committee at the University of the West of England, Bristol (HAS.22.11.040) and was preregistered with ClinicalTrials.gov (NCT05669053). Before recruitment, informed written consent was obtained from parents or guardians. Consent forms outlined who was conducting the study, the aim of the research and what it entailed, and the benefits and possible risks. It also ensured the highest level of confidentiality regarding the information shared by the participants, explaining that a unique participation code was assigned to each participant so that participants could not be identified. The complete anonymity of the participants was guaranteed regarding any reports or publications on this study’s results. The voluntary nature of their children’s participation was underscored, and the participant, parent, or guardian had the right to withdraw from the study at any time without providing a reason. Finally, parents or guardians could request the removal of their children’s data from the study up until the point of data analysis. With regard to compensation, participants were sent incentives totaling US $40 (US $10 after the first survey, US $15 after the second survey, and US $15 after the final survey). They were also entered into a sweepstake (free prize draw) to win 1 of 3 US $1000 prizes in the form Visa gift card claim codes.

The participants were recruited and enrolled via a US-based research agency. A sample of computer- and internet-literate girls and boys from diverse ethnic, socioeconomic, and geographic backgrounds (ie, from all 50 states) were recruited. Eligible participants were US-resident girls and boys aged 9 to 13 years who played Roblox games a minimum of 4 hours per week. They were excluded from the study if they had previously played the Super U Story game or if their parent or guardian did not provide written consent.

The research agency recruited participants online using their existing databases of adult participants and sample partners to ensure national representation. Sample partners are companies who own and manage panels with respondents who opt in to take surveys; these companies have been fully vetted as trusted and reliable sample suppliers by the agency’s field management team. In all cases, the sample was opt in, meaning that respondents had opted in to receive surveys from the research agency. The young people recruited into this trial were originally sourced through their parents or guardians. The research agency sent those parents or guardians who had at least one child aged between 9 and 13 years a link to a study information sheet and screening questionnaire. If a household had multiple children who met the inclusion criteria, the agency selected 1 child based on the age and gender most needed to meet age and gender quotas. Parents or guardians supplied demographic information on behalf of their children (ie, socioeconomic status, ethnicity, age, and gender) and completed the screening questionnaire to determine whether they met the inclusion criteria. Following the screening questionnaire, parents or guardians provided written consent. In addition, at the beginning of each of the study’s questionnaires, participants had the opportunity to read the study information (including their right to withdraw without penalty) and provide written assent. Participants were instructed to complete the study questionnaires on the device they usually played games on (ie, mobile phone, laptop or desktop computer, or tablet).

The research agency took several measures to ensure that the participants were quasi-anonymous and could not assume multiple identities. Their sample partners assigned each respondent a unique and anonymous ID number to safeguard against a respondent being screened more than once, and they used MyCleanID (RealDefense LLC) [55] and digital fingerprinting technology (ie, FingerprintJS) [56] to identify respondents’ devices and track survey interactions. Furthermore, each respondent was assigned a unique questionnaire link, which became inactive after it had been completed once to prevent duplicate submissions.

Super U Story was created through an industry-academic partnership between Toya, a gaming studio that creates video games for the online gaming platform Roblox, and the Dove Self-Esteem Project, the social mission for Unilever’s personal care brand, Dove, in consultation with the authors of this paper (ie, body image researchers). Toya owns the intervention. The intervention (version 1; March 4, 2023) is a narrative-based adventure video game designed for the Roblox platform aimed at children aged 9 to 13 years. Minor changes to character dialogue in 4 short scenes with nonplayer characters were implemented after the trial commenced (version 2; March 17, 2023). Playing Super U Story from start to finish takes approximately 20 minutes. The storyline was initially developed by Toya, which centers on The Academy, a school for kids with developing superpowers that comes under attack by a group of rogue ex-students intent on spreading negativity. As players progress through the game, navigating obstacles, learning about the unfolding narrative, and strengthening their newly acquired superpower to save The Academy, they are exposed to psychoeducational content through pop-up messages housed in a fictional social media platform named Flutter, nonplayer characters’ dialogue and conversations, interactive conversations between the players and nonplayer characters, and dialogue from The Academy’s trainers (Figure 1). Players reaching the end of the game are presented with 3 possible endings, 2 of which involve defeating the threat to the school. All 3 possible conclusions reinforce the psychological messages embedded throughout the game’s journey.

The development of Super U Story was guided by best practice for games that promote health outcomes, which recommends collaboration between game designers and academics to ensure that the game is grounded in evidence and theory [57]. As such, authors NP, HW, SH, and PCD provided Toya with recommendations and rationale related to theoretical content, such as the wording of key messages to target risk factors associated with the Tripartite influence model [40] and the protective factors connected to positive body image [41]. This included the development of the game’s characters to ensure a wide assortment of body types to encourage an appreciation of various kinds of bodies, considering body size, ethnicity, hair type, ability, facial and bodily features, and appearance-affecting conditions such as vitiligo. Drawing on their expertise, the game developers at Toya identified restrictions regarding how and where educational messages could be incorporated to maximize the length of time that players engaged with the game. A primary concern of Toya, and of developers more broadly, is to avoid their games appearing to be “edutainment,” thinly disguised educational software or “chocolate-covered broccoli,” an approach proven to be ineffective with children [58,59]. Toya’s valid concern about losing players who recognize and object to this technique needed to be balanced with embedding evidence-based psychoeducation in an activity that young people choose to engage in for purely entertainment purposes. As a result, most of the intervention’s content was embedded in gameplay options that were nonmandatory for players, a key feature of Roblox games that gives players the freedom to pick and choose the elements with which they wish to interact (Multimedia Appendix 1).

Toya engaged in their typical approach to user testing throughout the game development process, which involved regular testing with their user testing community via private servers on the Discord platform [60]. Authors NP, HW, and SH were invited to view 3 recorded sessions conducted on June 25, 2022; July 2, 2022; and July 23, 2022, for which the authors provided a set of questions for the sessions’ facilitator to ask the players regarding comprehension of the game’s narrative and key psychoeducational messages. Participants consisted of 33 regular Roblox players in the United States ranging in age from 7 to 17 years. The privacy of Discord’s user testing community is protected; as such, further demographic information for these players was not made available. In addition, the authors shared a short survey on the survey platform Qualtrics (Qualtrics International Inc) to assess Roblox players’ comprehension and acceptability of Super U Story with 4 young people in the United States ranging in age from 9 to 13 years (mean age 10.5, SD 1.91 years) recruited through convenience sampling (n=3, 75% boys; n=1, 25% preferred not to say). Following these 3 user testing sessions and receipt of the survey results, the game developers at Toya and authors NP, HW, and SH worked together to edit the game’s narration and dialogue to maximize players’ full comprehension of the storyline and psychoeducational messages. Multimedia Appendix 1 provides an intervention summary.

To isolate the key components of the intervention, Super U Story was compared to 2 control conditions: an alternative Roblox game (active control) and a web-based word search (attention control), described in the following sections.

The active control condition was assigned to play the Roblox game Rainbow Friends 2 Story (Color Story), a similar Roblox game that was not intended to target the outcomes of interest (ie, body image and related outcomes, the active ingredients of the intervention [61,62]). It should be noted that, following the trial, the name of this game was updated to “Colors (Story),” but the narrative and key game elements remained the same. As recommended when choosing an active control [63], other key features of Rainbow Friends 2 Story (Color Story) were evaluated to be a close match with those of Super U Story—both games are based on a similar narrative (ie, saving the world or The Academy), have the same approximate duration of gameplay, present challenges and obstacles to tackle, and have various interactive nonmandatory elements with which to engage.

The attention control condition was assigned an age-appropriate online word search where participants were required to find words related to animals [64] in as many word searches as they wished to engage with for up to 30 minutes, the maximum amount of time that participants in the intervention and active control conditions were instructed to play. The purpose of the attention control condition was to ensure that participants were provided with an activity of a similar duration as that of the intervention to occupy their time and attention but without any other components featured in the intervention [62,65]. In addition, given the physicality of gameplay in Roblox games (ie, avatars controlled by the player are in almost constant motion, engaging in movement such as running, jumping, and swimming, particularly when navigating obstacles), it is possible that this aspect could impact body functionality. As such, we included an attention control condition featuring a non-Roblox game to account for this potential effect.

The researchers had no contact with the participants during any phase of the trial to minimize the risk of bias. All communication occurred between the research agency and the participants. Participants completed online self-report questionnaires hosted on a secure platform, UNICOM Intelligence (version 7.5.1; UNICOM Systems, Inc), at 3 time points: baseline (days 1-2), intervention testing phase (days 8-9), and 1 week after the intervention (days 15-16). Questionnaire functionality underwent user testing by authors NP, SH, and HW and research agency staff before the trial commenced.

At 8 AM CST on days 1 to 2 (time 1, baseline [T1]), days 8 to 9 (time 2, intervention testing phase [T2]), and days 15 to 16 (time 3 at 1 week after the intervention [T3]), the research agency sent participants a link to the corresponding questionnaire. Questionnaires delivered at T1 and T3 were composed of trait measures of body esteem, body appreciation, internalization of appearance ideals, and social media literacy. The T2 questionnaire was composed of state measures of body satisfaction, body functionality, and mood. The participants had 36 hours to complete each questionnaire and were instructed to do so in one sitting. The research agency assigned each participant their own unique participant identification number to match participant responses over time.

Following completion of the baseline questionnaire, participants were randomly allocated via the UNICOM Intelligence program using least filled quota to the intervention, active control, or attention control group, which was executed by the research agency. Concealing participants from their assigned group was not possible given the nature of the intervention; however, the agency was concealed from the participants’ randomized arm as the allocation was automated. One week later, on day 8 at 8 AM, participants were sent questionnaire 2 along with a link to play one of three games: (1) Super U Story (intervention group), (2) Rainbow Friends 2 Story (Color Story; active control group), or (3) an online word search (attention control group). The state measures were completed immediately before and after gameplay (ie, T2a and T2b). When participants completed the first set of state measures, they were instructed to play the game that corresponded to their allocated condition for a minimum of 5 minutes and no longer than 30 minutes based on a summary of preliminary gameplay analytics provided by Toya before executing the pilot (ie, raw numbers were not provided by Toya). Participants were instructed to complete the state measures and play the game in one sitting. Acceptability and game engagement information was also collected from the intervention group participants at the end of questionnaire 2. The final questionnaire (questionnaire 3) was sent to participants 1 week later, on day 15. In total, 3 attention checks were embedded within the T1 and T3 questionnaires to assess data quality, and 3 manipulation checks were presented at the end of the T2 survey to assess intervention and active control participants’ level of attention during gameplay. Participants were given 36 hours to complete each questionnaire. The research agency sent reminder messages to those who had not engaged with each questionnaire (after 8 hours for questionnaire 1 and after 8 and 26 hours for questionnaires 2 and 3, respectively). Multimedia Appendices 2-4 provide the questionnaires at each time point. Following the completion of questionnaire 3, the research agency sent participants a debriefing document disclosing the study’s aims and objectives and contacts for free mental health resources. The procedure outlined previously was first trialed in a pilot with 136 participants (see the Pilot Study subheading in the Results section).

For this 3-arm study, based on an a priori sample size calculation, we aimed to recruit 322 participants per arm with complete data (N=966 in total) to have 80% power to detect a small standardized effect (Cohen δ=0.2) between the intervention and each control group using analysis of covariance (ANCOVA) conservatively assuming a correlation between baseline and outcome of at least 0.6. To account for dropout (similar studies testing microinterventions online show dropout rates of 34% at posttest [30]), phantom applications (ie, bots), and those who expressed an interest but did not fully engage (ie, manipulation check failures), the sample size was inflated by 35% to 493 per group (N=1479) to ensure sufficient power at posttest.

Analyses were conducted using SPSS (version 29.0; IBM Corp). Condition allocation was concealed from the data analyst throughout data preparation and hypothesis testing to avoid interpretation bias.

This real-world remote-based online gaming intervention expectedly had some technical challenges, which could not be resolved in real time despite best efforts. For these reasons, the analyses were conducted per protocol (PP) rather than being intention-to-treat (ITT) analyses (ie, eligible consenting randomized participants who did not exercise the right to withdraw and had no known protocol deviations were included in the analysis). The corresponding ITT analyses are provided as supplementary materials and are commented on in the Discussion section.

For hypothesis 1 (state-based measures), the primary analysis of state body satisfaction after the intervention (T2b) used ANCOVA with state body satisfaction before the intervention (T2a) as the covariate and randomized arm as the independent variable. Underpinning model assumptions, including the homogeneity of regression line assumption, were assessed, and appropriate simplification was undertaken if the parallel line assumption was justifiable. The η_p² summarized effect size in the ANCOVA model. A preplanned repeated-measure ANCOVA was used to compare the intervention group against the active control group at T2b, and the same preplanned ANCOVA model was used to compare the intervention group against the attention control group. The same analysis plan was used for state body functionality and mood.

For hypothesis 2, the same ANCOVA strategy used for analyzing the state measures was used for the trait measures. Specifically, for each trait measure (ie, body esteem, body appreciation, internalization of appearance ideals, and social media literacy), ANCOVA was used for between-group comparisons at T3 after controlling for the commensurate baseline measure at T1.

Hypothesis 3 (moderation of effects by age group and gender) was similarly considered by extending the ANCOVA models. Specifically for gender, the ANCOVA model was extended to include a main effect for gender and a gender by group interaction term. Age was dichotomized (≤11 years and ≥12 years) and included in the ANCOVA modeling as a main effect for age and the age by group interaction term. Moderation was assessed using the interaction terms.

For hypothesis 4 (engagement), we planned to use regression analysis and consider the degree of correlation between measures of engagement (ie, length of time of gameplay and number of Flutter messages recalled) in the intervention arm and for each outcome measure (T2b and T3) after controlling for the commensurate measure at T1 or T2a.

This 3-arm, fully-powered RCT evaluated the effectiveness of a purpose-built body image Roblox game, Super U Story, on children’s and adolescents’ state and trait body image and related outcomes. The findings partly support hypothesis 1—young people who played Super U Story experienced significant improvements in state body satisfaction immediately after playing Super U Story in comparison to participants in the active control group (Rainbow Friends 2 Story [Color Story]) but not when compared to the attention control group (word search). This effect was very small, in line with previous body image research using universal samples [20]. Unexpectedly, young people who played Super U Story did not experience improved state mood or state body functionality immediately after gameplay in comparison to the control groups. Moreover, contrary to hypothesis 2, intervention participants did not report improved trait body esteem, body appreciation, social media literacy, or reduced internalization of appearance ideals in comparison to the control groups at the 1-week follow-up. Body esteem and body appreciation were significantly lower among those who played Super U Story than among those who played Rainbow Friends 2 Story (Color Story; active control) at 1 week after gameplay. These effects were weak. Regarding hypothesis 3, the exploratory analyses found that effects were not moderated by age group or gender. These null findings are inconsistent with our hypothesis. Furthermore, we were unable to explore dose effects given the poor quality of the engagement data and, thus, were unable to test hypothesis 4.

Importantly, there was no evidence that Super U Story causes harm, meaning that, although there were no robust positive effects, we are confident that Super U Story does not negatively impact body image and related factors for girls and boys. On the basis of these findings, we conclude that Super U Story is ineffective in eliciting improvements in participants’ body image and related measures in comparison to both control conditions.

While the findings of this study were unexpected, they mirror unexpected findings in other body image intervention studies more broadly. Indeed, several micro- and standard intervention studies have also reported improvements in body image–related variables among the active control group [30,83,84]. Positive effects among the active control group have previously been attributed to demand or practice effects [84] or positive self-regard for participating in research [85], which may also be the case in this study. Relatedly, several studies have reported null findings for body image interventions [86-88], with a third of universal body image interventions found to be ineffective [89]. In their meta-analysis of stand-alone body image interventions, Alleva et al [13] noted that a considerable proportion of interventions that found negative or null effects for body image were not published or submitted for publication, limiting what can be learned from previous trials.

With these points in mind, we propose several possible explanations for why Super U Story did not have the expected impact on our key outcomes. First, the use of psychoeducation may have contributed to the nonsignificant findings. Key messaging was integrated into Super U Story (eg, via Flutter, the game’s social media platform) with information (eg, “Beware, some images on Flutter have been altered using editing tools. Don’t feel pressure to edit images of yourself, show the real you”) presented in a unidirectional way without opportunity for participants to engage in critical evaluation or practice new skills. A recent systematic review of digital interventions found that interactive interventions (ie, those that include additional building activities, such as written, listening, or reading activities or worksheets; assignments; and discussions) were most effective in improving body image outcomes among adolescents and young women [48]. In addition, another systematic review and meta-analysis looking at body image and media literacy interventions for young people found that interventions that induced and provided opportunities to resolve cognitive dissonance in participants were the most effective [90]. Research has shown that psychoeducational messaging is a weaker intervention strategy for changing behavior and attitudes (ie, body image) than other deeper learning activities in which participants have the opportunity to apply their knowledge. For instance, in a recent review, Guest et al [11] found limited evidence for psychoeducation as an intervention strategy for improving positive body image among children, with only 3 out of 9 psychoeducation-based interventions finding improvements across measures of body image. However, given the scope and nature of the Super U Story intervention (ie, brief, light touch, and fast-paced, with restrictions based on the platform used and software available), psychoeducation was chosen as an intervention strategy due to the ease of implementation in comparison to other strategies [11]. Furthermore, research has shown that this can be a useful approach in microinterventions; indeed, Matheson et al [30] found that mandatory psychoeducational messages delivered via microinterventions in both static and interactive formats were equally effective in immediately improving young people’s body image and mood in the short term.

Second, the intervention format may have contributed to the unexpected findings. Exposure to the key messaging via Super U Story was mostly optional given the typical conventions of Roblox games and the expectations of their users. The engagement data collected in this trial, albeit self-reported, provided some insight into how participants interacted with the game. The data suggest that participants experienced highly variable and often low-dose exposure. Notably, less than a third of the responses from participants about their learnings (105/361, 29.1%) related to the intervention’s key messaging. Previous successful microinterventions have included mandatory exposure to key messaging and activities [30]. In addition, player expectations and multiple distractions within the game may have contributed to the mixed findings. Players expect entertainment and escape from playing Roblox games. Overcoming obstacles (“obbies”) is central to that experience, but Super U Story’s obbies did not deliver key messaging and were likely a competing distraction. The Super U game developers were concerned about the “chocolate-covered broccoli” phenomenon (ie, losing players who object to thinly disguised educational messages [58,59]); as such, when writing the script for Super U Story, they were keen to avoid this risk by not enforcing mandatory exposure to the intervention’s psychoeducation messaging. However, because of this approach, the intervention’s messaging was likely to be insufficiently explicit to confer benefits, especially given the low levels of exposure.

Third, developmental factors in relation to processing educational content may have played a role. In this trial, the characters used in Super U Story were markedly different to the standard, blocklike characters typically featured in Roblox games, which was evident in some of the qualitative feedback we received. The storyline was also new to participants. According to the capacity model, how children process educational content is affected by their relatability to and familiarity with a narrative [91]. When narrative and educational content is less relatable and familiar, a higher cognitive load is required to process it. Participants in the Super U Story condition had no familiarity with the format of the characters and the storyline (inclusion criteria stated that participants must never have played Super U Story), which may have lessened the salience of the psychoeducational messages, consequently reducing the impact of the intervention. Conversely, the positive effects experienced by the active control group may be, in part, related to the familiarity of playing a game with the typical-looking Roblox characters. Going forward, given evidence that more familiar content reduces cognitive load [91], it would be useful to look at the impact of multiple engagements with Super U Story (ie, when participants are more familiar with the game). Moreover, microintervention studies, although with adult samples, have shown that multi-session and multi-activity approaches are effective at improving body image [23], further highlighting the importance of looking at the impact of repetitive gameplay.

Fourth, while the Super U Story acceptability scores were good, they have been typically higher for other similar interventions. For example, acceptability scores for the Brazilian body image chatbot Topity [32] and the Indonesian body image social media video series Warna-Warni Waktu [42] were generally of ≥85% (in comparison to the average acceptability score of 75.27%, [SD 4.83] for Super U Story). Coproduction (beyond user testing) and greater input from the target population (ie, those aged 9-13 years) at the earliest stages of development may have increased acceptability scores and potentially improved engagement levels. Indeed, evidence indicates that coproduction has been associated with improved engagement and adherence in other eHealth interventions [92,93].

This trial has several limitations. It was exclusively online (ie, participants completed all elements of the trial at home or in another non–laboratory-based setting), which was ecologically valid but, ultimately, limited the amount of control we had over the trial (eg, to troubleshoot problems), potentially impacting the quality of the data collected. Indeed, some participants indicated that they experienced glitches or struggled to return to the survey after playing the game, highlighting the challenges of collecting this type of data in a fully remote trial. We ensured that there was a “support” button embedded in the survey, which participants could access to receive help from a member of the research agency, but few chose to use this resource. A key reason for conducting the trial remotely was to enhance ecological validity (ie, participants were able to play the game in a familiar setting using their own device). The web-based design was also cost-effective and facilitated a more diverse geographic sample. However, it should also be noted that, in the review by Mahon and Seekis [48] of digital body image interventions, they found that tightly controlled, laboratory-based, and researcher-led studies tend to report significant results, whereas at-home settings result in more variable findings.

A further limitation was that we relied heavily on self-report estimates to measure engagement. In terms of capturing participants’ engagement with the game’s key features, we were unable to record this in a more objective way (ie, to track individual’s actual gameplay). Toya, the game developers, were able to collect engagement data on aggregate but were unable to do so for each individual player due to data protection regulations related to minors. Therefore, the quality of engagement data was unreliable (ie, participants relied on recall) and lacked sufficient detail. As a result, we were unable to conduct exploratory analyses to examine the impact of dose on body image and related outcomes. Future gaming trials would benefit from efficiently capturing engagement levels by tracking participants’ real-life gameplay to determine whether participants interact with any nonmandatory features (ie, receive the key messaging) [94] and calculate dose response. This is particularly noteworthy given that research suggests that self-reported adherence to digital interventions is inflated in comparison to objective adherence [94]. Crucially, obtaining fully objective adherence data (ie, not self-reported) could provide insights into whether engagement with key messaging was insufficient or the intervention was not potent enough to generate the desired effects [94]. Relatedly, in terms of metrics and engagement in the gaming world, the level of engagement evidenced in this study (albeit self-reported) was considered good. However, from an intervention perspective, it was very low, calling into question the fit of this game in improving children’s and adolescents’ body image.

An additional limitation was that baseline means across several of our variables were high, including social media literacy, mood, and body functionality, suggesting that many of our participants already had “good” levels before intervention exposure, consequently allowing for little room for improvement. It is possible that issues, for example, with mood and body functionality are less salient in this age group [12]. As such, future research with children could further explore these variables among younger and older cohorts to understand the optimal age for delivering such interventions.

Finally, this study used a PP analysis instead of an ITT analysis. While ITT is typically used for intervention trials, we opted for a PP approach due to concerns about the quality of the ITT data. Issues such as noncompliance, attention check failures, technical problems, and questionable data points (as detailed previously) led us to believe that the ITT analysis dataset contained data that were difficult to reconcile, whereas the PP analysis data had fewer unresolved queries. Therefore, we made the decision to report on the PP analysis dataset as clearly stated previously. While PP analyses are sometimes criticized for potentially showing more positive outcomes, our results from the PP analysis did not support the intervention’s efficacy. In fact, we concluded that the Super U Story intervention was not effective. For transparency, we have included the ITT results in Multimedia Appendix 7, and despite some differences, the overall conclusion that the intervention was not effective remains unchanged.

There are several strengths associated with this trial. It was adequately powered to detect small effects, and we included both an active control (alternative Roblox game) and an attention control (word search or non-Roblox game activity) to provide a fuller picture of the impact of Roblox games on body image and related factors. Using both types of control groups in a trial can help researchers understand and interpret the findings. Indeed, this study was developed at cost to have 2 control arms to permit a comprehensive assessment of efficacy. The prospective study has a preregistered primary research question, with a preregistered primary end point using a validated measure with good test-retest reliability. Sample size was determined a priori based both on power considerations and dropout rates informed by the literature; the target sample size was achieved. Analyses were conducted using well-established statistical techniques with due regard to statistical model assumptions to help ensure rigor, rely on statistical conclusions, and limit the impact of type I errors.

This study was also innovative in that we embedded a body image microintervention into a hugely popular gaming platform with a massive reach. Relatedly, we had the opportunity to inform experienced game developers as they created the body image game for Roblox. We were able to draw on their expert insight into game design and gameplay and understand what was feasible in terms of game development given the software available at the time. However, working with the developers involved compromise. Our priority as researchers was to maximize participant exposure to core messaging. In contrast, the game developers were opposed to including too much mandatory psychoeducational content for fear of losing players as, understandably, their business is to retain players until the end of the game. Finally, a notable strength of this study is the importance of sharing our learnings with the scientific community and not contributing to the file drawer problem or publication bias, in which studies without significant effects are not published [95]. Specifically, our detailed reporting of this study’s null effects contributes to improving the reliability of the wider evidence base [96] of intervention studies aiming to improve body image.

This is the first large-scale RCT assessing a body image intervention, Super U Story, embedded within a popular gaming platform, Roblox. Overall, playing Super U Story did not cause harm; however, there is a lack of evidence to suggest that Super U Story improves body image or related factors. Despite the lack of effects, this trial demonstrated that, with stakeholder collaboration, there is scope to explore delivering microinterventions to young people via large commercial gaming platforms to help address the need to reach young people at scale.