Social media has become a cornerstone of adolescent daily life, with 96% of US teens aged 13‐17 years reporting daily internet use and 46% engaging with platforms “almost constantly” [1]. These platforms (eg, TikTok [ByteDance Ltd], Instagram [Meta Platforms], Facebook [Meta Platforms], and X [X Corp]) offer adolescents unprecedented opportunities to interact with peers [2], explore their identities [3], and access global culture [4]. Although this study does cite from recent years (2023), we do recognize that social media data and information are constantly being updated, which has brought about a need for a study such as this to examine the type of content that adolescents post online.

Uncovering what specific content adolescents choose to post online is crucial, as it reflects lived experiences, developmental priorities, and cultural contexts [5,6]. These methods capture displayed content, which includes content analysis and ethnography. Within content analysis, there is objective coding but with limits to what meets our criteria [7]. Ethnographic studies of social media to date primarily focus on a narrow topic [8]. To address this gap, this study focuses broadly on cyberethnography of adolescents’ social media postings. Cyberethnography, by definition, is a type of research methodology that involves viewing media online and observing interactions in a digital space [9]. In this study, we conceptualize social media posting as self-generated content (SGC), defined as material that is created and shared by adolescents themselves on social media platforms. Though there is similar terminology, such as user-generated content (UGC), SGC differs in the sense that it is specific to what the individual participant posts versus just any form of content that is posted (retweets, reposts, etc). Our study also builds on previous social media coding tools such as the Application Programming Interface (API) or Python (Python Software Foundation) to use manual human coders instead of data scraping for data collection. Our research group proposes a cyberethnographic codebook designed to systematically categorize adolescent self-generated social media content across platforms, grounded in the lived experiences of teens’ posts [10].

Adolescence is a pivotal developmental period marked by milestones such as identity exploration [11,12], intimate friendship-building [13,14], and cultural socialization [15], all of which are commonly mediated through social media. Our codebook creates greater insight into these developmental periods through creation of a self-generated content categorization (coding) scheme informed by Bronfenbrenner (1979)’s ecological model [16]. Ranging from individual experiences to broader cultural systems, we unpack layers of contexts that may emerge from teens’ self-generated posts in the following section.

Engagement with cultural trends, politics, global issues, and social norms situates teens within broader digital and societal contexts. Adolescents prepare for adulthood by understanding social norms and gaining insight into the world around them [17,18]. By posting sociocultural content, adolescents explore civic engagement and political views and test their values and beliefs within the context of society. Engaging with these topics online and participating in larger communities allows adolescents to express and reinforce their values, helping them feel understood and validated within their social and societal contexts [19]. This sense of validation fosters a greater sense of belonging and safety [20], which in turn enhances their self-esteem and emotional well-being [21].

Moving on to interactions with family, peers, and romantic partners reflects immediate social ecosystems. One of the core developmental tasks during adolescence is the formation and maintenance of meaningful interpersonal and social relationships. Social media plays a significant role in this process, serving as a platform where adolescents can share aspects of their personal lives and sustain social connections [22,23]. These interactions often begin within the family unit, where adolescents engage in intimate relationships and shared activities. It is common for adolescents to post content about family events or milestones, as these interactions form the foundation of their social networks, with 44% of US teens reporting that they post content about their family on social media [24]. As adolescents spend more time outside the home, particularly in school, they learn to navigate and build friendships with classmates and peers in environments separate from their families [25].

Then we will review institutional activities that support adolescent development by providing structure, fostering social connections, and creating opportunities for personal growth [26-30]. Participation in these activities also contributes to identity development by allowing youth to explore and express aspects of themselves within institutional frameworks; joining a basketball team, for example, offers the opportunity to learn the sport while also conferring the identity of a basketball player. Furthermore, these activities offer social opportunities for adolescents to gain in-group affiliation (eg, being a member of the junior varsity basketball team). Social media may extend this process by providing a space to publicly display and reinforce these identities [2,31].

The next topic being adolescent behaviors, specifically displayed on social media, including self-presentation, identity exploration, and sharing of personal achievements, reflects adolescents’ developmental imperative to construct a coherent self-concept and assert autonomy [11]. Guided by the psychosocial task of identity formation—a hallmark of adolescence [32]—teens use platforms to experiment with roles, express authenticity, and document milestones. For instance, self-presentation through curated selfies and photos, or candid mental health disclosures, allows adolescents to negotiate how they are perceived by others, while identity-related posts enable them to explore and affirm marginalized aspects of their identity [33,34]. Furthermore, these activities scaffold identity integration, bolster self-esteem, and enhance emotional regulation, processes that underpin socioemotional well-being [22].

Finally, in the digital age, social media provides a key avenue for maintaining peer relationships. Adolescents use social media not only to interact with friends but also to update their social status, share group activities, and express romantic interests [23]. For example, posting pictures or updates about time spent with friends or significant others allows adolescents to communicate and reinforce social bonds. Posting relational content online is crucial for adolescents’ socioemotional well-being [23]. Adolescence is a time for learning social norms and building identity, and sharing experiences with peers, family, or romantic partners on social media reinforces a sense of belonging. By publicly validating their social identities, adolescents strengthen their connections to important social groups [35]. These interactions also provide feedback and validation, which are essential for boosting self-esteem and feelings of belonging [36]. Moreover, posting relational content helps adolescents navigate social norms and satisfy their social needs. These fulfilled needs promote overall well-being by fostering secure attachments, emotional resilience, and a positive sense of self.

Building on the developmental foundations with the existing research on adolescents’ social media use, we introduce an adolescent-centered, cyberethnographically informed codebook that objectively captures adolescents’ SGC across 4 major platforms. This study was created to observe adolescent social media and their related well-being, with our ultimate goal being to create reliable, replicable categories to be used in future analyses about how posting specific content on social media influences adolescent well-being. This work will, in turn, allow parents or guardians and potentially policymakers to make more informed decisions about youth in relation to social media and then ultimately how these decisions will affect the well-being of the adolescents.

To categorize the themes in adolescents’ online posts, the research team created a detailed codebook aligned with Bronfenbrenner’s ecological systems model (Figure 1) [16]. Our codebook was designed to be applicable for posts made across 4 platforms, including Instagram, Facebook, X, and TikTok.

The initial version of this codebook was informed by the lab’s prior research on adolescent posting behaviors. Between 2019 and 2022, the lab conducted a study that directly observed 145 youth over 2 years through social media monitoring and dense cyberethnographic field notes [37]. These field notes were used to identify common patterns and themes across participants, informing the development of the codebook for the present study. The team identified 5 overarching themes in adolescents’ social media posts that reflect developmental processes unfolding across each ecological level (Figure 1), including self-expression, relationships, activities, environment, and online behavior. Through these 5 categories, the team drew from template analysis and framework analysis to create the concepts for the codes using the data that were previously collected by this study team [38]. Table 1 presents the complete codebook of themes, including subcodes and examples. Multimedia Appendix 1 includes sample posts and the applicable codes for each post.

These 5 major themes formed the preliminary codebook, which was first implemented in April 2023. Over the first year of coding, the codebook was applied to all collected data and continuously refined to address emerging patterns and discrepancies, ensuring that codes both accurately reflected the data and had strong interrater reliability. This refinement process was done through frequent team meetings to discuss and compare coding between posts reviewed by multiple coders, all of which active coders were required to attend, and if they could not make it, they would be required to watch the recording during their next shift. Adding a new code to the codebook consisted of having a new theme brought up during individual coding that didn’t fit in with any existing ones. Then the team would keep a list of all emergent themes, and if it was brought up more than one time by more than one person, the team would then need to reach a consensus to agree to add in that new code to the codebook. All updates were carefully documented in a shared document between the team to maintain a timeline of revisions for future reference. The codebook was finalized once no new emergent codes were identified by any coder during the coding process, indicating theoretical saturation. After multiple iterations, the codebook was finalized in June 2024. Since the codebook has been finalized, no additional codes have been created; however, if a subject does come up that appears to not fit into a category, the team will meet to discuss what the best codes are that do fit that. All subsequently collected data were coded using this final version.

To ensure consistency across the dataset, all data coded before the codebook was finalized (April 2023-June 2024) was systematically back-coded according to the final codebook. Using the documentation of all codebook changes, the study team created a detailed guide for coders that included a comprehensive list of changes and instructions on how to address these changes while back-coding. The back-coding process, which ran from July to December 2024, took place alongside ongoing data collection. See Figure 2 for a timeline of codebook development and application.

Research Electronic Data Capture (REDCap; Vanderbilt University) is our primary data collection tool, chosen for being accessible, flexible, and secure. REDCap is a free, web-based, open-source platform that is Health Insurance Portability and Accountability Act (HIPAA)–compliant and hosted on our institution’s servers. REDCap allows research teams to create multiple customizable forms for data collection and coding tracking and provides storage for the repository of coding data. Although we do recognize that there are other tools that could be used for this project, REDCap has a special team at the University of Wisconsin–Madison that is ready and able to help with issues or questions on programming. This has been incredibly helpful for project creation and execution. This team had previously used Qualtrics (Qualtrics International Inc) for a different project and found that the data collection was fairly easy; however, when it came to data exportation and analysis, this platform was not as user-friendly as REDCap.

To view participants’ social media accounts, we created multiple study-specific profiles on each platform (Facebook, Instagram, TikTok, and X). Study accounts use fictitious names and photos designed to resemble typical user profiles and are periodically updated with posts. This approach reduces the likelihood of being flagged as bots or inadvertently revealing participants’ involvement in the study. Multiple profiles were also created in advance to serve as backups in case of accessibility issues. With parental consent and minor assent, we request to follow participants’ accounts on each platform using the study profiles during enrollment and conduct six-month follow-ups to verify that the information remains current. All research team members use the same accounts for data collection.

To ensure consistency in coding, our team first defined the unit of analysis—the portion of a social media post that would be coded. In this study, the unit of analysis is the entire self-generated post, including its media (eg, photos and videos), audio, and captions. Because our focus is on content created by participants, comments are not thematically coded. However, on platforms where engagement is visible, coders record the number of comments the participant writes, reacts to, or responds to, providing additional insight into how participants interact with their own posts. Each platform varied in the information that is to be collected. From Instagram having regular posts to reels to highlights to stories, we created a variety of forms to encapsulate this information in a meaningful way for this study. Instructions for the coding procedures can be found in Multimedia Appendix 2.

Our team uses a holistic approach, using participants’ shared social media accounts and other publicly available data to properly code each post. When needed, coders are instructed to gather contextual information from any of the participant’s social media accounts followed by the study team. Coders may also reference other public information, such as usernames of tagged individuals, to support coding determinations. For example, if a post features the participant alongside another similarly aged individual, it might be unclear whether this person is a friend or a family member. By reviewing the participant’s social media history, a coder may observe that this individual only appears in family-related posts. Additionally, a publicly visible username may reveal a shared last name, suggesting a familial relationship and allowing for more precise coding.

Coders are encouraged to explore social media platforms and the broader internet to contextualize posts to identify online trends, interpret online vernacular, or understand niche or unknown references. For instance, if a participant mentioned going to an “LARP event,” a quick Google search could reveal that “LARP” stands for “live-action role play.” Coders are also instructed to use search tools and platform-specific features—for example, TikTok’s shared public audio function—to identify trends or clarify content. However, coders are strictly prohibited from using personally identifiable information in any online searches and are instructed not to collect or record data from accounts that have not explicitly consented to participate in the study.

This project and team are a part of a larger longitudinal observational study that is conducted over the course of 24 months, using a mixed methods study design consisting of both qualitative and quantitative data collection [39]. There are 3 principal investigators (PIs) who man individual teams varying in size from 5 to 20 research staff at any time, which are made up of undergraduate students, graduate students, postdoctoral students, and professionals.

The larger longitudinal observation study is collecting well-being data at different time points throughout the study. Some of the well-being metrics being collected include Media and Technology Usage and Attitudes Scale (MTUA), General Anxiety Disorder 7-item (GAD-7), Emotional Responses to Social Media Experiences Measure (ERSMEM), 8-item Patient Health Questionnaire (PHQ-8), and more [40-43]. These metrics are collected at 3 time points over the 24 months and will be used to connect to the participants’ social media data at different time points throughout the study.

The participants in this study are adolescents aged 13‐15 years from the state of Wisconsin who have at least one active social media profile on Facebook, X, Instagram, and/or TikTok. These participants were sourced from local events in Wisconsin such as the state fair, Wisconsin Science Festival, and more, where team members from the study talked to potential participants and their parent or guardian about the study and enrollment. Our team used the census data from 2020 for the state of Wisconsin for our targeted enrollment demographics. Each participant, when joining the study, shares their social media platforms and is followed by the study-specific profiles mentioned above. This team then goes into each participant’s profile every month to code and collect the data.

Our team began coding study data in April 2023, with the initial coding conducted by research staff with backgrounds in adolescent medicine, public health, and psychology. Beginning in September 2023, undergraduate student research assistants were hired to assist with data collection. The size of the coding team has fluctuated over the course of the study based on workload and coder availability. As of the time of writing, 28 coders have contributed to data collection, each participating for at least 4 months. The standing study team for this project is built of PI (ES), 5 research staff (KB, LB, TF, KK, and CMC), and alternating undergraduate student research assistants.

First, our study team developed a comprehensive coding guide to support coders in both learning the coding process and referencing updated procedures throughout the project. The guide was designed to be highly visual and easy to navigate and included screenshots of social media and data collection platforms, step-by-step coding instructions, and troubleshooting strategies. Any updates to the coding process are immediately reflected in the guide, which is used as the definitive reference to ensure consistency across coders and throughout the data collection process.

Coders begin training by meeting one-on-one with the lead study team member who reviews the coding process and explains each code in detail. Coders then practice applying this codebook to a sample social media post with the assistance of their supervisor to assess their understanding. Using the coding guide and codebook, coders then complete a training exercise by coding 24 preselected social media posts (6 from each platform). Upon completion, a REDCap report identifies agreements and discrepancies between their coding and the answer key, created by their supervisor. Each coder then meets with their supervisor to review results post-by-post, discussing agreements, discrepancies, and refining their understanding of the codebook.

Coders who achieve at least 80% agreement with the master coder begin official data collection under close supervision to address any lingering questions. While most coders reach this threshold after the first round, those who do not repeat the process with a new set of posts. As of writing, no coders have required more than two rounds, though a third set of posts was prepared as a contingency.

Coding is conducted monthly for each participant, with each 30-day interval following enrollment representing an observation period. Participants are followed for 24 observation periods as part of a larger longitudinal, multimethod study [39]. Content is coded from participants’ Facebook, Instagram, TikTok, and X accounts (Figure 3). Something to note is that each post can be assigned more than one thematic code, as many posts are complex and include a variety of subjects.

We assign coding using REDCap’s Data Resolution Workflow, a feature typically intended for formally flagging data issues (eg, missing, incorrect, or invalid values). This workflow enables users to open a query attached to a specific data field, assign it to another user, and track its resolution. Although not designed for this purpose, our team adapted the feature to assign coding tasks to individual team members.

Coding is organized by “participant-months,” where each unit represents one month of coding for a single participant. At enrollment, each participant’s full coding schedule was programmed into REDCap for the duration of their study participation. When a participant-month became due, it appeared on REDCap’s calendar.

To make assignments, the data manager reviews the calendar daily to identify participant-months scheduled for coding. For each assignment, the manager opened a query in a form on the participant’s profile, specifying the participant-month to be completed. Coders accessed their assignments through the data resolution dashboard by filtering queries by their username.

Given the volume of data, number of coders, and duration of the study, it was essential to establish rigorous procedures to ensure data integrity. To support accuracy and consistency, our team implemented annual interrater reliability checks, double-coding procedures, and weekly alignment meetings.

Because 2 of our PIs are mandated reporters, our study team has both a legal and ethical obligation to comply with mandated reporting requirements [45]. To support this, we developed a standard operating procedure for how to identify and report content on participants’ social media profiles that might meet reporting criteria. Our standard operating procedure ensures that any reportable content is identified and escalated to the investigators in a timely and responsible manner.

For the purposes of this study, concerning content is defined as posts that referenced suicidality or self-harm, sexual or physical abuse, or threats of harm or violence toward others. Such content can appear in the participant’s own words or in reposted material that suggests the participant may be at personal risk of harm. A made-up example of this would be a participant posting “I’m going to school with a gun tomorrow,” which would require prompt action from our team. Notably, concerning content does not include all illegal or risky behaviors, which do not always meet the threshold for mandated reporting [46].

Institutional Review Board approval was obtained from Application Review for Research Oversight at Wisconsin (ARROW) under Institutional Review Board number 2022‐1280. Informed consent is given through a survey through REDCap that each participant’s parent or guardian must sign before being enrolled in the study. All participant and related data are stored in the secure University of Wisconsin-Madison REDCap and are deidentified when exported for analysis. Participants can receive a total of up to US $300 for completing all study-related activities.

This study was funded in 2022 and began data collection in early 2023, with the codebook still in development, with the study concluding in 2027; however, there is a possibility of continuation. Figure 2 outlines the timeline from beginning to the expected end of the study. There are 369 participants at the time of this writing, with 336 actively enrolled and 33 withdrawn.

Figure 4 illustrates the trend in participant-months requiring coding. The workload peaks around the midpoint of the study (mid-2025) and then declines as participants reach the end of the 2-year enrollment period. Month-to-month variation reflects the staggered and fluctuating nature of participant enrollment.

Annual interrater reliability assessments produce 5 AC1 scores, one for each platform (Facebook, Instagram, TikTok, and X) and one overall score. Year 1 results, shown in Table 2, indicate that Instagram had the highest agreement (0.89) and X the lowest (0.87), with all scores exceeding the 0.80 acceptability threshold. In Year 2 to date, X shows the highest agreement (0.96), while Facebook has the lowest (0.95).

Applying this method included important considerations, including the team, participant social media use, and policies impacting social media use.

The implications of our methodology are many. First, the provision of granular, researcher-observed data regarding self-generated social media content not only densely phenotypes adolescent online behavior but also allows for identification of more precise online experiences that may have positive or negative impacts on a multitude of physical and mental outcomes. The impact of social media on the health and well-being of adolescents is of concern to adolescents, parents, policymakers, and the public [53,54]. Objective identification of risky and healthy online behaviors can aid in future development of interventions, from parental, adolescent, and clinician-based education to detection and amplification at the design level. The latter, however, would require close partnership and transparency between researchers and social media companies, which may need to be enforced at a policy level.

Across other research studies, there have been a variety of different approaches taken to code adolescent social media. One study focused on coding race- and ethnicity-related content (text and images) in Facebook public posts through a hybrid inductive-deductive approach [57]. Which is similar to our study in that we used a hybrid approach but focused mostly on psychosocial and/or well-being–related content across 4 platforms, and we also included private posts as well. Another study focused on health-risk–related content, which also followed a theoretical theory (deductive approach) first followed by an inductive approach, so it was very similar to our approach that we used [58]. As for content coding of profiles versus just posts, one other paper coded users’ profiles and then further developed a codebook for sexualized behaviors on Facebook, which draws to our study as well because we not only collected posting data but also coded profiles for each account per platform [59]. Finally, some of our team’s previous social media research work includes creating a codebook that was designed for YouTube video content, which ultimately helped us to have an idea of how to navigate creating a codebook for social media data collection [60].

This project offers both methodological and theoretical advances to adolescent social media research. First, it adds to the field’s previous evidence using self-reported screen time or broad “active–passive” classifications by objectively analyzing the actual posts adolescents create. Each participant’s content (video, text, photo, etc) is captured across multiple platforms (TikTok, Instagram, Facebook, and X) over a 2-year span, providing a granular, longitudinal view of their digital lives that is unattainable with single-platform or cross-sectional designs. Second, the study introduces an adolescent-centered cyberethnographic coding framework. The codebook was applied by rigorously trained coders, ensuring high interrater reliability while eliminating recall bias. This participatory approach yields a faithful record of what adolescents choose to share online. Third, the coding scheme is grounded in Bronfenbrenner’s ecological model, capturing posting themes at 5 nested levels, including individual, interpersonal, institutional, environmental, and systems. By aligning observable social media behavior with a well-established developmental theory, the study links online self-generated content to the broader contextual layers that shape adolescent well-being. Together, these innovations furnish a holistic portrait of adolescents’ online expression and lay the groundwork for future research examining how specific posting patterns relate to socioemotional outcomes.

Our study team is already in the process of running initial analysis on the descriptives of these data that we are collecting. Beyond a descriptive analysis, we have been working on cross-analyzing the engagement of the participants’ peers on their social media posts in relation to the participants’ overall well-being from different time points throughout the study. These are just a few examples of how this data will be used; however, overall, we are planning to run many different types of analyses on the well-being data that are being collected and the SGC that is also being coded by this team. We will be able to show how one impacts the other and vice versa, which is information that will hopefully help parents or guardians and policymakers.

This formative research approach that was taken with our lab has allowed for us to have insight and collect data that will give a new perspective on social media and adolescents. This, in turn, will help our research team to further show the connection between social media and socioemotional well-being. Specifically, this team plans to write a descriptive paper on the data that is being collected. We then plan to work with the other teams on the bigger project to connect participants’ well-being with the social media data that we have collected. This team works with undergraduate and graduate students, and postdoc and research professionals across the University of Wisconsin–Madison, who will also use the data that we have collected to answer differing research questions. This approach will aid future studies in project creation and execution on studies similar in nature. The foundation that has been laid out by this team should be a guidance to further the research on the effects of social media on adolescents and within other age groups as well. This research strategy could also be extended to conducting research in different ethnically diverse populations across the globe. This research is critical to understanding the well-being of individuals who participate in social media use, especially with the ever-changing nature of social media. This team will continue to publish research outcomes from the data collected from this methodological approach and will carry on studying the well-being of adolescents and their interactions with social media.