Beyond Digital Literacy: Building Youth Digital Resilience Through Existing “Information Sensibility” Practices

Abstract

Youth media consumption and disordered eating practices have historically been subjects of moral panics, often resulting in protective, deficit-based interventions like content removal. We argue for interventions which instead equip youth to evaluate and manage risks in their online environments, building upon their existing “information sensibility” practices. Drawing upon ethnographic research and intervention testing with 77 participants in the US and India, we analyze how youth (aged 13–26), including those with diverse political perspectives and those recovering from disordered eating (DE), engage with online news and health information. Participants generally algorithmically encountered (rather than searched for) information online, and their engagement was shaped more by social motivations—like belonging—than truth seeking. Participants interpreted online information collaboratively, relying on social cues and peer validation within their online communities. They demonstrated preference for personal testimonies and relatable sources, particularly those with similar social identities. We propose resilience-building interventions that build upon these youth online information practices by: (1) leveraging peer networks, promoting critical information engagement through collaborative learning and peer-to-peer support within online communities; (2) developing social media sensibility, equipping youth to critically evaluate information sources in situ; (3) providing pathways offline, connecting youth to desired in-person communities; and (4) encouraging probabilistic thinking.

1. Introduction

Youth media consumption and disordered eating (DE) practices have historically been subjects of moral panics (Gooldin 2008), often resulting in protective, deficit-based interventions like content removal and parental controls. We argue for interventions which instead equip youth to agentially evaluate and manage risks in their online environments, building upon their existing practices. This paper develops strength-based interventions for youth digital resilience by analyzing how youth (aged 13–26) engage with online information and testing interventions designed based on these findings. Strength-based interventions focus on how to leverage people’s existing abilities and practices to increase resilience against harm, rather than focusing on their deficits. Resilience-based approaches seek to maintain people’s autonomy and ability to self-regulate when facing harms, instead of paternalistically shielding them from encountering risk (Park et al. 2024).

Most research on youth digital literacy and DE follow medical approaches by centering analysis on individual encounters with “harmful” content, resulting in individual-oriented interventions which do not match the culturally mediated and socially oriented way that youth encounter, evaluate, and apply information. These interventions are often deficit-based, aimed at identifying, labeling, and removing “harmful” content through moderation to protect youth. We argue that a focus on content (1) fails to account for the contextually mediated nature of harm (that a piece of content can be helpful or neutral for one youth while being harmful to another) and (2) does not result in solutions which equip youth with the tools to evaluate content harms in situ.

Moving beyond traditional literacy-based paradigms, we argue for interventions based upon youth “information sensibility” practices—how youth develop a socially informed awareness of the value of information encountered online, relying on folk heuristics1 of credibility (Hassoun et al. 2023). Information sensibility differs from traditional information literacy, which often relies on individual linear models of engagement, the mastery of core competencies like fact-checking and lateral reading, and the assumption that accurate information is always the highest-quality or highest-value information (and that validating content accuracy is people’s primary information engagement goal) (Mackey and Jacobson 2011).

We instead engage anthropological analyses which frame digital literacy and DE practices as fundamentally relational (rather than individual). By explaining how dietary science has long been embedded in moral (often sexist and racist) frameworks, we argue for relational analyses of how youth make sense of eating-related health information online—particularly marginalized youth, who often doubt medical models of health and eating that have historically been discriminatory. Building on research explaining how DE practices are mediated by desire for social belonging, we analyze how youths’ desires to identify with emerging health-related online identities affect their information sensibility practices.

Information sensibility stresses the social nature of information processing for youth, who often encounter, rather than actively search for information, and sees social motivations as primary drivers in how they engage with and interpret it. Youth understanding and evaluation of online information is socially informed: heavily influenced by their social context, including their peers, family, and online communities. Youth interpret information together, as aspirational members of social groups, rather than as isolated individuals. Their informational and social needs are inseparably entangled. Youth assess information not just for its accuracy or facticity, but also for its value to their personal lives, relationships, and their ability to navigate their social environments. The social utility of information shapes its importance and purpose for them. This social significance of information is often as important as, or even more important than, its accuracy.

Our findings expand previous work on the social appeal of misinformation (Duffy et al. 2019; Geeng et al. 2020; O’Connor and Weatherall 2019) by detailing the social influences affecting youth information engagement (e.g., Herrero-Diz et al. 2020). We suggest that social support and belonging may be key solutions to the problem of youth misinformation susceptibility. Many participants’ core problem was not misjudging misinformation (solved via literacy-based competencies) but rather a desire for social belonging, which they sometimes solved by sharing misinformation. Interventions targeting youth susceptibility to misinformation, therefore, must go beyond classroom-based competency-building paradigms to support the underlying social and emotional needs driving youth misinformation sharing.

1.1. Research Questions

We conducted two ethnographic studies with 77 participants in the US and India, investigating their online information practices around news and health information. Study 1 asked the following research questions:

(RQ1) Seeking information—Why, where, and how do youth seek information online?

(RQ2) Assessing information—Under what circumstances do youth attempt to assess if information they engage with online is accurate, credible, and high-quality? How do they perform such assessments?

(RQ3) Susceptibility to misinformation—What factors affect the susceptibility of youth to misinformation?

From Study 1, we found:

• Participant informational and social needs were inseparably entangled.

  (1)

  Participant information journeys rarely began with a truth-seeking search query.

  (2)

  Participants used information to orient themselves socially and define their emerging identities.

Study 2 included participants recovered from DE and asked the following additional eating- and health-related research questions:

(RQ4) Trusted health information—How do youth find trusted information on eating and health? What are their information sensibility practices?

(RQ5) Healthy or harmful behaviors—How do those patterns and encounters with online eating and health information lead to healthy or harmful behaviors?

(RQ6) Misinformation and authoritative information—What constitutes “misinformation” and authoritative health information in youth’s online contexts?

From Study 2, we found:

2.

Participants used similarity to themselves as a trust heuristic to determine whether an information source was trustworthy, asking if the content creator was “like-minded and like-bodied” (what we term n = 1 thinking).

3.

Participants crowdsourced credibility, using distributed trust heuristics to confirm with peers whether a source was trustworthy, instead of relying on more traditional markers of institutional or scientific authority.

4.

Participants sought to demonstrate online that they were living a holistically “healthy” life, in which performing the “right” body-based practices signified their “goodness” (and resulted in better access to social capital and material resources).

5.

Even when participants found useful online resources and communities, they believed that the “good life” was ultimately lived offline and wanted to decrease the time they spent online.

In both studies, we found that information encounter—whether for news or health information—was socially driven: participants generally encountered (rather than searched for) information online, and their engagement with it was shaped more by social motivations like belonging and identity formation than truth seeking. Further, social sensemaking shaped trust assessments: participants interpreted online information collaboratively, relying on social cues and peer validation within their online communities. They demonstrated preference for personal testimonies and relatable sources, particularly those perceived to share similar social identities (“like-minded, like-bodied”). These practices could increase potential digital-safety risk, particularly from following health guidance from such sources. These findings highlight limitations of content-focused interventions and emphasize the potential for strength-based approaches built upon the social embeddedness of online information practices.

Based upon these ethnographic research findings, we worked with designers and product managers at a technology company to create mockup interventions to test with participants in follow-up interviews. From the ethnographic findings and intervention-testing results, we propose resilience-building interventions that:

(1)

leverage peer networks, promoting critical information engagement through collaborative learning and peer-to-peer support within online communities;

(2)

develop social media sensibility, equipping youth with the skills to critically evaluate information sources in situ, situate creators’ motivations, consider potential harms from following their advice, and navigate social influences online;

6.

provide pathways offline, connecting youth to in-person communities;

7.

encourage probabilistic thinking, countering some of the digital safety risks increased by n = 1 thinking.

Next, we review existing literature to argue for a relational approach to building youth digital resilience, particularly with respect to eating and health information. We then describe our methodology, results, intervention testing, and discuss the implications of our findings for developing strength-based interventions for youth digital resilience.

1.2. Youth Information Literacy and Disordered Eating

While “media influence” has long been a concern in DE literature (Bordo 1993), the “always on” cadence and algorithmically powered delivery of online content give rise to new conditions shaping youth eating and health information engagement (Marwick and Boyd 2010; Buckingham 2008; Pater et al. 2016). Social media is a key site where youth learn how their bodies should visually reflect adherence to dietary and fitness cultures (Weinstein 2018; Marks et al. 2020). Wearable technologies and digital health tracking applications make health vitals, calories burned, and steps walked constantly referenceable (Eikey 2021; Schüll 2016). These technologies are premised on the assumption that “more information = better decision making” (Tregarthen et al. 2015), implying that “poor health” comes from literacy or data deficits rather than access inequities or discriminatory normative definitions of health (Neff and Nafus 2016).

Algorithmic recommendations present emergent challenges for youth engaging with eating and health content online (Griffiths et al. 2024; Costello et al. 2023). As Eaton describes (2016), social media platforms make decisions about what users see: “When I scrolled on the app, I made choices, but the algorithm made just as many, if not more”. Algorithms “work too well in catering to user preferences, bombarding users with the content they need to facilitate their self-harm” (Lai 2022). Lai (2022) argues:

For users with preexisting body image issues, seeking out one or two fitness or healthy recipe videos could fill their feeds with similar videos, and those who continue watching similar content could easily be led to content explicitly promoting eating disorders. Such regular exposure has the potential to trigger or worsen disordered eating behaviors.

Research on Internet use and DE frequently focus on pro-anorexia (“pro-ana”) content, portraying extreme fasting as a lifestyle choice (Logrieco et al. 2021). “Thinspo” and “thinspiration”, content that inspires people to be thin, has “a negative impact on viewers with regard to body image” (Yamamiya et al. 2005). Based on a survey of pro-ana Internet searches and websites, Greene and Brownstone (2021) find “search terms with references to thinspiration and thinspo are associated with the most harmful Website content”. Such content may induce a “toxic spiral” where content becomes intensified and more explicitly harmful (Austin 2021). Interventions tend to be content-focused (Dedrick et al. 2020): removing “harmful” content or replacing it with “healthy”, body-positive or body-neutral content, testing the effects on individuals’ mood and self-image in controlled, lab-based environments (Cohen et al. 2019).

Studies provide limited insight into how and when technology companies can (or should) moderate a user’s informational health ecosystem to prevent harm (Gillespie 2020; Lenhart and Owens 2021). Interventions which identify and remove harmful content alone have demonstrated limited evidence-based success, contributing to “creating isolated and polarized communities” and making search terms inaccessible for those seeking support—particularly at-risk youth (Zhang et al. 2024; Chancellor et al. 2016). Trigger warnings evidence little to no effect and can be countertherapeutic for those with trauma histories (Greenfield 2012; Jones et al. 2020). Skills-based information literacy approaches like fact-checking, widely taught in schools, show similar limitations (Mackey and Jacobson 2011; Breakstone et al. 2021; Gordon et al. 2021).

These content removal and literacy-based interventions may not be effective because conventional paradigms of information literacy do not fully capture what youth actually do online (Boyd 2017) or account for the contextually mediated nature of harm (Yeshua-Katz and Martins 2013). In their everyday lives, youth are motivated by more than what is “true”: they want to discover, learn, and share what is meaningful to them and their social communities (Hassoun et al. 2023). Youth also encounter information in situ and make sense of it with their peers (Eynon and Malmberg 2012; Evans et al. 2010; Feng et al. 2022), making lab-based studies measuring individual encounters inadequate simulations of their actual online practices. Further, most content is neither intrinsically harmful nor helpful—content harms differ based on an individual youth’s sociocultural context, user state, and mode of encounter (Warford et al. 2022).

1.3. Relationality of Disordered Eating Practices

Medical approaches to DE generally focus on individual pathologies, often including “culture” only as a risk factor or reductive proxy for ethnic categorization (Anderson-Fye et al. 2015). Studies of youth Internet use and DE largely follow this individual-focused medical model. We argue these approaches under-account for how youth use socioculturally mediated eating and health practices—including DE—to make meaning: “how the practices of body discipline and affective modulation shape a person’s sense of being in the world” (Lester 2021). We suggest that for youth, these meaning-making practices are not just individual: they are relational social practices performed with communities making sense of online information together.

Existential questions driving DE practices are likewise culturally specific and mediated (Pike and Borovoy 2004; Pike and Dunne 2015). Anthropological studies in India have long integrated questions of food and eating with purity, the body, and social distinction (caste) (Appadurai 1981; Khare 1992; Marriott 1989; Raheja 1988). Recent studies have analyzed everyday practices of eating, belonging, and identity-formation (Mines 2009; Rao 2009; Pandian 2009). Researchers in the US have begun analyzing how subjective experiences of anorexia and broader cultures of dietary restriction and moderation are shaped by religious idioms, symbols, and ideologies (Bell 1985; Cheney et al. 2018; Giles Banks 1996; Twigg 2011; Gressier 2021).

DE practices must also be historically situated, shaped as they are by the enduring influence of gender discrimination and colonialism. In eighteenth-century England, anorexia communicated a woman’s purity and delicateness (Bart and Scully 1979; Showalter 1985). The Victorian pathologizing of women’s DE co-produced modern dietary medicine (Brumberg 1988; Williams 2020): appetite, a feminine part of the body, needed to be combated by “masculine” self-control (Bartky 1990).

British dietary medicine, widely used in North America, also emerged partly from Britain’s colonial project in India. Early dietary medicine was invested in preserving white European bodies amidst growing wealth from colonization, overtly reinforcing ethnoracial discrimination (Dacome 2005; Albala 2005). It, for example, viewed Indian vegetarianism as a cultural weakness resulting in bodily and psychic impoverishment (Ghassem-Fachandi 2012; Philip 2004; Roy 2002, 2010; Zook 2000). Dieticians claimed that “hot” foods of the colonies would raise English “passions” to the detriment of their physical wellbeing (Arnold 1993; Stoler 2002). As appetite control became a British cultural expectation, dietary medicine idealized thin, white, female bodies as cultural containers of national virtue while casting black and brown bodies as dangerously insatiable (Hamel 2018; Ames 1691).

Indian indigenous medical, philosophical, and religious perspectives on food continued to flourish despite colonial rule (Alter 2005; Langford 2002). Engagement between Buddhist, Hindu, and Jain philosophies developed syncretic cultural ideals around ahimsa (nonviolence), eating, and moral practice. Ayurvedic conceptions of body and health have continuously informed local articulations of disease, even as they became blended with Western biomedical ideas (Das 2015; Ecks 2013; Solomon 2016). The dominant health regime remains medical pluralism, “the availability of different medical approaches, treatments, and institutions that people can use while pursuing health”, though socioeconomic factors like healthcare privatization, caste and religious discrimination, and financial inequality create treatment access disparities (Khalikova 2021; Banerjee 2009).

In the US, as anorexia and bulimia nervosa diagnoses gained public attention in the 1990s, researchers argued that media images promoted damaging body ideals, principally aimed at women (Wolf 1990). The belief that anorexia is an “upper class white woman’s illness” has led to medical and cultural ignorance about eating disorders among male-identified, non-binary, queer and trans individuals (Duggan and McCreary 2004; Feldman and Meyer 2007, 2009). Due to omission of men from studies and female-based diagnostic criteria, statistics on DE prevalence in men in most pre-2017 studies are gross underestimates (Wu et al. 2020). Yet in a sexist double-bind, because DE has been treated as a women’s health problem, treatment and research is chronically underfunded. DE in India is also under-researched, resulting in a lack of “culturally sensitive instruments for diagnosis” (Vaidyanathan et al. 2019). Our research focuses on these understudied populations while also attending to the unique harms this historical treatment of DE has had on women-identifying people.

Ultimately, DE practices are imbricated in historically shaped sociocultural systems of belief and practice. We must understand how youth evaluate and apply eating and health information within secular, syncretic, and cosmopolitan frameworks that incorporate alternative medicine, nutritional logics of “macros”, trending topics, popular beauty ideals, and religious beliefs (Bansal 2012; Kumar 2019; Lukose 2009; Nakassis 2016; Poonam 2018). Based upon this context, we analyze how medical pluralism continues to shape perceptions about what is “healthy” and “harmful” in India. We also demonstrate the lasting effects of these discriminatory dietary science histories in India and the US, showing how youth’s awareness of these historical inequities affects their trust in and evaluation of “authoritative” online health and eating information.

1.4. Health-Related Online Identities

Anthropological analyses deconstruct the causal explanation that young women uncritically consume (Western) media images of thin bodies, leading to DE behaviors (Bray 1996). They also demonstrate that DE practices are not simply Western imports stemming from fat-phobia and body image issues (Littlewood 1995; Ritenbaugh et al. 1996). These critiques urge focus on the relational motivations shaping DE practices. One such motivation is the social belonging that illness categories mediate (Eli and Warin 2018; Lavis 2015; Warin 2010).

Emerging health-related online identities and the sense of community belonging they make possible affect youths’ information sensibility and DE practices. Researchers have observed the multiplication of cultural identities around food, exercise, and wellness (Charmaraman et al. 2021; Jenzen 2017). From sub-cultures of carnivory to warring factions of vegan YouTubers, a growing number of eating-related identities are available to youth via online engagement (Bissell 2018; Hamblin 2018).

While research has shown how online communities become misinformation vectors (Hassoun et al. 2024), such communities might also provide support for youth digital resilience. During the pandemic, youth developed new ways to care for each other through online spaces—in part by sharing health and well-being information—amidst the increased digitization of social life (Unni and Weinstein 2021). Mental health discussions in online communities, like sharing coping stories, have demonstrated psychosocial benefits for youth (Yuan et al. 2023; Niu et al. 2024). We can learn from what youth are already doing in online communities dedicated to eating and health practices and identity-formation (Smith and Keel 2017).

In short, our research interrogates the prevailing wisdom that youth would be less susceptible to DE or political misinformation if they were more capable of decoding media messaging. Youth are not simply “fact-checking” when evaluating information online. They are weighing a host of social inputs derived from their social identities and communities when deciding whether a given piece of content or information is meaningful to them.

Next, we describe our materials and methods, detail the results of our ethnographic research and intervention testing, and discuss their applications for developing youth digital resilience.

2. Materials and Methods

2.1. Overview of Studies 1 & 2

We conducted two ethnographic studies. The first (documented in Hassoun et al. 2023) studied 35 internet users aged 13–24 from the US. Study 1 asked about youth information-seeking practices in general, with a secondary focus on news information. We recruited participants who reported mainstream (n = 19) and alternative (n = 16) political beliefs2. This was conducted between February and June 2022.

To examine whether these general findings held true in a specific, sensitive topic area and beyond the US, Study 2 asked if and how youth in the US and India apply the practices of “information sensibility” identified in Study 1 to information about eating and health. Half of Study 2 participants were recovering from disordered eating (DE) practices. This second study (briefly introduced in Xu et al. 2024) focused specifically on youth eating and health informational practices. For Study 2, we performed in-person ethnographic research, including walkalong interviews and participant-observation, with 42 youth aged 18–26 in the US and India between June and October 2022.

Ethnographic research helps uncover how youth interactively incorporate learned ideas about the news, eating, and health into their online informational practices. Our ethnographic approach responds to calls for sociotechnical (Marwick 2018) and sociocultural (Lloyd 2018) analyses of youth digital-safety risks to help researchers, governments, platforms, and youth develop better interventions (Brown and Duguid 2017).

The rest of this section presents detailed methods for Study 2; detailed methods for Study 1 can be found in Hassoun et al. 2023.

2.2. Study 2 Participants and Recruiting

We recruited participants from the US and India (

Table 1. Study 2 participant location.

	Urban	Suburban	Semi-Rural	Total
USA	11	5	5	21
India	15	3	3	21
Total	26	8	8	42

and

Table 2. Study 2 participant gender identity.

	Male-Identifying	Female-Identifying	Trans	Non Binary	Total
USA	5	10	4	2	21
India	10	11	-	-	21
Total	15	21	4	2	42

). In the US, we selected participants through the platform respondent.io. We chose participants from the broader New York area. 25% of participants reported a household income (HHI) of less than USD 40,000, 30% earned between USD 40,000 and 60,000, 25% had an earned between USD 60,000 and 100,000, and 20% earned more than USD 100,000. In India, we used a local recruiter and included participants in Delhi and nearby peripheral areas. Incomes ranged from USD 7000 to 48,000. In both India and the US, all but one participant attended at least some college.

Because most youth who experience DE are not formally diagnosed (particularly in India, where DE is less pathologized and diagnosed than in the US), we recruited based on self-reported experiences with DE (rather than formal medical diagnoses).3 Ethical concerns also informed the decision to recruit participants in recovery, rather than those actively in distress or experiencing DE. Participants, however, largely did not see recovery as a linear or fully complete process. They were able to speak in detail about how their online experiences actively affected their eating and health practices.

We did not examine specific eating and health information or DE habits among 13–17 year olds for ethical reasons, deciding that such a study might expose participants to undue harm. This affects our ability to generalize about young adolescents across Study 1 and Study 2 and limits our eating and health observations to young adults aged 18–26. We did not observe sharp differences within 18–26 year olds in Study 2, but observed gradient effects, which we discuss later. Since Study 1 included early adolescents aged 13–17, the age, period, and cohort effects we observed are detailed in Hassoun et al. 2023.

Researchers in the US and India had extensive prior experience studying the eating and health cultures of their research context and were fluent in the languages spoken by participants. Field and interview guides were designed to account for cultural differences4, and the way researchers spoke about and contextualized DE in each site therefore differed. In both contexts, aspects of diet are (sometimes dangerously) politicized. For example, in India, eating meat, or even rumors of eating meat, have caused vigilante violence against Muslims and Dalits. Meat-eating in the US carried different symbolic meaning: masculinist discourses about meat-eating affected US participant experiences, sometimes to harmful effect.

We treated some “baseline” screening questions differently as a result. In India, we focused on eating behaviors while protecting participants’ safety by not probing for specific details about dietary choices (like eating meat). Based on existing research on food culture in India, we focused our recruitment on behaviors that could proxy for intensity of interest in the subject, for example, time spent thinking about food and engaging in exercise.

Additionally, public discourse about the body and health is articulated differently: in an Indian context, it is far more appropriate to talk explicitly and openly about weight, weight gain, weight loss, and bodily dissatisfaction in a way that might be inappropriate in a US context. While experiences with weight and health stigma overlapped significantly, participants often used different vocabulary, which we accounted for and analyzed (e.g., US youth spoke more about “energy” while Indian youth referred to “balance”). Categories of “eating disorders” or “disordered eating” are far less prevalent in India, and there can be significant stigma around discussing or experiencing mental health issues. In contrast, US participants regularly used medical terms like “bulimia”, “anorexia”, “depression”, and “disordered eating”. Researchers were sensitive to these differences.

Participants used many overlapping platforms, but one of the most commonly used image- and video-sharing platforms among participants in the US did not exist in India.

2.3. Study 2 Research Methods

Participants’ ability to self-report their own actions, design desires, and mental models, particularly in a controlled lab setting, is limited (Bernard 2017). These self-reports are shaped by social norms, generating gaps between user narration and practice that require qualitative analysis methods to explain.

As a result, in both studies we chose ethnographic methods that probe cultural values that shape human behavior, unearthing “that which is left unsaid” (Rosaldo 1993). We sought to understand youth’s information-seeking practices in situ, to understand the effects of family values and influences, social and economic conditions, and technology use in the context of everyday life. Study 1 included three phases of data collection, using dyadic interviews (Phase 1), digital diary exercises (Phase 2) and 1:1 interviews with an intervention testing component (Phase 3). It also included a data analysis and intervention design workshop between Phases 2 and 3. Details can be found in Hassoun et al. 2023.

Study 2 also employed a three-phase, mixed-methods approach with an intermediate data analysis and intervention design workshop:

Phase 1: Diary Exercise. Participants created visual representations of their conceptions of “healthy” and “unhealthy” with collages of images drawn from their social media environments.

Phase 2: In-Person Ethnographic Research. Researchers performed participant-observation and semi-structured interviews, conducting home and community visits and accompanying participants on a self-selected eating or health-related activity (e.g., a grocery shopping trip, jogging in the park).

Data Analysis and Intervention Design Workshop. Following ethnographic data collection, researchers performed preliminary data analysis to identify emergent participant needs. To contextualize these findings within existing platform moderation practices and design realistic potential youth digital resilience interventions, we conducted a design workshop with product managers and policy specialists specializing in health and youth digital-safety at a large technology company. We co-designed interventions to test with participants in Phase 3. Details about the design process can be found in the Supplementary Materials.

Phase 3: Online Follow-Up Interview and User Testing. Online follow-up interviews with participants were conducted to test and discuss potential interventions derived from the workshop, gather further qualitative data, and engage in member checking using collaborative reflection techniques (Urry et al. 2024).

2.4. Analysis Methods

Our analytic approach in Study 2 was reflexive, iterative, and inductive, following a grounded theory approach to qualitatively code data (Charmaz 2006; Mills et al. 2006; Saldaña 2021). Rather than testing participants for deficits or evaluating them against preset frameworks, we adopted an emic lens (Scarduzio 2017) to learn about what participants were doing with information online and for what purpose—from their own perspective and in their own words.

Two researchers coded each interview or field note during the data collection process. We performed consensus coding throughout, comparing codes assigned by researchers after each iteration and resolving divergence through discussion (Richards and Hemphill 2018). We used an open, descriptive coding method during our first coding pass. We then used a constant comparative method during subsequent focused analytical coding passes (Corbin and Strauss 2015), tabulating each discrete code (e.g., admitting mistakes is a sign of trustworthiness) against each participant whose interviews or diary submissions generated that code and comparing the underlying data.

Once data collection and preliminary coding were complete, we held several collaborative analysis sessions. We performed multiple theoretical coding passes, writing short memos to articulate overarching themes. We then iterated those interpretations through discussion. We referred regularly to the codebook to ensure that our analysis reflected the complete dataset and drew conclusions about how analyses applied to different populations within our dataset. We used the codebook to test data saturation and reflexively mitigate cognitive biases during data collection and analysis.

2.5. Limitations

Readers should consider several limitations of our study’s methodology when interpreting the findings. Qualitative research, while providing rich, deep insights into social phenomena, has inherent limitations. First, our small study sample, drawn from two geographically defined areas, cannot claim global representativeness. We studied a small number of experiences in-depth. We recruited to ensure that our sample did not skew toward a specific demographic and attempted to collect a diverse sample. To mitigate the over-representation of white, affluent, and female participants in prior studies, we prioritized diversity across gender and sexual identity (male, nonbinary, queer), race (non-white), geographic density (urban, peri-urban), income levels, political leaning, and eating and health behaviors. We propose that further explicit diversification across other characteristics and increased sample size could have expanded the study’s relevance.

Second, the study relied on self-reported data from participants through diary exercises and interviews. These data are subject to inherent limitations such as recall, observer, and social desirability biases. The use of both diary exercises and participant-observation was intended to mitigate these biases, enabling cross-referencing and in-depth probing into gaps between what participants said and what they did. Third, because we only studied those recovering from DE, our analysis cannot address the experiences of those actively experiencing DE. This gap could affect the generalizability of our analysis to all stages of the DE process. A study involving participants with active eating disorders would have changed our methodology and de-escalation procedures, and likely the research findings. It is unclear, based on existing research, whether participants actively experiencing DE would have given more or less specific insight into online harms or misinformation susceptibility (for example, due to the denial many in active DE experience), but our findings almost certainly would have differed in some way. It is possible that digital diaries would have been more illustrative of youth DE online ecosystems, while interviews would have been more fraught and potentially less reflective.

Finally, intervention design was conducted in partnership with a technology company and was therefore limited to testing interventions on the company’s platforms. We showed interventions on two types of surfaces where youth frequently seek and encounter information: a search engine and a video platform. This excluded other popular platforms such as image-sharing platforms, which many participants reported using.

2.6. Research Ethics

Our study went through a systematic internal review and approvals process to ensure that participants and parents of minor participants provided informed consent and were protected from undue risk. We followed human research ethics protocols to ensure participants fully understood the study and their ability to withdraw from it. When interviewing minors, we ensured there were always two adults present. A parent/legal guardian completed consent forms on behalf of their child and provided additional verbal consent. We also verbally explained the study process to both the minor and the parent, confirming they each understood the study. Each participant received USD 250–500 as a thank you for their 5–6 h of participation. This reflects best practices from our local research partners.

To design our research, we conducted a broad, interdisciplinary literature review, interviewed domain experts, and sought external review, following guidance on research with at-risk users (Bellini et al. 2024). We worked with a licensed psychologist specializing in DE to review prospective participants’ screener responses, to confirm that we excluded participants whose responses indicated risk for re-traumatization, and co-designed a de-escalation procedure. Participants were assigned pseudonyms during data collection (different from those used in this paper) and personally identifying information (PII) was removed from interview and diary data.

The results presented in the next section pertain to Study 2. To highlight differences between both studies, we first discuss the findings from Study 1.

3. Results

In this section we detail the findings from Study 2, which focused on youth eating and health information, explaining the challenges participants encountered and the information sensibility practices they used to manage these challenges. We indicate where and how they relate to Study 1 findings (

Table 3. Study 1 findings.

Challenges	Practices	In Study 2
Information Overload Too many articles, facts, opinions and ads, often too wordy, too long, and too negative	Surrogate Thinking Finding a trusted go-to source and relying on the source’s opinion	Yes
Misrecognition Lack of ability to speak to their generation, seen in tone, lack of controls, vibes	Everyday Experience Privileging on-the-ground crowdsourced reporting, lived experience, first-person narratives	Yes
Social Error Fear of making mistakes, taking unpopular positions, or actions that lead to social harm	Crowdsourcing Credibility Using comments, likes, and other peer discussions to evaluate content	Yes
	Good Enough Reasoning Limiting effort to finding just enough—a few facts, numbers, whatever is easy—and then moving on	No

), which analyzed youth information-seeking (and encountering) challenges and practices more broadly. In the last subsection, we discuss how content generally could not be classified into “healthy” and “harmful” categories—these were context- and participant-specific.

3.1. Challenges

While Study 1 participants felt a sense of information overload, Study 2 participants felt a sense of information overgeneralization or misrecognition (

Table 4. Study 2 findings.

Challenges	Practices	Study 1 Equivalent
Misrecognition by Overgeneralization Sense that institutions do not understand or serve their specific health needs	n = 1 thinking Assessing health information based on personal testimony of like-bodied people online rather than probabilistic measures	Surrogate thinking, everyday experience
	Distributed Trust Trusting identity- or interest-based community opinion (sourced via likes and comments) rather than centralized institutional authority	Crowdsourcing credibility
Individualistic Conformity Desire to conform to social norms, but needing to appear like conformity was not important	The “good life” Signaling “holistic health” rather than thinness, but still requires performance of bodily conformity to hard-to-attain cultural norms	Response to fear of social error
Meaningful Connection Craving deep connection with others and a sense of social belonging	Getting offline Privileging online over offline sociality	Response to general desire for social belonging (over truth seeking)

). While Study 1 participants also felt a general sense of misrecognition, the misrecognition that Study 2 participants felt was specific: that institutions, specifically medical institutions, did not recognize and serve their specific bodies. They sought information and trusted information they encountered based on how well they felt it recognized them.

Study 2 participants also used their engagement with algorithms to make sure that the information they encountered online was increasingly tailored to them, attempting to solve both information overload and misrecognition challenges. In alignment with Study 1—which found that participants tended to encounter, rather than search for, information—Study 2 participants wanted to actively search for tailored eating and health information less, by training their algorithms to deliver tailored content to them.

The fear of social error also took on a more specific form in Study 2—we term the specific version of this challenge individualistic conformity. Participants felt afraid of being judged for trying too hard to conform—to look “skinny” or like they were trying to cultivate an online persona—but knew they needed to do so to fit in and experience positive social feedback about their bodies and practices. They described it as feeling like a tightrope walk, where social media needed to be carefully cultivated to avoid social sanction in either direction.

Study 2 participants also felt an overarching need for meaningful connection, mapping on to the general motivation driving participant information sensibility practices in Study 1: the desire for social belonging. In the sections that follow, we describe the information sensibility practices that Study 2 participants performed in response to these challenges, connecting them to Study 1 information sensibility practices where relevant.

3.2. n = 1 Thinking

Study 1 found that youth value personal testimony over institutional authority, using personal relatability and authenticity as key trust heuristics. We termed this heuristic “surrogate thinking” (finding a trusted go-to source and relying on that source’s opinions) and “everyday experience” (privileging on-the-ground crowdsourced reporting, lived experience, and first-person narratives). Of the trust heuristics discovered in Study 1, Study 2 found youth used surrogate thinking and everyday experience most when evaluating eating and health information. Participants assessed social media content creator credibility based on how similarly and authentically creators felt to participants’ personal experience. Participants were largely not trying to become experts on nutrition or fitness, but experts on their own bodies. The wider generalizability of health information was not important: their goal was to find health advice that worked for at least one other person and could therefore potentially work for them.

We term this “n = 1 thinking”5: when youth assess health information based on personal testimony of like-minded, like-bodied people online rather than probabilistic measures. In part due to youth’s dwindling trust in health authorities and their comfort with online spaces, they sourced credible eating and health information via the “real” experiences of individuals online.

Participants gave concrete explanations for this evaluative process. They cited that both experts and non-experts have begun to acknowledge that universalized Western medicine lacks all the answers, particularly for nutrition and wellness. Further, many participants thought about eating and health through the lens of identity and were wary of how science had historically harmed and pathologized marginalized people like them. Many explained that health information is not universal—it is based on white, male bodies—and that therefore the advice of someone similar to them would be more likely to work than medical advice (e.g., that a young Black woman’s advice is more likely to work for me, as a young Black woman than NIH guidelines). With n = 1 thinking, participants who felt marginalized by generalized medical advice or mainstream images of fitness could access potential answers to their health goals. They spoke about “trusting their gut”, expressing that eating and health-related information had a unique relation to the individual body and that the consequences of “getting it wrong” felt higher than with other kinds of information.

When watching a video about someone else’s experience with a new diet, Lacey (23, US) said she did not worry if that person’s explanation did not sound plausible or stand up to scrutiny: “I might not believe the science behind your explanation but I believe your experience is legit. I accept that some things just don’t have an explanation”. Shivani (24, India) also expressed a need for authenticity, which content creators she watched signaled through relatability, that they are “just like you”. For Jewel (26, US), “Knowing to trust individual people is the important thing”: she finds beauty in the “moment of recognition that I know what the person is going through”. Amira (23, India) was drawn to Tahajjud prayer testimonials, specifically first-person accounts from individuals sharing anecdotes about what they prayed for and how the prayers manifested over the months of their prayer practice.

If a diet or workout did not work, it meant that it was not right for participants’ bodies, but not necessarily that the diet/workout itself was flawed. The faith in one person’s experience (n = 1) still held. After being diagnosed with a medical condition Amulya (23, India) “researched” on image- and video-sharing platforms to become an expert on what her diagnosis meant for her. She watched videos from a creator, hoping for weight loss tips she could follow to manage the condition. When the method did not work, she said: “I didn’t lose 10 kgs in 10 days … so it just didn’t work for me. But it could work for somebody else”. She needed to evaluate content “according to my body type and allergies”. Sascha (24, US) said: “everyone has different genetics, different bodies, and different lifestyles” and that if a content creator is “a teeny tiny influencer or a buffed-up gym bro, I’ll just keep scrolling”.

Brad (22, US) said that both he and his girlfriend would be deemed overweight in “doctorese” and pointed out that the BMI6 is not a reliable measure of health. For Brad, skepticism of the medical establishment went hand-in-hand with prioritizing personal experience: “I haven’t experienced health problems because of what I eat”, which was, for him, a far better indication of how healthy his relationship to food and eating was than his BMI.

Participants verified sources by checking if a creator was like-minded and like-bodied—if a creator met this criteria, their singular experience could be proof enough (n = 1). Study 2 participants asked the following questions when encountering health information online:

• Does the creator think like me?
  • Do we have similar backgrounds?
  • Do we share the same politics?
  • Do we have similar gender, racial, and class identities?

• Is their body like mine?
  • Do we have similar body types?
  • Are we at the same place in our health journeys (or could I arrive at theirs)?
  • Do we have similar symptoms?
  • Have we responded in similar ways to the same diet or fitness trends in the past?

Participants found sources of information online from like-minded, like-bodied communities that they lacked offline—especially people of color (POC), transgender (trans), and disabled participants. But online, participants could choose to be part of communities where they felt included. Korean-American Jinny (22, US) said that in the suburbs, the popular girls were always white and skinny. Online, Jinny followed creators who had bigger bodies and were healthy and happy at their size. She said: “No one trusts the medical industry. I did [fitness app] and I was not going out, no sugar, no pasta, and it was depressing”. Shivani (24, India) felt ostracized because of her weight: “fat is unacceptable in Indian society”. Shivani found a depression meetup group to talk to others like her. Jean (23, US) grew up in a conservative area where they did not know anyone else who was trans. Jean used image- and video-sharing platforms to learn about top surgery and body-neutral workouts. Taera (19, US), a woman of color, lived in a white community in an outer borough, and her parents body-shamed her. Online, Taera found like-minded fans of black cottagecore.7 Zara (26, US) felt exhausted navigating subways with a cane and did not know any differently abled people like her in real life, but found supportive communities, like differently abled hiking groups, online. Even if it was just one other person who experienced “success” in losing weight, our participants were willing to emulate them, in hopes that the advice would yield similar results for them.

For participants who felt under-represented in their offline communities, social media was a space where they could feel connected to and represented by people like them. Ariana (24, US) had saturated her feed with “healing Black girl content”. She shared that encountering Black representations of health via the algorithm, which she lacked in her mostly White community, felt healing and healthy. As Jean (23, US) said, “I have to wade through a lot of ’macho masc content’ to get to the type of trans bodies I relate to.” They used search terms like “queer workouts” and “trans strength training” to find the right fitness creatorss in preparation for their top surgery. They checked for similar mind and body to determine credibility. Lower-income participants in both India and the US found that video content teaching people how to be healthy without money was validating and useful.

Indian participants used video-sharing platforms for localized voices on health and eating—where the local became a useful proxy for like-mindedness and like-bodiedness. Participants noted positive differences between “global” esthetics, notions of status, performative photos, and aspirational symbols of class and “localized stories” representing local and niche identities, place-based vlogs, and everyday esthetics and local codes of “truthiness”. Some Indian participants felt that religious approaches to working on oneself—like those in the language of homeopathy, Ayurveda, and Unani—felt more trustworthy and relatable. When participants accessed localized results that felt credible and immediately helpful, they felt good. But when participants, like Arjun (23, India), could not find an answer, they dove back into social media and n = 1 thinking to find creator surrogate thinkers to tell them what to do (see Figure 1).

Personal relatability played a large role in trust. Shivani (24, India) said that “Finding relatable content always helps. Seeing people on [video-sharing platform] going through the same things made me feel less alone and more normal.” Sanjay (21, India) said that “I listen to [creator’s] interviews on [video-sharing platform] because somehow his experiences and my experiences are the same… I sort of thought that I have a link with his experiences, so he’s kind of my mentor.” Sangita (20, India) sought out exercise tips from “authentic” people: “If someone is suggesting something to you and it has worked for them, it gives a little bit more authenticity to it… I don’t want experts and their advanced workouts, you don’t know which one is authentic or if it will work for you.” For Sascha (24, US), if an creator says “this is what happened to me, this is how I changed … it gives you more background and I can see if and how they bettered themselves.” Krish (21, India) shared about the like-minded and like-bodied creators he followed: “if they can do it, I can do it too”.

This valorization of personal testimony exemplifies a larger shift participants expressed about social media. Instead of a social space to catch up with friends they knew offline, they used it as a personalized source of information fed by an algorithm that knew them intimately. They were generally comfortable sharing their own photo-sharing posts with researchers, but were hesitant about sharing their algorithm-suggested content feeds, because those feeds felt deeply personal. Many expressed caveats like “OK this is a weirdly specific interest but…”.

N = 1 thinking helped historically marginalized participants, but could also lead to misinformation susceptibility and testing out harmful remedies with the young person as the test subject. When using n = 1 thinking, nothing can be proven false or unlikely. Conversely, many participants did take a semi-scientific approach to n = 1 thinking, stringing together data points from a series of individuals who passed their similarity checks on social media to determine whether medical advice might work for them. The main risk with n = 1 thinking is that participants were not doing any follow-up research to what they heard online; rather, they immediately implemented the advice in their own lives. There was no friction to encourage them to vet the health information before copying someone online. Another risk is that n = 1 thinking encouraged “magic bullet” thinking (e.g., believing that they could get ’shredded’ in 2 weeks if they just followed the same diet and exercise program), potentially leading to harmful diet and exercise practices.

3.3. Distributed Trust

Participants relied on what we term “distributed trust” to crowdsource credibility, an information sensibility practice identified in Study 1, observable in the eating and health specific context of Study 2. Participants used this distributed trust heuristic by monitoring likes, comments, and reviews, particularly in interest-based communities, which felt pre-vetted and authentic. Shifting from institutional trust to distributed trust, participants called upon communities’ decentralized authority to determine validity, because approval of the crowd signified credibility. Participants’ already beloved online communities provided anonymous reassurance and an intimate sense of trust at scale.

Participants specifically used this distributed trust heuristic from communities online where they felt a sense of social belonging. Jinny (22, US) thought she had missed her chance to meet BTS8 member, Jungkook, when a karaoke bar claimed he made a surprise visit to a BTS fan event. To confirm this claim, she crowdsourced her online BTS community for consensus: “I looked at the comments and asked my other friends, and everyone thinks that venue was known for lying about this kind of stuff.” Jinny trusted the online BTS communities because “it’s only ARMYs [committed BTS fans] who reply”. She knew she had shared values and could read shared social codes in the space, increasing its trustworthiness. She repeatedly experienced and participated in the community’s social practice of working to decipher information—giving her a sense of shared responsibility for making sense of what goes on, rather than an individual sense of responsibility. Making false claims in the space was also risky, because it meant you could be removed or socially ostracized, increasing her trust in the crowd.

Jheel (23, India) saw the same messaging (about the type of diet and exercise dancers follow) repeated by many bhangra9 creators she followed: “I don’t feel like they’re saying anything wrong ‘cause I’m hearing it from every YouTuber”.

Participants also used distributed trust as a second check for verifying the credibility of a source—in tandem with n = 1 thinking—asking:

Do they think like me? [n = 1 thinking]

Is their body like mine? [n = 1 thinking]

Does this community of people like me trust it? [distributed trust]

For example, a popular creator’s diet videos (Figure 2) had more than 2.4 million views when he claimed drinking coffee and lemon accelerated weight loss. The creator showed his trim body in his normal-looking kitchen as evidence of the diet trend’s success (n = 1 thinking). He gained the faith of followers by appearing as a more familiar, local, and “relatable” source.

Participants showed how comment sections were used to debate the legitmacy of diet trends and were therefore useful for deciding whether to trust a creator (Figure 2). Commenters shared their doubts, personal successes, and even admiration of the creator’s body, creating an interactive forum to determine trustworthiness (Figure 3).

Amira (23, India) used comments to vet posts about healing prayer and therapeutic action, checking to see if commenters are Muslim like her and believe in the practices being recommended by the poster. She once reached out to a woman commenter on social media to see if she was real and would respond—which she did—confirming Amira’s belief in the content. Accessing shared experiences of mood and emotion at simultaneous points in the video with other commenters made watching this content deeply therapeutic for Amira.

Brad (22, US) also found a relatable surrogate thinker through his existing communities’ advice, whose popularity in those pre-vetted communities made the creator feel trustworthy (Figure 4). Suggestion algorithms then led him deeper into similarly fringe, and potentially dangerous, eating and health-related communities—which associated eating raw meat with a particular form of masculinity, related to sexist political ideologies (Muller et al. 2024). This led him to both follow their eating and health advice and start building a health-related political identity around these communiess’ principles.

Participants were skeptical of institutional trust and trust in the information literacy trust heuristics they learned in school. As Jewel (26, US) explained:

I was told not to use Wikipedia as a source. Then I was told to look up who is publishing the piece, and to ask myself about the author’s intentions and to critique the reading. But this was flawed because in my school, [US news outlet] was cited as a valid source but the [US news outlet] is neoliberal propaganda. Also, the way history was told was accurate in terms of events but not in a politically critical way … so now I am reframing my relationship with sources. I’m asking myself, ‘Who has authority?’ For example, “SuperSize Me”—it’s about the food we consume and how we should think about obesity as a moral issue. Obesity is a problem, but not how we are talking about it, as if underweight is the opposite of obesity. But this is a risk factor more than obesity. If you look at the J curve of health. You are at a higher risk post-surgery if you are too skinny vs. obese.

Jewel did not generally seek out health information, but nevertheless consumed a lot of it—she said that “I rely on the algorithm to direct it to me.” Her favorite content creator “can talk about how systemic oppression like ableism, fatphobia, transphobia, etc., impact the ways we exercise and the ways we diet”. Her negative experiences with doctors during her transition made her critical and skeptical of medical advice from doctors. For example, doctors told her that she would not experience wetness after bottom surgery10—but in her experience this was false, and she found many people online who also personally experienced wetness following surgery.

Using distributed trust, youth exercise less faith in single institutional authorities, and instead, rely on crowdsourced community knowledge.

3.4. The “Good Life”

Participants employed coded vocabulary to discuss and share DE practices, focused on the “good life” rather than “thinness” or other explicitly body-specific descriptors. Participants shared that it was taboo to have anorexia or to diet, but common to broadcast “lifestyle” achievements on social media to signal that they were living “the good life” to others. In both India and the US, this “goodness” was signaled through body-based eating and health practices, but not stated explicitly as such. Rather than signaling a cultural shift away from body-based moral judgments, the “good life” appeared to be a new way of framing the longstanding judgments and extreme expectations youth across generations have faced about their bodies and appearances.

Participants did not reference “anorexia” or “bulimia” when recounting their restriction/binge cycles. They talked about “energy”, “mental health”, “feeling good”—but their social media feeds reinforced the idea that certain bodies fit the esthetics of “the good life”. They felt pressured to be constantly performing this good life online, which many found harmful. Participants felt like performing that good life also meant that you were a “good person.” As Franny (24, US), articulated:

I thought to be a worthy person you have to have this body. Generally, I was doing a lot of restricting. I was a swimmer, and I would swim an hour and a half in the morning and afternoon. My mom wouldn’t allow me to eat afternoon snacks. This led to a misconception of how much food I needed or how to listen to your body.

Participants found that language about thinness being taboo did not actually remove the need to look thin, but instead added another layer of signification work to social media posts: you had to look “healthy”, but not look like you cared about looking thin. Language of holistic, “balanced” mental health and well-being replaced language about thinness.11 Ariana (24, US) said she was “just trying to be healthy all across the board”. Jivika (22, India) expressed: “You shouldn’t be too fat or too thin, I just want to maintain.” Johannes (25, US) said that “Being healthy means being conscious. If one is overweight, you have to take a look at how you feel mentally”.

For participants, having the “good life” required you to broadcast it. It was not enough to have it all—you had to also show that you had not given anything up to get there. In the US, taking care of one’s health was a value, and a fit body signified that one had the “right” type of life to maintain it. Chase (24, US) quit his strenuous job because he could not perform the right kind of life: “It didn’t leave me time to get to the gym or go hiking.” Now he does calisthenics and yoga “obsessively” and dreams about being a fitness creator. Chase described how he obsessed over looking fit with ease on social media, and also had to perform that he was not working too hard to achieve the perfect body and work-life balance.

Fit bodies were a marker of “balance” for Indian participants. In India, participants openly talked about achieving “balance” in diet and exercise. But looking like you were achieving balance took extreme work. Krish (21, India) said he worked on his body—“but not too much”—and his startup in tandem. Dhruv (24, India) believed health was about being motivated to slowly, slowly (“dhire dhire se body acha ban jayega”) and continuously work toward weight loss and fitness goals and slowly cut out “bad stuff” (“dhire dhire se khatam ho jayega”). He said that if you have the process and routine, you will see the results without being stressed (“tension mat lena”). Women participants expressed that they should not be too fat or too thin but, as Isha (23, India) put it, “curvy in a good way”.

As a result, US and Indian participants expressed that appearance indicated how one is living life. The body was grounds for judgments that affect what one was granted and denied. For participants, pursuing this “good life” image was not irrational—performing it successfully with the “right body” came with access to real social benefits. Brenda (24, US), for example, was moved to the front of their dance class when they lost weight. Mohammed (19, India) commented that “everyone is attracted to the steroid type of body”, that women he was interested in only seemed interested in men with that body type. After losing weight, Amira (23, India) noted that “all clothing suits me—girls should be a proper weight and look good.” Sangita (20, India) detailed how she was more socially accepted and received less criticismwhen she lost weight: she said you have to “keep yourself in a way that they can’t comment on”. Sangita, her mother, and her boyfriend all believed that Sangita would be happier if she could conform enough to social standards to keep people’s comments at bay.

Conversely Jinny (22, US) described how she “didn’t get into a club because I’m not a skinny white girl”. Taera (19, US) recounted comments on their posts: “All of that sabotage [fat shaming comments] would be there and I would swallow myself with hate.” Shivani (24, India) stopped going out altogether, because “People judge me for being fat. I’ve seen it in their eyes, when they ‘x-ray’ you—from people who don’t even know me”.

Chase (24, US) articulated that content removal would not change this: “It’s not the content that’s toxic. It’s the comments on people’s bodies”.

As Lester (2021) argues: “The motivation behind an eating disorder is ‘how do I exist as a good person’ for people who may feel like they aren’t valued as a full person.” Youth recovering from disordered eating were not alone in pursuing “goodness” in this way. Participants were all trying to figure out how to be and appear like a “good person” through eating and health. They were acutely aware of how their self-presentation on digital platforms invited outsiders to judge their lives and mediated access to benefits.

3.5. Getting Offline

Finally, participants found supportive communities online, but ultimately believed that the “good life” was lived offline. As discussed, for participants who felt underrepresented in their offline communities, online spaces felt more accessible and healthy because they saw their bodies represented—but those participants still did not want to live life online.

All participants expressed a gut instinct that time they spent on social media was not good for them—that they should be “out there” in the “real world”. Taera (19, US) expressed:

Staying on [video-sharing platform] is a hindrance to my life because of lack of sleep and the bad feeling of not listening to my intuition… I got addicted to it so easily. It’s so much fun learning hacks to your life, but that time could be spent different. I can feel my intuition that I need to get off it and I can’t. I will be on it until 4 or 5 in the morning.

BTS online fan communities fostered in real life (IRL) connections for Taera. She met “women of all races, sexual identities, body types, and walks of life. I could be different, and we are all worthy and special”.

Jean (23, US) said “[video-sharing platform] makes my brain feel like death. I have to delete it when I get too scrolly.” Nikhil (21, India) said “During the lockdown, I felt no energy and spent so much time on my phone late at night. I had a ‘self-realization’ that I needed to get out there and change my life”.

Broadcasting the “good life” on social media ultimately did not feel like the real good life. Participants sought offramps from online to “real life.” They valued and craved in-person interactions, especially after the pandemic. Priya (24, India) said (Figure 5):

I had a mild clinical depression but I can’t tell [my parents]. They would blame me. But I searched on [search engine] for meditation and found [program]. It’s a happiness program where you go in person and do meditation. I go with my dad.

Shivani (24, India) craved “real interaction” (Figure 6), expressing that:

Videos are a one-way community: you can just hear the other person and not talk to them. I found the [anonymized] app where you have a group call and have someone to talk to. I realized that people out there have similar problems, or problems bigger than mine.

Sanjay (21, India) used [search engine] to find local meditation groups and videos posted by those groups. When participants were able to translate healthy aspects of their online lives into meaningful connections offline, they felt better (see Figure 6).

These offline connections are ultimately what made participants feel a holistic sense of community and acceptance.

3.6. Beyond Healthy/Harmful Content Labels

We started Study 2 by asking how youth distinguished helpful from harmful content, asking them to make collages of both (Figure 7 and Figure 8).

But it became clear that delineating between universally helpful and harmful content was not their key goal, nor was it analytically useful. Deeper needs for personalization and belonging drove their eating and health behavior, and helpful/harmful distinctions varied based on personal experiences. The same content could be helpful or harmful depending on the context, and where the participant was in their journey.

As Franny (24, US) put it: “I was never not bingeing and restricting… But during the pandemic, I realized it wasn’t just periods of eating really healthy and falling off the wagon… I realized there were two cycles that continually happened. It was one big pattern of bad”.

Content unremarkable or helpful to one person could be triggering for someone else. Watching shopping “hauls” on video-sharing platforms during the pandemic reinforced the restriction/binge eating cycles that Franny had learned from her mother and grandmother. But video-sharing platforms later helped her find peer resources for cooking healthy on a budget, which did not trigger her.

Jean (23, US) explained that for them, recovery content, which many participants found incredibly helpful, was a pathway into DE practices: “This recovery stuff is more triggering. When I was 12, it helped me learn how people were engaging in [anorexic] behaviors”. Jean believes they know the difference on social media between recovery accounts that are going to be helpful and harmful after working through their contextually specific triggers in therapy.

Anil (22, India) was inspired by professional athletes’ bodies, content which other participants found “healthy”, but this led him down a path to an early heart attack (see Figure 9).

Participants wanted to train their algorithms to ensure that they only encountered “healthy” content in their social media and information ecosystems. They specifically wanted guidance on the right vocabulary to individualize their online searches and algorithms. Amira (23, India) struggled to cope with her “anger issues”—she found solace in prayer testimonials on video-sharing platforms, but lacked the vocabulary to connect her spirituality to mental health sources. Sammy (25, US) had an “eating disorder thing” that he could not name—but he knew what it was not. When he went online to find out more about his symptoms, he found that medical diagnostic content could be triggering: “When I searched it up those two ones, bulimia and anorexia, were in my results, and it was very intense. I can’t really find other ones”. Dhruv’s (24, India) social media was filled with content about “paper skin”, a term used to praise muscular, veiny bodies. He struggled to find more about where this term came from, why it was popular now, and understand whether it was achievable for him. Krish (21, India) wanted to try out a new fat-burner pill, but searched for videos to crowdsource reviews and found they were negative. It was helpful for him to know that he was “not the first person to think this way”.

4. Intervention Testing

Based on Study 1 findings and Study 2 preliminary analysis, we created interventions to test in follow-up online interviews with Study 2 participants (N = 42), using design mockups (Figure 10, Figure 11, Figure 12, Figure 13 and Figure 14). We coded 98 unmet needs articulated by participants and categorized them into five clusters, which served as the basis for intervention concepts.12 These five clusters were: discernment, diversification, social learning, self-protection, and personalization. Each intervention corresponds to one cluster (

Table 5. Tested Interventions and Corresponding Participant Needs and Practices.

Intervention	Needs Cluster	Practices (Study 1 or 2)
View As See through the eyes of trusted creators	Discernment	Surrogate thinking (1) Good enough reasoning (1)
Explorer Remix Browse outside your algorithm	Diversification	Surrogate thinking (1) Good enough reasoning (1)
Search Together Search results tailored for your communities	Social learning	Crowdsourcing credibility (1) Distributed trust (2) Getting offline (2)
Personal Safe Search Filter out unwanted search terms	Self-protection
Location-Based Search Filter search results specific to your region	Personalization	Crowdsourcing credibility (1) n = 1 thinking (2)

).

We detail aggregated participant reactions to the interventions, describing the two negatively received interventions (View As and Explorer Remix), the two mixed-reviewed interventions (Personal Safe Search and Search Together), and the one positively received intervention (Location-Based Search). We then summarize participant reactions to all interventions.

4.1. View As

The discernment cluster summarizes needs participants articulated around understanding whether creators and the information they shared were trustworthy. Needs included confirming whether creators were real or a character, understanding their “rabbit-holing” pathways, knowing whether content has been edited or not, and understanding whether content is “quality” or just popular.

Since Study 1 findings also indicated that participants wanted surrogate thinkers to help navigate information overload, we theorized that an intervention allowing participants to see their trusted creators’ online ecosystem might help participants with their discernment needs while supporting their preferred trust heuristics. The View As intervention (Figure 10) would allow users to see what their preferred creators saw on their social media feeds (no actual creator data were used in the design mockup). Aman (21, India) liked the idea: “it’s useful, helpful because I want to know what he [creator] is watching. I want to know more about them—what they’re into beyond what they show to the world—it’s like a background check”. But Brenda (24, US) thought it was “too invasive”. Most participants were not interested in knowing about the creator’s online life to this extent, but they did want to know whether the creator made money off of the content the creator posted.

Figure 10. View As intervention mockup. The user is able to see what creator Isabela Smith’s feed looks like (left) compared to the user’s own feed (right).

Figure 10. View As intervention mockup. The user is able to see what creator Isabela Smith’s feed looks like (left) compared to the user’s own feed (right).

The failure of this intervention could be explained, in part, by the information sensibility practices discovered in Study 1—that is, that participants only wanted good enough information to make a decision about a creator, and “View As” led to information overload (more information than they wanted or felt they needed). Once a creator had become a surrogate thinker for them, they did not feel the need to learn more about the creator beyond what they followed the creator for in the first place (e.g., for specific types of news or health information).

4.2. Explorer Remix

The diversification cluster aggregates articulated participant desires for new informative content. Needs included hearing perspectives on health and eating beyond those of their family members, how to “reset” one’s algorithm on social media, immediately seeing other points of view from the video or content they were consuming (for some participants, beyond accepted medical science), content helping participants think critically about the eating ideas that they grew up with, and seeing creators with different body types.

In response, Explorer Remix (Figure 11) would allow participants to browse separately from their usual recommendations, seeing “outside” their algorithms. Participants were largely uninterested in seeing content outside of their usual interests. Jivika (22, India) said:

I don’t see why you’d want something outside of your feed. My brother is really crazy about cars, and likes watching videos about cars, but I’m not open to that knowledge. I’m not that much into cars.

Amira (23, India) echoed a similar sentiment:

There is no need to explore different types of content–there is no need to see everything. If you’re getting bored, you can use this to find something new, but I typically switch to a [Netflix] series.

Participants in both Study 1 and Study 2 valued trusted, interest- or identity-based groups (distributed trust) and creators (surrogate thinking), looking for markers of social (crowdsourcing credibility) or bodily (n = 1 thinking) similarity to evaluate information. Outside of their trusted, algorithmically moderated information ecosystems, these information sensibility practices are harder to implement. It therefore makes sense that stepping outside of these spaces via interventions like Explorer Remix would hold little interest or value to them, especially given how much care and effort they put into training their algorithms to enable them to encounter information specifically tailored to them.

Figure 11. Explorer Remix intervention mockup. The user can toggle on the “Explorer Remix” mode in their settings, seeing content different from that suggested on their main profile feed.

Figure 11. Explorer Remix intervention mockup. The user can toggle on the “Explorer Remix” mode in their settings, seeing content different from that suggested on their main profile feed.

4.3. Search Together

The social learning cluster summarizes a set of needs around discussing health information with like-minded and like-bodied people, community-based ways of searching for and vetting online information, and developing deep and meaningful online and offline relations. The Search Together intervention sought to meet those needs.

Search Together would show results tailored to participants’ online communities, letting them see search terms and results that people in their communities searched for (Figure 12). Participants liked seeing these, but ultimately wanted more interpersonal interactivity. As Zara (26, US) put it:

“It would be cool if these search groups were also connected to something like a forum board where people could also post. This makes me feel like these are folks I don’t just want to search with, but what are they thinking in the realm of this topic.”

Figure 12. Search Together intervention mockup. The user can choose to see what search terms (e.g., “running support”) people in their selected communities (e.g., “Learning to run”) used.

Figure 12. Search Together intervention mockup. The user can choose to see what search terms (e.g., “running support”) people in their selected communities (e.g., “Learning to run”) used.

Malia (26, US)—like many participants—described a desire to move beyond searching for information together to actually getting together. She wanted something that “takes you offline and into real life”. Aneet (21, India) said: “I would definitely use the community results because if there are people going through the same thing, and sharing their knowledge and results, then I can use that. Those people could advise me”.

Interestingly, participants’ desire for getting offline (Study 1) took precedence over their information sensibility practices of distributed trust (Study 2) and crowdsourcing credibility (Study 1). Seeking meaningful connection offline (Study 2) could be a solution for the challenge of information overload (Study 1), potentially more attractive to youth than online solutions.

4.4. Personal Safe Search

The self-protection cluster summarizes a set of needs around avoiding unhelpful online content. Needs included avoiding recommended content centered around body image, “What I ate in a day” and “this is what worked for me” videos and images, and labeling of food as “good” or “bad”. The specific information to avoid varied from participant to participant, requiring customization to individual needs, but the desire to avoid some kind of content seemed broadly shared. The Personal Safe Search intervention attempted to address this self-protection cluster.

Personal Safe Search would allow participants to filter out unwanted terms from their results (Figure 13). US participants liked this feature, but Indian participants were generally uninterested in blocking content. Ariana (24, US) said: “I love that you can search with intention, but more than just blocking out content, I like that you can hone it down, so this helps streamline that search.” Sanjay (21, India) wanted to use it for “healthy” ingredient searching: “I like this ‘exclude custom words’ option—if I was making a healthy smoothie, I’d exclude ‘protein powder’ to find protein shakes that use natural things.” Pooja (23, India) was uninterested in blocking, instead wanting something that allowed her to search more easily for what she wanted: “I would rather have an interface where I can get options not to block, but to specify the thing I’m looking for”.

Figure 13. Personal Safe Search intervention mockup. The user can choose to exclude terms from their search results.

Figure 13. Personal Safe Search intervention mockup. The user can choose to exclude terms from their search results.

This desire for tailored content enhanced our understanding of information overload (Study 1) as a challenge. Information sensibility practices to reduce overload did not rely on reducing content in general—they relied on reducing general content. Participants did not want less information—they wanted information tailored to them.

4.5. Location-Based Search

The personalization cluster aggregates participant needs for customization of health and eating content to their local context. Participants wanted help finding and encountering information more socioculturally tailored to them, rather than general health and eating information, as well as help applying that advice in their everyday life.

Our corresponding intervention, Location-Based Search would allow users to filter search results specific to their region. In the mockup (Figure 14), we left the definition of “your region” open-ended, and asked participants what definition of “region” would be most useful (e.g., country, state, “within 5 miles/kilometers”). Participants liked the idea of searches based on region, but wanted to be able to adjust location on a sliding scale based on what they were searching for (rather than specifying what fixed scale would be useful in general). Sanjay (21, India) said: “Whatever I’m searching, I want results in my region—the content should be available in India. Things are different in India, so there’s no point in searching abroad first. I want to search regionally first, and then globally if I can’t find anything”.

Figure 14. Location-Based Search intervention mockup. The user can choose to show search results from their region.

Figure 14. Location-Based Search intervention mockup. The user can choose to show search results from their region.

Maya (23, India) speaks Hindi mainly, and searches for information on her medical condition, yoga, and diet content. She wanted a Hindi language filter with a caveat: “I like this idea for finding recipes or products, especially in Hindi. But it can’t return formal (shudh) Hindi; it needs to return local or colloquial Hindi that would actually be useful.” Zara (26, US) was unsure about how she would use the intervention: “it would be tricky to know when to toggle that on”.

Dhruv (24, India) felt it was important to “improve your search power”, via using specific terms and targeted queries, even on social media. He felt that in order to avoid negative content, you have to “hack” your algorithms: “there’s a pattern to search—to learn how to find healthy content, and like, and subscribe, and that will tailor your [social media].” While Dhruv recognized the need for an Indian perspective when it comes to search, he did not initially understand that this was a new concept. In fact, he said he already uses this feature via “Search nearby results” and using “.in vs. .com.” He felt that Gen Z already has workarounds for finding some version of localized results, so this felt redundant and less exciting. However, he felt like the onus was on the user to learn these workarounds—such that a toggle would make this an easier process. Dhruv emphasized that there is a BIG difference in Western and Indian perspectives, for example, on turmeric, supplements, and protein.

Place-based content felt credible because another user from the same region was more likely to be similar to the participant—participants used Location-Based Search as a proxy for finding information from like-minded/like-bodied people (n = 1 thinking, Study 2).

4.6. Intervention Testing Summary

In summary (

Table 6. Tested interventions and aggregated participant feedback. We list interventions with largely negative feedback with a negative sign (−) mixed-to-medium feedback with a plus/minus sign (+/−), and positive feedback with a plus sign (+).

Intervention	Participant Feedback
View As − See the feeds of trusted creators	Didn’t care about knowing more about the creator than they were already exposed to(other than if the creator was monetized or not).
Explorer Remix − Browse outside your algorithm	Felt that content outside of their algorithms would be irrelevant to them.
Search Together +/− Search results tailored for your communities	Liked interacting online, but ultimately wanted the intervention to help them meet offline.
Personal Safe Search +/− Filter out select search terms	US users wanted to tailor their results against potential triggers. Indian users did not have the same discourse around triggering content. Participants largely encountered information. When searching, most wanted more good information, not less bad information; they wanted to use filters to help find better content.
Location-Based Search + Filter search results specific to your region	Liked opting into searches based on region, but wanted to be able to adjust location on a sliding scale. Felt someone from the same region was more likely to be like-minded and like-bodied.

), the Location-Based Search intervention was the most positively received intervention. It gave participants a sense of control and agency over their search results and aligned with their existing practice of trusting information from people who are “like-minded and like-bodied”, as location was seen as a useful proxy for similarity. This intervention would also likely be the easiest to implement, though may be limited by both content creators’ and users’ location and privacy settings. Careful consideration and user control over the balance between localization and privacy is crucial to the successful application of an intervention like this.

The Search Together intervention received mixed feedback because, while participants liked the idea of interacting with their online communities, they ultimately wanted help to take those interactions offline. The Personal Safe Search intervention received mixed feedback because participants generally wanted more tailored information through filtering, not to block content (especially Indian participants). Providing basic content filters or better transparency and education about search engines and algorithmic recommendation systems could help meet participants’ personalization and self-protection needs. Peer- and AI-assisted learning techniques could also support these needs, guiding users in their interactions with online information ecosystems. Given the growing intergration of AI in tools like search, implementation seems possible (though potentially fraught, due to the propensity for both peers and AI to share inaccurate information).

The View As intervention was negatively received because participants felt it was too invasive and gave them irrelevant information. The Explorer Remix intervention was negatively received because users felt that content outside of their algorithms would be uninteresting and irrelevant. The View As intervention would be difficult to implement as it is uncertain whether content creators (particularly those invested in the use of misleading content) would consent to a View As setting. Explorer Remix would be easier to implement but, based on these findings, unlikely to be used—except potentially in the form of a “private browsing” session when participants would prefer not to have a particular session included in their algorithmic recommendations.

5. Discussion

Our findings across Study 1 and Study 2 indicated common challenges that youth face and practices that youth employ to navigate information online, which have significant implications for designing effective interventions to promote youth digital resilience. Both studies demonstrated that youth information seeking is deeply intertwined with social needs and identity formation. Participants rarely engaged in purely truth-seeking behavior; instead, they used information to orient themselves socially, define their emerging identities, and connect with like-minded peers. They also showed a strong preference for personal testimonies and relatable sources over traditional markers of authority, and they relied heavily on social cues and peer validation within their online communities to assess credibility. These shared findings point to a critical need to move beyond traditional, literacy-focused interventions and adopt strength-based approaches that leverage the social embeddedness of online information practices.

Literature on social media’s effects on youth, particularly those focused on DE, often assumes that content can be classified as “healthy” or “harmful” and effectively moderated (Ambwani et al. 2019). But our findings demonstrate that these distinctions do not exist as intrinsic qualities of content: they only make sense in contextual interactions between the user and the content. Much work on DE focuses on an individual pathology, evaluating an individual’s psychological state, family history, and reactions to therapeutic interventions, often in clinical or lab settings (Holland and Tiggemann 2016, 2017; Ioannidis et al. 2021; Ioannidis and Chamberlain 2021). To date, much intervention attention has focused on the moment of crisis, when a user is facing immediate harm. These interventions model the path to crisis as a funnel, asking how to intervene before users reach the crisis point at the end. This model imagines youth’s journeys as linear and their crisis points as homogenous. It also leads interventions toward solutions like content removals by platforms, suggesting that the best way to improve the digital resilience of youth is to grade pieces of content as inherently “harmful” or “helpful”, removing those that fall into the former category. Our findings, however, show that different youth had different journeys and relationships to content, mediated by the social contexts in which they encountered and shared information.

5.1. Summary of Design Recommendations

We found that meeting participants’ information needs meant understanding their underlying motivations for consuming and sharing content. Interventions should help youth assess whether content is helpful for them and provide them with resources. We propose supplementing content-grading interventions with interventions that support and re-channel that desire for a sense of purpose and belonging. This approach, we propose, helps support a wider range of youth (beyond those formally diagnosed with eating disorders) navigate an online information ecosystem saturated with eating and health information.

These common findings directly informed our design recommendations. First, recognizing the importance of peer networks and the prevalence of crowdsourcing credibility (Study 1) and distributed trust (Study 2), we recommend leveraging peer networks (6.1). This involves creating interventions that facilitate collaborative learning and peer-to-peer support within online communities.

Second, given the finding that participants often encountered information (Study 1) through algorithmic suggestion and tailoring (Study 2) rather than actively searching for it, we recommend developing social media sensibility (6.2). This means equipping youth with the skills to critically evaluate information sources in situ, understand creators’ motivations, and navigate social influences online. Implemented successfully, this sensibility could help them engage in their algorithmic tailoring work with more attention to the quality of sources they choose as surrogate thinkers (Study 1).

Third, acknowledging participants’ desire for meaningful offline connections and their belief that the “good life” is lived offline (Study 2), we recommend providing pathways offline (6.3). Interventions could connect youth to in-person communities and resources, such as local support groups or workshops, based on their online interests. Finally, recognizing the potential dangers of “n = 1 thinking” (Study 2), we recommend encouraging probabilistic thinking (6.4). This involves helping youth understand the limitations of personal testimonies citing everyday experience (Study 1) and the importance of considering broader evidence to enhance their good-enough reasoning (Study 1). All interventions would require careful consideration of digital privacy, safety, and security risks such as hackability, abuse, and infiltration. We discuss some of these risks.

These findings highlight limitations of solely literacy-focused interventions and emphasize the need for strength-based approaches built upon the social embeddedness of online information practices. We propose that interventions should not primarily focus on “cleaning up” the online ecosystem through content removal, but instead give high-quality information to youth in key moments when they are reaching for it and support them by helping them build the meaningful peer-to-peer relationships they crave.

5.2. Leveraging Peer Networks

Within social media, participants needed localized content and practical advice from communities they trusted. Participants also wanted communities for long-term learning and resources for putting that learning into practice. When participants accessed localized results that felt credible and immediately helpful, they felt more confident in their decision-making.

A version of the “Search Together” intervention described in Section 4.3 might help youth find supportive communities to synchronously discuss information with. When they were left to search alone, their chances of encountering potentially harmful misinformation seemed just as likely as on social media, so peer support mechanisms could help them use their existing information sensibility practices to avoid digital-safety harms. Location-based filters could also help younger users find eating and health information that feels more relevant to them.

Features that highlight community-sourced information—like ratings, reviews, and shared search histories—could help youth identify credible sources and connect with supportive peers. Our findings suggest that information should be community-sourced where possible, given youth’s reliance on crowdsourced credibility (Study 1) and distributed trust (Study 2). Disabled and minoritized participants wanted in-the-moment information to make their cities more physically accessible and guides to places hospitable for Black, Indigenous, and People of Color (BIPOC). Interventions could elevate community-based knowledge sources like @DisabledHikers, which provide up-to-date info about accessible trails.

Youth desire for identity- and interest-based communities could be supported through interventions strengthening peer-to-peer networks: researchers studying sexual minority youth found that they often join online communities to express themselves and feel less isolated (Charmaraman et al. 2021). These communities foster a sense of belonging and identity, and, when designed well, could provide peer-to-peer support, helping youth navigate mental health challenges. In doing so, these communities could also help protect youth from misinformation susceptibility, which is often increased by loneliness and other challenging emotional states (Hassoun et al. 2024).

Research indicates that users of online mental health communities are more engaged when posts are about social experiences and emotional expression (Liu and Kong 2021). Interventions could encourage users to share detailed content, express distinct emotions, and use images to enhance engagement and support.

However, there are potential risks involved in this recommendation. Over-reliance on peer networks could amplify harmful viewpoints and reinforce misinformation if communities are already entrenched in those views. The social benefits youth receive from such communities might make it hard for them to disengage and find more supportive communities.

5.3. Developing Social Media Sensibility

Participants needed a trusted way to evaluate new vocabulary they encountered on social media, especially when social media content used diagnostic words. Participants needed help in the moment directly after learning something new, where they were most receptive to new perspectives. They also needed ways to vet creators they found on social media. Useful interventions could help them successfully surface from social media algorithms, ask clarifying questions, and pause to reflect before diving back in.

Possible interventions could involve provenance tracking or a language tree of search queries which build on the vocabulary youth use on social media to better guide them to new terms and searches. Historically and globally contextualizing how a term became prominent and in which contexts could help young users understand emerging, confusing terms. Other potential interventions include tools that provide contextual information about content creators, such as their funding sources or potential biases, and prompts that encourage users to reflect on the information they encounter.

Study 1 found that participants liked having go-to sources (surrogate thinkers) that provided information they trusted. The like-minded, like-bodied checks undertaken by Study 2 participants helped them vet creators and determine whether the creators qualified as a go-to source. Rather than trying to push youth away from their go-to sources, helping them better evaluate content creators online could offer better and more impactful digital-safety support.

Social media sensibility interventions could help youth understand how algorithms work in situ, in the specific context of their social media environments: how content they see is promoted or recommended, the motivations and potential biases of content creators (Breakstone et al. 2021), and the effects of deceptive design elements on social media platforms (Costello et al. 2023; Montag et al. 2019). Interventions could also address the algorithmically mediated power of social influence online, like learning how “likes” affect perceptions of social norms (targeting their “crowdsourcing credibility” practices). Education should focus on how these dynamics and affordances could impact their everyday experience on social media platforms.

This recommendation has potential drawbacks. It still focuses on individual-level interactions, which may limit its efficacy. It could place an undue burden on individual users to engage actively with everything they encounter online, which is not feasible or sustainable—especially given the finding that youth employ good enough reasoning (Study 1) whenever possible, given the information overload they face.

Interventions in the literature that potentially counteract this risk address the surrogate thinking (Study 1) information sensibility practice. These include recruiting opinion leaders within online communities to promote positive health behaviors and counteract negative content, as peer influence can be a powerful tool for shaping online behavior (Valente and Pumpuang 2007). Using network analysis to identify go-to sources, “core pro-recovery users” that other users find supportive could be recruited or supported as behavior change agents (Wang et al. 2018).

5.4. Providing Supportive Pathways Offline

Participants valued resources that connected them to offline communities when they needed in-person resources and support. Participants wanted to connect with people offline and activate a sense of being seen and valued “IRL”. Providing youth with pathways offline using community-sourced information could meaningfully build their resilience to digital-safety harms. Developing connective online-offline interventions—like a digital community bulletin board that weights results toward local sources (e.g., in-person activities and events)—could help youth find and strengthen the offline relationships they sought based on the empowering connections with like-minded and like-bodied people they found online. This recommendation (along with the peer-to-peer network recommendation) is supported by literature on social-emotional learning, which focuses on helping youth make responsible decisions, increase their self- and social-awareness, build their relationship skills, and foster ability to manage their emotions and behaviors (Durlak et al. 2011).

These pathways could offer another source of support for their crowdsourcing credibility (Study 1) and distributed trust (Study 2) information sensibility practices, allowing youth to process information with in-person communities that meet their need for identity-based social belonging. Offline interest-based communities seemed less likely to form around extreme political viewpoints or misinformation sharing than online ones. Offline groups tended to be oriented around mental health discussions and kinetic activities (e.g., gardening, sports, hiking, knitting).

This recommendation risks excluding youth who lack access to offline resources or who prefer online interactions. More structured and interactive online communities, like synchronous online camps for younger youth, could augment this pitfall (Darvasi 2017; Ringland et al. 2016). This recommendation also has the same potential pitfalls as leveraging peer networks with regard to offline communities that are entrenched in harmful views and practices (e.g., supremacist groups, workout groups with pro-ana goals). It could also be abused, hacked or infiltrated, making it potentially difficult to responsibly implement.

5.5. Encouraging Probabilistic Thinking

As discussed, the problem with n = 1 thinking is that information cannot be proven false or unlikely. Encouraging probabilistic thinking (how likely something is to work, what percentage of people like them support it) may help to offset reliance on personal testimony. Youth need help putting n = 1 personal testimonials into broader context. Interventions could include sources that highlight probabilistic thinking (e.g., “86% of peer-reviewed studies conclude that there is not enough information to know if the Keto diet is effective or safe for anything other than epilepsy.”) or communities that vet claims (e.g., voting mechanisms in community spaces or underneath user-created content). Interventions could provide this probabilistic and voting information alongside personal anecdotes. Youth do think probabilistically—when they see n = 1 content on social media, they ask themselves, “What are the chances this would work for me?”—but more exposure to probabilistic thinking could help them make an educated guess without experimenting on themselves. One risk of this recommendation is that it could undermine the trust and connection that youth feel with relatable sources, which is a key aspect of their information sensibility.

The belief underpinning n = 1 thinking—that there is no universal right answer to most health questions—could also help protect youth from certain kinds of misinformation susceptibility if oriented more probabilistically. In Study 1, we found that the desire for certainty drove many participants toward misinformation’s definitive answers to difficult questions. The ability to accept ambiguity and a probabilistically likely, though not definite, answer helped participants not seek certainty in misinformation sources.

In Study 1, we found that participants engaged in “good-enough” reasoning—limiting effort to finding just enough information to fit their context-specific needs (e.g., data to support them in proving a family member wrong in an argument or obtaining background information on what their peers were discussing). Helping youth use this heuristic to support thorough decision making by making probabilistic information easily available could help offset n = 1 thinking. Because information overload—as discovered in Study 1—is such an issue for youth, such probabilistic information must be simple and easy for youth to access and parse.

This recommendation is especially crucial in the age of GenAI. GenAI tools—if used to generate highly personalized content that caters to individual biases and preferences—could make youth less likely to consider broader probabilistic data, particularly if such an algorithmically tailored GenAI tool becomes a surrogate thinker. Its authoritative answers and singular voice could undermine the impulse and ability to think probabilistically or consult other sources. Conversely, teaching youth to prompt GenAI tools for probabilistic evaluations could support this intervention. Given the reliance of youth on social cues, future research could explore the role of GenAI in youth information sensibility practices, as GenAI tools were not widespread during our field work.

5.6. Algorithmic Interventions

Finally, it is necessary to stress the importance of algorithmic suggestion and responsibility. While it is important to equip youth with strength-based, autonomy-preserving approaches to digital resilience, interventions should also ensure that social media platforms facilitate transparent and independent assessments of the impact of their algorithm design on youth (Costello et al. 2023). As Costello et al. (2023) argue, the focus should be on the design elements of these algorithms and the harms produced by them (e.g., infinite scrolling, social pressure and rewards, notification cadence, individualized content feeds), not just the content on the platform. While Costello et al. reached this conclusion from a legal-pragmatic evaluation, our ethnographic research produced the same recommendation: a focus on algorithmic mediation and its effect on information engagement—rather than content classification—from a youth-centered emic approach. Given the increasingly central role of GenAI algorithms in shaping online information ecosystems, future research and resulting frameworks should include algorithmically mediated engagement if they hope to positively impact youth digital safety and resilience.

5.7. Period, Age, and Cohort Effects

We hypothesize that youth engagement with information online is shaped by a combination of period, age, and cohort effects.13 The tendency to encounter information rather than actively search for it is likely a period effect experienced across age groups, but is more pronounced among younger generations, forming a generational gradient. Similarly, the desire for social belonging often outweighs the pursuit of accurate information for all ages, a period effect particularly strong in adolescents, indicating an age gradient. Adolescence is also marked by a heightened need for peer-based orientation and identity construction through information practices, as well as a stronger initial influence of family on beliefs, which shifts towards peer influence with age. The specific ways youth validate information, such as crowdsourcing credibility, are cohort effects enabled by social media affordances not available to prior generations. Furthermore, a greater reliance on everyday experience and n = 1 thinking over traditional expertise is a cohort effect potentially stemming from a decline in institutional trust during Gen Z’s formative years.

5.8. Future Work

We recommend researchers build on our findings by investigating strategies to improve upon interventions to encourage digital resilience, disseminate them to youth, and promote their adoption. It would also be valuable for future research to evaluate interventions in different geographies and digital contexts, including but not limited to the predominant forms of digital public squares that are centrally moderated (e.g., TikTok), community moderated (e.g., Reddit), or privately moderated (e.g., chat applications) (Goldberg et al. 2024).

There is also a timely need to study how large language models (LLMs) could better customize these interventions to tailor them to individual youth. For example, LLMs have demonstrated utility in personalizing educational messaging for youth—though not without serious limitations (Gill et al. 2024)—and could be used to surface probabilistic generated responses to youth queries for information to address n = 1 thinking. Conversely, the risks posed by youth interactions with LLMs should also be studied in relation to this paper’s findings.

6. Conclusions

This research underscores the limitations of solely literacy-focused or content removal interventions and highlights the necessity for strength-based approaches based upon the social dimensions of youth online information practices. Our findings suggest that digital resilience among youth can be fostered by: (1) leveraging peer networks to encourage collaborative learning and peer-to-peer support within online communities, (2) cultivating social media literacy skills to critically evaluate information sources and navigate social influences, (3) providing pathways offline, and (4) encouraging probabilistic thinking. We advocate for interventions that move beyond content removal, focusing instead on providing high-quality information to youth at critical junctures in their online journeys and supporting the development of meaningful peer-to-peer relationships. By addressing the social and contextual factors that shape online information engagement, we can empower youth to navigate their digital ecosystems with greater resilience and agency.