Introduction

The exponential growth of digital short video platforms, such as TikTok and YouTube Shorts, has transformed how information is consumed, often concentrating users’ exposure to algorithmically selected content1. This reliance on tailored content delivery has intensified concerns about information fragmentation—a condition where limited exposure to diverse perspectives distorts public discourse and narrows individual worldviews. As of early 2024, TikTok surpassed 1.56 billion monthly active users globally2, and YouTube Shorts recorded over 70 billion daily views across more than 100 countries3. The global scale of engagement highlights how these platforms have become dominant gatekeepers of information, raising urgent questions about users’ ability to critically navigate personalized media ecosystems.

Media literacy is widely recognized as essential for empowering users to critically evaluate media content. Yet, significant gaps remain in understanding how media literacy operates within the hyper-visual, rapid-consumption environment of short video platforms4,5,6. Existing research has emphasized the importance of developing visual-media literacy, given that over 80% of digital content consumed today is visual7,8. However, much of the media literacy literature remains anchored in traditional news or long-form content contexts, overlooking the unique cognitive, emotional, and algorithmic pressures exerted by short video environments9,10. Thus, how media literacy functions against algorithmically reinforced fragmentation in short video platforms remains an unresolved empirical and theoretical gap.

The COVID-19 pandemic further revealed the fragility of media literacy in high-velocity, high-emotion media environments. Even among educated users, the ability to discern credible from false information was found to be highly inconsistent, especially in developing regions with weaker educational infrastructures11,12. Recent studies13 stress that possessing media literacy skills does not necessarily guarantee resilience against misinformation in real-time crises, pointing to the need for deeper cognitive, emotional, and contextual training beyond traditional literacy models. This critical limitation remains under-theorized in short video research, where speed and emotional resonance are even more pronounced.

Although global initiatives like UNESCO’s Media and Information Literacy (MIL) framework have improved verification skills by around 25–30%14, existing interventions often fail to address the rapid, visually saturated, algorithm-driven environments that characterize today’s dominant social platforms. Current frameworks are insufficiently adapted to the immersive, swipe-based engagement patterns of platforms like TikTok and YouTube Shorts.

Generation Z, the most active demographic on these platforms, faces unique vulnerabilities. Global statistics indicate that over 63% of Gen Z users engage daily with TikTok, Instagram Reels, or YouTube Shorts15. Yet, studies show that cognitive dissonance, emotional affinity, and platform-induced confirmation biases often overpower rational media evaluation processes16. Despite increasing global awareness of MIL’s importance17,18, around 68–70% of users still struggle to consistently differentiate credible from misleading content post-training13,19, signaling major unresolved gaps in both the design and delivery of existing media literacy education, especially in the context of dynamic, short-form media.

Given the rapidly evolving nature of digital media ecosystems, there is an urgent need for interdisciplinary, context-sensitive research frameworks that address not only users’ cognitive evaluation capacities but also their emotional engagement patterns, platform-specific usage behaviors, and algorithmic personalization effects20. Manca et al21. call for “glocal” literacy frameworks—combining global standards with local realities—but most current studies inadequately theorize how these global principles interact with localized media ecosystems shaped by unique technological and cultural factors.

Thus, this study seeks to bridge these critical theoretical and empirical gaps by focusing specifically on media literacy’s role in mitigating information fragmentation within the algorithmic environments of digital short video platforms. It not only advances theoretical integration across Media System Dependency Theory (MSDT), Information Processing Theory, Social Cognitive Theory (SCT), and Dual-Processing Theory, but also addresses practical imperatives for developing next-generation media literacy interventions attuned to today’s fragmented and emotionally charged media landscapes. Specifically, this study aims to bridge these gaps by addressing the following research questions:

  • RQ1: How does media literacy affect users’ perceptions of information fragmentation on short video platforms?

  • RQ2: How does media literacy influence the diversity of media exposure and critical engagement among Generation Z users?

  • RQ3: How do technological proficiency, social influence, and platform characteristics moderate the relationship between media literacy and media engagement outcomes?

Employing Partial Least Squares Structural Equation Modeling (PLS-SEM) with a rigorously stratified sample of 473 Generation Z users in Guangdong Province, China, this study undertakes a pioneering analysis of the intricate cognitive, emotional, and platform-driven dynamics influencing media literacy outcomes on digital short video platforms. The novelty of this study lies in its deliberate and unprecedented focus on situating media literacy within algorithmically curated, hyper-visual short video environments—a domain critically underexplored despite its profound implications for global information ecosystems and democratic discourse.

By systematically integrating Media System Dependency Theory (MSDT), Information Processing Theory, Social Cognitive Theory (SCT), and Dual-Processing Theory, this research constructs a comprehensive, interdisciplinary framework uniquely tailored to the fragmented, emotionally charged, and personalization-driven realities of contemporary media consumption. Unlike traditional models anchored in long-form media or static digital environments, this study captures the dynamic, high-velocity interaction patterns that define Generation Z’s engagement with platforms such as TikTok and YouTube Shorts.

The study’s contributions are twofold. Theoretically, it redefines media literacy research by embedding cognitive control, emotional engagement, and social reinforcement within the context of algorithmic personalization—addressing major conceptual blind spots in existing media literacy scholarship. Practically, it generates actionable insights for educators, policymakers, and digital platform designers. For educators, it underscores the urgent need to reframe curricula toward real-time critical engagement skills applicable to fast-moving, visual-first media. For policymakers, the findings offer empirical grounding for regulatory initiatives aimed at enhancing transparency and promoting diversity within content recommendation algorithms. For platform designers, the results highlight pathways to develop interface features that foster critical evaluation and exposure to diverse perspectives, thereby strengthening user agency in an era of algorithmic gatekeeping. By addressing a rapidly evolving yet critically neglected media context, this study advances the frontier of media literacy scholarship and offers essential strategies for mitigating the systemic risks of information fragmentation across global digital societies.

Literature review

Media literacy: global context and research gaps

Over the past two decades, media literacy has been widely recognized as a fundamental competency for navigating increasingly complex information environments22,23. Global initiatives such as UNESCO’s Media and Information Literacy (MIL) framework, the European Commission’s Digital Competence Framework, and national programs in countries like Finland, Australia, and Canada have emphasized critical evaluation, source verification, and digital navigation skills as essential for democratic participation and civic engagement24,25. Despite these efforts, substantial gaps persist between theoretical ideals of media literacy and its practical operationalization, particularly in digitally saturated environments dominated by algorithmic curation.

Firstly, most global media literacy programs have historically focused on traditional, text-based news media, underestimating the rise of hyper-visual, short-form content ecosystems such as TikTok, Instagram Reels, and YouTube Shorts4,20. The cognitive, emotional, and behavioral demands of these fast-paced platforms differ fundamentally from conventional news environments, necessitating updated conceptualizations of literacy that accommodate visual persuasion techniques, rapid information processing, and algorithmically enforced exposure patterns5,10.

Secondly, the dominant models of media literacy education tend to emphasize individual cognitive skills—such as fact-checking and critical reasoning—without adequately addressing the emotional, social, and systemic dimensions of media engagement4,26. Emerging research underscores that emotional arousal, social influence, and platform design significantly shape users’ information processing behaviors, often overpowering cognitive control mechanisms27,28. Yet, existing frameworks largely omit these critical dimensions, resulting in interventions that fail to translate into resilient media behaviors in practice.

Thirdly, few empirical studies have systematically investigated how media literacy operates within algorithmically personalized content environments, where selective exposure and filter bubbles are structurally reinforced29,30. Although Media System Dependency Theory and Information Processing Theory have been applied separately to media studies, their integration in understanding literacy outcomes in short video platforms remains underdeveloped. As a result, there is a limited understanding of how cognitive engagement, emotional responses, and social environments interact to shape users’ critical media practices in contemporary digital spaces.

Within the Chinese context, studies show a high penetration of short video platform use among Generation Z15, but also reveal persistent vulnerabilities to algorithmic biases and misinformation, even among digitally savvy youth16,31. Media literacy initiatives in China remain predominantly focused on formal education settings and have not yet adequately addressed the unique challenges posed by short-form, algorithmically curated content ecosystems32. Thus, this study responds to these global and localized gaps by systematically examining how media literacy influences perceptions of information fragmentation, diversity of media exposure, and critical engagement among Generation Z users in Guangdong Province, China.

Theoritical understanding

This study constructs a multi-theoretical framework to explain how media literacy influences perceptions of information fragmentation, diversity of media exposure, and critical engagement within digital short video environments.

Media System Dependency Theory (MSDT) provides a systemic foundation by conceptualizing how reliance on media sources shapes individuals’ cognitive maps and behavioral responses33. In algorithmically curated ecosystems such as TikTok and YouTube Shorts, users’ dependency extends beyond information acquisition to encompass identity construction and social validation34. This deepened reliance exacerbates selective exposure, contributing to the perception of information fragmentation as users repeatedly encounter homogeneous, reinforcing content35. Media literacy—comprising knowledge of media operations, critical analysis ability, and bias recognition—emerges here as a disruptive force. By fostering skepticism toward algorithmic recommendations and encouraging critical evaluation of information sources, higher media literacy levels are theorized to weaken users’ dependency on algorithm-driven streams, subsequently reducing their perception of fragmentation7,36. For example, a media-literate user encountering repetitive political narratives on TikTok is more likely to seek alternative viewpoints, breaking the cycle of echo chambers. However, MSDT primarily addresses media dependency at the systemic level, offering limited explanation for how individual cognitive processes enable resistance, thus necessitating integration with cognitive theories.

Information Processing Theory extends the understanding by focusing on how individuals encode, store, and retrieve information amid the abundance of digital stimuli37. On short video platforms characterized by rapid visual consumption, shallow processing is common, often leading to acceptance of surface-level information without scrutiny. Media literacy acts as a cognitive scaffold, enhancing selective attention and fostering deeper cognitive engagement, which in turn supports more deliberate evaluation of diverse content. As users’ cognitive engagement increases, they are theorized to encounter a wider diversity of media exposure, seeking out heterogeneous perspectives rather than remaining confined to algorithmically suggested streams38. For instance, a user equipped with strong critical skills may intentionally diversify their feeds by searching for opposing views on trending topics. Furthermore, the relationship between media literacy and critical engagement is theorized to be mediated by cognitive engagement: those who exert greater mental effort in information processing are more likely to rigorously evaluate credibility, authenticity, and intent behind content. Nevertheless, while cognitive control is vital, it does not fully account for how emotional experiences and intuitive judgments shape engagement behaviors, necessitating further refinement through dual-process models.

Dual-Processing Theory further elaborates on the cognitive mechanisms by differentiating between heuristic (System 1) and analytical (System 2) modes of thinking39,40. The rapid, emotionally evocative nature of short video content frequently activates System 1 processing, leading users to uncritically absorb information that aligns with pre-existing beliefs41. Media literacy, by contrast, encourages the activation of System 2, prompting reflective scrutiny and critical evaluation of content claims. Consequently, critical engagement is conceptualized as a behavioral outcome of this cognitive shift toward deliberate information processing. As an example, a media-literate user might recognize sensationalist headlines in YouTube Shorts and actively seek source verification before forming an opinion. However, even among users with high media literacy, emotional responses to content can still profoundly influence processing depth, highlighting the need to integrate socio-emotional dimensions.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Research model.

Social Cognitive Theory (SCT) introduces these environmental and interpersonal moderators by emphasizing that behavior is shaped by interactions among cognitive, social, and environmental factors42. In digital media contexts, peer norms, community standards, and platform affordances significantly shape whether users apply their media literacy skills. Social influence—conceptualized as peer reinforcement or discouragement of critical media behaviors—is posited to moderate the relationship between media literacy and critical engagement. Users embedded in peer networks that prioritize critical thinking are more likely to question content credibility; conversely, environments emphasizing passive entertainment may suppress critical application even among media-literate individuals21,27. Moreover, technological proficiency moderates how effectively media literacy translates into critical behavior: users with greater proficiency in navigating platform interfaces and identifying source metadata are better equipped to operationalize their skills19. Usage patterns—including frequency, duration, and the variety of content consumed—also influence outcomes, as users exposed to a more heterogeneous media diet are less likely to experience information fragmentation43. For instance, frequent engagement with diverse educational and political content on TikTok may counteract the narrowing effects of algorithmic filtering. Finally, platform characteristics—such as personalization algorithms, content ranking systems, and user interface design—moderate the extent to which media literacy fosters exposure diversity. While some platforms promote content heterogeneity, others reinforce filter bubbles even for media-literate users30.

By integrating systemic dependency (MSDT), cognitive processing (Information Processing and Dual-Processing Theories), and socio-environmental modulation (SCT), this study advances a holistic model (Fig. 1) explaining how media literacy can mitigate information fragmentation, foster diversity of media exposure, and enhance critical engagement in short video ecosystems.

Hypothesis development

MSDT posits that individuals’ reliance on media systems significantly influences their perceptions and behaviors33. In the context of digital short video platforms, users with higher levels of media literacy—encompassing knowledge of media operations, critical analysis abilities, and recognition of biases—are better equipped to understand how media content is produced and disseminated. Empirical studies support this theoretical linkage. For instance, Ref44 found that increased media literacy is associated with greater skepticism toward social media content. This suggests that media-literate individuals are less likely to perceive information as fragmented because they can critically assess and integrate information from various sources. Reference38 demonstrated that media literacy enhances the ability to identify fake news, often a product of fragmented information ecosystems. While previous research has not explicitly addressed the direct relationship between media literacy and perceptions of information fragmentation, the connection is implied through MSDT’s emphasis on media reliance and critical engagement.

H1

Higher levels of media literacy (knowledge of media operations, critical analysis ability, and recognizing biases) are negatively associated with users’ perceptions of information fragmentation.

Information Processing Theory explains how individuals encode, store, and retrieve information, suggesting that cognitive abilities directly influence media engagement45. Media literacy enhances these cognitive abilities46,47, enabling users to actively seek out diverse sources and process complex information effectively. As a result, media-literate individuals are more likely to expose themselves to a broader array of content48, counteracting algorithmic content curation that often limits exposure diversity. Media literacy education has been linked to increased online political engagement and exposure to diverse perspectives49. Reference50 found positive effects of media literacy on media knowledge, critical thinking, and behavioral outcomes. Additionally, news media literacy is associated with improved current events knowledge and internal political efficacy51. Mass media, compared to interpersonal discussions, exposes individuals to more dissimilar political views52. Despite some gaps in cohesive conceptualization within media diversity research53, scholars advocate for a more inclusive and intersectional approach to media literacy education54. Furthermore, media health literacy is crucial for empowering individuals to engage with health information in our increasingly digitized world55.

H2

Higher levels of media literacy are positively associated with the diversity of media exposure.

Building on Information Processing Theory, media literacy also influences critical engagement with digital content56. By improving cognitive processing skills, media literacy enables individuals to critically evaluate digital content, assess credibility, and scrutinize sources57. This enhanced cognitive engagement leads to more frequent questioning of information and deeper critical interactions with content. Studies indicate that media literacy education can enhance critical engagement and improve online civic participation among youth49,58. As digital media becomes increasingly prevalent, the ability to assess information credibility is crucial57,59. Exposure to information about media ownership can increase skepticism and critical evaluation of news60. However, evaluating online information is complex, involving trust, agency, and context-dependent factors61. Models like the MAIN model offer heuristic approaches to understanding technology’s effects on credibility assessment. While youth develop digital competencies for personal goals, these skills may not always transfer to academic settings62. Overall, media literacy education is a promising approach for fostering critical thinking and enhancing digital engagement among young people.

H3

Increased media literacy enhances critical engagement with digital content.

Information Processing Theory further emphasizes that cognitive engagement—the mental effort invested in processing information—is a key mediator between knowledge acquisition and critical thinking45. Media literacy enhances cognitive engagement by equipping individuals with the skills to deeply process and analyze information63. Empirical evidence supports this mediating role. Media literacy education increases youth online political engagement58 and civic participation64. Critical digital literacy, which includes understanding utopian and dystopian perspectives, facilitates civic engagement26. Assessments of media literacy skills show that critical questioning habits become more complex after media literacy courses, indicating enhanced cognitive engagement65. Therefore, cognitive engagement acts as a bridge between media literacy and critical engagement.

H4

The relationship between media literacy and critical engagement is mediated by cognitive engagement.

SCT posits that emotional arousal influences learning and behavior by affecting attention and retention processes42. Exposure to diverse media content can elicit strong emotional responses, which, in turn, enhance cognitive processing and critical engagement with the content. Research has found that high-arousal emotions, both positive and negative, increase behavioral engagement such as commenting and sharing on social media66,67,68. Emotional intensity, rather than whether the emotion is positive or negative, appears to be a key driver of engagement69. Additionally, exposure to diverse media content can evoke stronger emotional responses, leading to increased critical engagement4,70. Individual differences, such as personality traits and moral foundations, also influence emotional responses and subsequent engagement behaviors71. These findings highlight the complex interplay between media content, emotional responses, and engagement behaviors.

H5

Emotional responses to content mediate the relationship between diversity of media exposure and critical engagement.

Information Processing Theory emphasizes that users’ cognitive engagement with information is not only influenced by their individual literacy levels but also by the diversity of media stimuli they encounter45. Media literacy, by equipping individuals with critical evaluation skills and selective attention, fosters exposure to a broader range of content and perspectives56. Diverse media exposure, in turn, heightens emotional responses by presenting users with conflicting, novel, or challenging information that stimulates deeper affective engagement72. Empirical studies support this dynamic: greater exposure diversity enhances emotional arousal and encourages reflective reactions rather than passive acceptance44,73. Similarly, diversity in media consumption is associated with higher levels of emotional and cognitive activation, essential for critical civic behaviors74. However, the effect is complex: if diversity is absent or poorly navigated, emotional responses may remain shallow or polarized75,76. Based on this, the framework theorizes that diversity of media exposure functions as a mediating mechanism linking media literacy to emotional engagement with content. Thus, it is hypothesized:

H6

Diversity of media exposure mediates the relationship between media literacy and Emotional responses.

Social Cognitive Theory (SCT) emphasizes that social environments critically shape the application of individual competencies42. In digital contexts, peer norms and community dynamics influence whether users apply media literacy skills to critically engage with content. Supportive environments that reinforce skepticism and inquiry enhance the application of media literacy77,78, while entertainment-driven peer cultures may suppress critical behaviors28. Evidence from online learning and financial literacy contexts confirms that peer influence can either amplify or diminish critical engagement79,80. Therefore, it is hypothesized:

H7

Technological proficiency moderates the relationship between media literacy and information fragmentation perception.

Building on Media System Dependency Theory (MSDT), individuals’ media dependency shapes not only information access but also perception patterns33. Usage patterns—particularly the frequency, diversity, and intentionality of engagement—are theorized to moderate how media literacy reduces information fragmentation. While simple increases in consumption may not predict literacy81, varied and intentional exposure enhances critical capacities82,83. For example, users diversifying across platforms show greater resistance to fragmented narratives84,85. Thus:

H8

Social influence moderates the relationship between media literacy and critical engagement.

SCT further emphasizes the role of environmental affordances. Platform characteristics—algorithmic design, content curation transparency, and interface architecture—shape users’ exposure to diverse viewpoints42. Empirical studies show that platforms differ significantly in promoting content diversity: while short-term exposure may be varied, long-term reliance often narrows user experiences29,86. Thus, even high media literacy may be constrained or facilitated depending on the platform’s affordances27,87. Accordingly:

H9

Usage patterns moderate the relationship between media literacy and perceived information fragmentation.

Finally, beyond platform design, individual usage behaviors—such as browsing duration, frequency of search, and engagement with varied content—critically determine the operationalization of media literacy. Research shows that intentional and diversified engagement reduces selective exposure and enhances information processing19,82. Passive, repetitive consumption, however, can lead to heuristic shortcuts, undermining critical skills43. Thus:

H10

Platform characteristics moderate the relationship between media literacy and perceived information fragmentation.

Study methodology

Study context

Guangdong Province represents a strategically appropriate and theoretically compelling context for investigating media literacy and information fragmentation among Generation Z users of short video platforms. As one of China’s most economically advanced and culturally diverse provinces, Guangdong exhibits both exceptional digital connectivity and profound socio-economic stratification88. With an internet penetration rate exceeding 80%, and widespread adoption of platforms such as Douyin and Kuaishou, users in Guangdong engage intensively with algorithmically curated content ecosystems89. However, this digital sophistication exists alongside pronounced disparities in educational access, urban-rural development, and media consumption habits across different subregions31. Such internal heterogeneity provides a rich setting for examining how media literacy competencies interact with information fragmentation risks under varied social and technological conditions.

Moreover, Guangdong’s strong integration into both national and global media flows intensifies users’ exposure to competing informational narratives, ideological contestations, and diverse cultural frames32. The province’s dynamic information environment amplifies the relevance of key constructs under investigation—particularly the need for critical engagement with diverse media sources, the mitigation of algorithmic echo chambers, and the management of emotional responses to heterogeneous content. Users in Guangdong, navigating rapid shifts between local, national, and transnational digital cultures, face intensified cognitive demands in processing, evaluating, and contextualizing short-form content compared to users in more homogenized or less digitally saturated regions90.

Selecting Guangdong thus allows the study to test the theoretical propositions of Media System Dependency Theory, Information Processing Theory, and Social Cognitive Theory under conditions of heightened media complexity, socio-economic variability, and accelerated digital platform use. It ensures that findings are grounded in a media ecosystem where information fragmentation pressures are not merely hypothetical but are observable and consequential for everyday cognitive, emotional, and behavioral practices.

Sampling

The sample size determination was based on a power analysis conducted using Cohen’s91 guidelines, targeting a medium effect size (f² = 0.15), a statistical power of 0.80, and an alpha level of 0.05, with up to 10 predictors. According to G*Power software92, these parameters indicated a minimum sample of 118 participants for achieving reliable results in multiple regression analysis. However, PLS-SEM, the primary analytical method employed in this study, is generally more robust with larger sample sizes to ensure greater reliability and validity in path modeling, especially in studies examining complex relationships93. Given this, we set an initial target of 500 participants, aiming to enhance statistical power and ensure model robustness.

Our sample size is further supported by prior research in similar domains. Studies on media literacy and digital engagement, such as those conducted by Huang4 and Tommasi5, have utilized samples of 400 or more respondents, establishing a precedent for adequate sample sizes in comparable studies. Such sample sizes enable both robust multivariate analysis and subgroup examination, ensuring comprehensive insights into generational media behaviors and attitudes. Accordingly, the final sample of 473 respondents aligns with the methodological rigor of similar studies and fulfills the requirements for PLS-SEM, enhancing the reliability of the findings.

To ensure representativeness, a stratified random sampling approach was implemented. Age-based strata were established to capture varied perspectives within Generation Z, specifically through three subgroups: Adolescents (13–18 years, 150 respondents), Young Adults (19–24 years, 173 respondents), and Older Young Adults (25–30 years, 150 respondents). Each subgroup was further balanced by gender, educational background, and geographic distribution (urban, suburban, rural) to accurately reflect Generation Z users of digital short video platforms in Guangdong Province. This stratification, while strengthening representational balance, also addresses the potential variability in media literacy and platform engagement across subgroups within Generation Z.

The sample size and selection process were also shaped by practical constraints, including participant accessibility and resource limitations. Utilizing Wenjuanxing, a prominent Chinese online survey platform, facilitated broad reach across Guangdong, enhancing recruitment efficiency while ensuring diversity in demographic representation. Despite a two-week data collection period and follow-up reminders to non-respondents, logistical factors, including survey fatigue and participant availability, contributed to the final sample size of 473 respondents. This is consistent with the minimum recommended sample and aligns with the best methodological practices.

While the sampling method aimed for regional representativeness, we recognize the limitation in generalizing findings beyond Guangdong Province.

Instrument items development

This study’s questionnaire was systematically designed to capture the constructs outlined in the conceptual framework, ensuring theoretical alignment, linguistic accuracy, and cultural relevance. The questionnaire consisted of several major sections corresponding to key constructs, each with clearly defined dimensions. Media literacy was conceptualized as a higher-order construct comprising three dimensions: (1) knowledge of media operations (e.g., understanding algorithmic curation and ownership structures), (2) critical analysis ability (e.g., evaluating source credibility and identifying misinformation), and (3) recognition of bias (e.g., detecting framing or agenda-setting techniques). Items measuring these dimensions were adapted from validated scales developed in prior studies on digital literacy and critical media literacy51,89.

Perception of information fragmentation was measured using the Information Fragmentation Scale7, capturing two dimensions: (1) perceived homogeneity of content streams and (2) perceived lack of exposure to alternative viewpoints. These dimensions reflect users’ cognitive awareness of content diversity or isolation within short video platforms.

Diversity of media exposure was operationalized through two components: (1) breadth of content types encountered (e.g., political, entertainment, educational) and (2) range of ideological viewpoints accessed (e.g., exposure to differing opinions on controversial issues). Items were adapted based on frameworks used in media consumption diversity studies51.

Critical engagement with digital content was measured through the Critical Media Engagement Scale51, covering (1) credibility verification behaviors (e.g., fact-checking) and (2) reflective skepticism (e.g., questioning content motives). These dimensions aligned with hypotheses regarding the depth of cognitive processing during digital interactions.

Technological proficiency, a proposed moderator, was assessed using items adapted from the Digital Proficiency Scale94, focusing on two dimensions: (1) operational skills (e.g., navigating platforms, managing settings) and (2) strategic skills (e.g., customizing algorithmic feeds or using privacy settings to curate content exposure).

Social influence was captured using the Social Influence Scale95, divided into (1) perceived peer norms supporting critical evaluation and (2) observed peer behaviors favoring diverse media engagement.

Platform characteristics were measured through items assessing (1) algorithm transparency (e.g., users’ perception of control over recommendation systems), (2) content interactivity features (e.g., ease of discovering new content), and (3) user interface intuitiveness.

All adapted items underwent a careful cultural and linguistic validation process during the translation and back-translation stages96,97. Where necessary, minor modifications to wording were made to ensure contextual relevance for Chinese Generation Z users (e.g., replacing references to “Facebook algorithms” with “Douyin recommendation systems”). A pilot study was conducted with 30 participants from the target demographic to assess item clarity, reliability, and content validity.

Filtering questions were embedded throughout the survey to verify participant eligibility and maintain data quality, such as confirming active use of short video platforms within the past three months. Ethical protocols were rigorously followed, including voluntary participation, informed consent, and strict anonymity protections, in accordance with the ethical guidelines of Holloway et al97..

The final data set was analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM), ensuring appropriate treatment of measurement models and complex structural paths.

Data analysis

The demographic profile of the 473 respondents in this study reveals a balanced representation across various categories (refer to Fig. 2), providing a comprehensive understanding of Generation Z users of digital short video platforms in Guangdong Province. The demographic data provides a well-rounded and representative snapshot of Generation Z users of digital short video platforms in Guangdong Province. The balanced age and gender distribution varied educational and occupational backgrounds, frequent platform usage, diverse content preferences, and geographical representation ensure that the findings of this study will be robust and reflective of the broader population.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Demographics profile.

Measurement model

The measurement model statistics (Table 1; Fig. 3) and discriminant validity results (Tables 2 and 3) provide an in-depth understanding of the constructs and their relationships within the study. Both Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA) were employed to ensure that the constructs exhibited strong psychometric properties. Specifically, factor loadings, composite reliability, average variance extracted (AVE), and variance inflation factor (VIF) were assessed to validate the reliability, convergent validity, and discriminant validity of the measures.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Measurement model (Lower order construct).

The Exploratory Factor Analysis (EFA) results demonstrated that all items loaded highly onto their respective constructs, with factor loadings (FL) ranging between 0.772 and 0.914. All outer loadings (OL) exceeded the 0.70 threshold93, confirming good item reliability. The VIF values for all indicators were well below the cut-off of 5.0, ranging from 1.603 to 3.352, indicating no significant multicollinearity issues among the items98.

The Confirmatory Factor Analysis (CFA) supported these findings. Composite reliability (CR) values ranged from 0.881 to 0.940, and Cronbach’s alpha (CA) values ranged from 0.797 to 0.915, both exceeding the recommended threshold of 0.7099. The AVE values for all constructs ranged from 0.680 to 0.758, exceeding the 0.50 benchmark100, thereby establishing strong convergent validity.

Table 1 Measurement model statistics.
Full size table

Table 2presents the heterotrait-monotrait ratio (HTMT) values for discriminant validity. According to Henseler et al.101, HTMT values should be below 0.85 to indicate adequate discriminant validity. Most HTMT values in this study are below 0.85, confirming that the constructs are distinct from each other. For instance, the HTMT value between Critical Analysis Ability (CAA) and Critical Engagement (CE) is 0.614, and between Diversity of Media Exposure (DME) and Emotional Responses (ER) is 0.847, both within acceptable limits.

Table 2 Discriminant validity (HTMT).
Full size table

Table 3 provides the Fornell-Larcker criterion (FLC) values for discriminant validity. The diagonal elements (square roots of AVE) are higher than the off-diagonal elements in the corresponding rows and columns, indicating that each construct shares more variance with its indicators than with other constructs100.

Table 3 Discriminant validity (FLC).
Full size table

Table 4 provides the outer loading composition for the higher order construct Media Literacy (MLL). The loadings are as follows: CAA on MLL is 0.894 with a standard deviation of 0.019 and a T statistic of 48.158 (p < 0.001); MO on MLL is 0.839 with a standard deviation of 0.028 and a T statistic of 29.450 (p < 0.001); and RB on MLL is 0.912 with a standard deviation of 0.013 and a T statistic of 67.852 (p < 0.001). These high loading values, all significant at the 0.001 level, indicate that the higher order construct MLL is well represented by its lower order constructs (CAA, MO, and RB).

Table 4 Out loading composition for higher order construct.
Full size table

In summary, the measurement model exhibits strong reliability and validity, supporting the robustness of the constructs used in this study. The discriminant validity confirmed through HTMT and FLC further validates that the constructs are distinct and accurately measured. The higher order construct MLL is also validated through significant and high outer loadings on its lower order constructs93,100,101.

Model fit and predictive relevence

The structural model statistics (refer to Fig. 4) provide critical insights into the relationships between media literacy and various outcome variables within the study. The model fit statistics in Table 5 indicate the explanatory power and predictive relevance of the constructs used.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Structural Model (Higher order construct).

The R-square value for Information Fragmentation Perception (IFP) is 0.268, which indicates that approximately 26.8% of the variance in IFP is explained by the model’s independent variables. While this value may appear modest, it is not uncommon in social science research, where behavioral and attitudinal outcomes are often influenced by numerous factors outside the model. In studies of complex, multifaceted phenomena—such as media engagement and information fragmentation—R-square values tend to be lower, reflecting the diversity of influences beyond those directly included in the model93,102.

Table 5 Model fit statistics.
Full size table

Moreover, smaller R-square values can still provide meaningful insights, especially when coupled with significant path coefficients and robust theoretical backing103. In this study, the Q²predict value of 0.215 for IFP adds support for the model’s predictive relevance, indicating that the constructs meaningfully contribute to understanding Information Fragmentation Perception within this specific context. The RMSE and MAE values of 0.895 and 0.640, respectively, also confirm a good fit, with lower error measures reinforcing the model’s reliability in capturing the core relationships of interest.

Structural model results

Table 6 presents the structural model results, summarizing the path coefficients, significance levels, and effect sizes (f²) for the hypothesized relationships.

Media literacy (MLL) significantly predicted information fragmentation perception (β = 0.310, p = 0.031, f² = 0.026); however, the positive direction contradicted the hypothesized negative relationship, thus H1 is not supported. In contrast, media literacy positively influenced diversity of media exposure (β = 0.529, p < 0.001, f² = 0.310) and critical engagement (β = 0.287, p = 0.002, f² = 0.034), supporting H2 and H3, respectively.

Regarding mediation effects, cognitive engagement significantly mediated the relationship between media literacy and critical engagement (β = 0.224, p = 0.001), supporting H4. Similarly, emotional responses mediated the relationship between diversity of media exposure and critical engagement (β = 0.107, p = 0.002; H5), and diversity of media exposure mediated the relationship between media literacy and emotional responses (β = 0.153, p < 0.001; H6).

The moderating effects produced mixed results. Technological proficiency (β = −0.094, p = 0.232; H7) and usage patterns (β = 0.034, p = 0.694; H9) did not significantly moderate the relationships as hypothesized, thus both H7 and H9 are not supported. However, social influence positively moderated the relationship between media literacy and critical engagement (β = 0.060, p = 0.013; H8), while platform characteristics negatively moderated the relationship between media literacy and diversity of media exposure (β = −0.062, p = 0.015; H10), supporting both H8 and H10.

In total, seven out of the ten hypotheses were statistically supported, providing substantial empirical validation for the theoretical model proposed in this study.

Table 6 Structural model Statistics.
Full size table

Discussion on the findings

The structural model findings provide nuanced insights into the relationships between media literacy and digital media engagement outcomes on short video platforms. Although H1 was statistically significant, the positive association between media literacy and information fragmentation perception contradicts the initial hypothesis. Rather than reducing perceptions of fragmentation, higher media literacy appears to heighten individuals’ awareness of content silos, algorithmic biases, and selective exposure patterns. This finding aligns with emerging research suggesting that critical media consumers, by virtue of their enhanced evaluative skills, may perceive the fragmented nature of digital media environments more acutely44. Rather than resolving fragmentation concerns, media literacy may sharpen users’ cognitive sensitivity to the structural biases of digital ecosystems, an important theoretical refinement of Media System Dependency Theory33.

The strong positive relationships found in H2 and H3 reaffirm the proposition that media literacy broadens informational horizons and strengthens critical digital behaviors. Media literacy fosters diverse media exposure49,84 and promotes reflective consumption, supporting models of critical media engagement57,58. These findings validate the integration of Information Processing Theory and Dual-Processing Theory into the framework, confirming that media-literate users shift towards more analytical, deliberate engagement39.

The mediation hypotheses (H4, H5, H6) offer further theoretical contributions. Cognitive engagement emerges as a key psychological mechanism through which media literacy enhances critical evaluation, corroborating Polizzi’s26 findings. Meanwhile, emotional responses to diverse content4,70 suggest that emotionally resonant media encounters can deepen engagement, emphasizing that media literacy interventions should target both cognitive and affective pathways. The mediation effect of diversity on emotional responses (H6) especially highlights that broader exposure facilitates richer emotional and evaluative interactions, underscoring the need to design media environments that encourage exposure to varied perspectives44,74.

The moderating hypotheses yielded mixed results, requiring deeper theoretical reflection. Technological proficiency (H7) did not significantly moderate the relationship between media literacy and information fragmentation perception. Although prior studies suggest that digital skills are crucial for navigating online environments94, this finding suggests that operational proficiency alone may be insufficient. Users may possess technical navigation skills without critically challenging algorithmic biases or seeking exposure diversity29. This highlights an important nuance: critical media competencies are not purely technical but involve deeper cognitive and evaluative capacities beyond operational familiarity22. Future research could incorporate more differentiated models of digital literacy—such as distinguishing between functional proficiency and critical proficiency—to better understand these dynamics.

In contrast, social influence (H8) significantly moderated the relationship between media literacy and critical engagement. Users embedded within supportive social environments that value critical media consumption were more likely to translate literacy skills into critical behaviors28,79. This reinforces Social Cognitive Theory’s assertion that environmental reinforcement plays a crucial role in shaping individual behaviors42. Interventions targeting peer environments—such as peer-led media literacy programs—may thus offer particularly powerful means of enhancing critical engagement.

Platform characteristics (H10) were found to negatively moderate the relationship between media literacy and diversity of media exposure, suggesting that even highly literate users are constrained by systemic platform architectures. Algorithmic curation, interface designs, and engagement-maximization logics can restrict users’ exposure to diverse content, despite users’ intentions to seek varied perspectives27,87. This finding challenges the assumption that user agency alone can overcome algorithmic biases and highlights the necessity of structural interventions, such as greater algorithmic transparency or user-centered recommendation systems104.

The non-significance of usage patterns (H9) provides further insight into the limitations of equating engagement frequency with literacy outcomes. Rather than the amount of time spent on short video platforms, it is the intentionality and quality of engagement that determine critical media behaviors. Habitual, entertainment-driven consumption tends to reinforce selective exposure and passive engagement, while purposeful and reflective use fosters diversity and critical awareness105. This distinction between habitual and intentional use suggests that future studies should operationalize usage patterns not merely as quantitative frequency metrics but as qualitative indicators of media purpose, motivation, and depth of interaction. Incorporating these behavioral dimensions would enable a more accurate assessment of how platform engagement translates into literacy-driven outcomes.

Finally, the findings suggest potential extensions to the model. Future research could investigate additional moderators such as individual skepticism, critical thinking disposition, or personality traits (e.g., openness to experience) to better understand variations in media literacy outcomes44,106. These factors may interact with literacy skills to either strengthen or attenuate critical engagement and exposure diversity.

In summary, these findings demonstrate that media literacy is a powerful but contextually bounded force. Its success depends not only on individual competencies but also on social support systems, technological structures, and emotional engagement pathways. Effective interventions must thus adopt a multi-layered approach, addressing personal, social, and systemic barriers to meaningful digital engagement.

Implications of the study

Theoretical implications

This study provides substantial theoretical contributions by addressing critical research gaps within media literacy scholarship, uniquely integrating Media System Dependency Theory (MSDT), Information Processing Theory, Social Cognitive Theory (SCT), and Dual-Processing Theory. Previous studies often examined these theories independently, neglecting the dynamic interplay among cognitive, emotional, social, and platform-specific factors in digital short video contexts. By synthesizing these theories, this study introduces a holistic, multi-theoretical framework, significantly advancing scholarly understanding of media literacy within algorithmically driven digital environments, particularly among Generation Z users.

Notably, this research contributes to MSDT by uncovering an unexpected positive relationship between media literacy and perceived information fragmentation (β = 0.310, p = 0.031). This contradicts traditional expectations that increased media literacy uniformly reduces dependency on fragmented sources. Instead, findings suggest that media-literate individuals develop heightened awareness of content biases and selective exposure dynamics, leading them to perceive information ecosystems as more fragmented. This expands MSDT by redefining media dependency as a nuanced cognitive state where heightened literacy promotes critical recognition of media limitations rather than mere disengagement from dependency44.

Further, the substantial positive relationships between media literacy and diversity of media exposure (β = 0.529, p < 0.001) and critical engagement (β = 0.287, p = 0.002) significantly enhance Information Processing Theory and Dual-Processing Theory. This confirms that media literacy fosters deliberate cognitive engagement and the systematic evaluation of diverse information sources, reducing selective exposure effects49,84. Additionally, by identifying cognitive engagement as a vital mediating mechanism (β = 0.224, p = 0.001), this study empirically validates theoretical assumptions that deeper cognitive involvement facilitates critical and reflective media interactions26.

Moreover, integrating emotional responses as mediators (β = 0.107, p = 0.002) into Information Processing Theory constitutes an innovative theoretical extension. The emotional dimension has frequently been overlooked in traditional cognitive media literacy frameworks. This study demonstrates that emotionally engaging content actively enhances critical reflection and deeper cognitive processing4,70. Thus, this finding expands existing cognitive models, highlighting the necessity of integrating emotional factors within comprehensive media literacy theories.

The study also significantly enriches Social Cognitive Theory by empirically substantiating social influence as a crucial moderating variable in the relationship between media literacy and critical engagement (β = 0.060, p = 0.013). This confirms SCT’s premise that environmental and social contexts profoundly influence the application of media literacy skills, with supportive peer norms and community behaviors amplifying critical media practices28,79. Conversely, the non-significant moderating effects of technological proficiency (β = −0.094, p = 0.232) and usage patterns (β = 0.034, p = 0.694) provide vital theoretical clarification, challenging widespread assumptions in Information Processing Theory that operational technical skills or increased media exposure alone enhance critical media evaluation29. These results suggest a need to distinguish clearly between functional digital literacy (basic technical skills) and critical digital literacy (evaluative, cognitive capacities) in theoretical models22.

Finally, the negative moderation by platform characteristics on media literacy’s relationship with exposure diversity (β = −0.062, p = 0.015) significantly extends SCT. It highlights how algorithmic designs and interface structures constrain users’ application of media literacy skills, even among highly literate individuals. This underscores a crucial theoretical refinement: media literacy must be viewed as operating within a complex socio-technological ecosystem, where individual agency interacts dynamically and sometimes adversarially with technological architectures27,87. This finding calls for further theoretical development around structural affordances and constraints within digital platforms.

Overall, the theoretical contributions of this study lie in providing a robust, integrated, and nuanced theoretical framework. It highlights the intricate interplay of cognitive, emotional, social, and technological dimensions, positioning media literacy not merely as an individual competency but as a dynamic and contextually dependent skill-set essential for navigating contemporary digital environments.

Practical implications

This study offers actionable practical implications across multiple stakeholder groups, including educators, policymakers, digital platform designers, and content creators, providing empirically-grounded strategies to enhance media literacy and address information fragmentation among Generation Z.

For educators, findings underscore the necessity of comprehensive curricula integrating cognitive, emotional, and evaluative dimensions of media literacy. Concrete examples include UNESCO’s Media and Information Literacy (MIL) curriculum and Finland’s national MIL educational programs, both of which prioritize critical questioning, reflection, and emotional engagement. For instance, Finland’s “Facts vs. Fiction” media literacy initiative successfully incorporates interactive workshops promoting both cognitive evaluation and emotional reflection, significantly enhancing students’ abilities to critically interpret diverse digital content23,24.

Policymakers can utilize these findings to design targeted regulatory frameworks enhancing media literacy. The European Union’s Digital Services Act (DSA) exemplifies such policy by mandating algorithmic transparency, empowering users with greater insight into content recommendation processes, thereby supporting critical engagement and reducing passive information fragmentation25. Additionally, policymakers could invest in community-level initiatives such as Australia’s national media literacy strategy or Canada’s MediaSmarts programs, emphasizing social contexts and community involvement as pivotal factors in strengthening media literacy outcomes.

For digital platform designers, findings highlight specific platform-level interventions to mitigate structural constraints on media literacy. Practical measures include greater transparency features (e.g., TikTok’s “Why am I seeing this?” tool or YouTube’s recommendation transparency measures), which help users consciously diversify their information exposure. Moreover, platforms might implement customizable algorithmic controls, allowing users active participation in curating content recommendations, enhancing both cognitive and emotional engagement with diverse content104.

Content creators and media organizations can leverage the study’s findings on emotional mediation by designing emotionally resonant media literacy campaigns. Effective examples include the BBC’s and PBS’s emotionally engaging public media literacy campaigns, which utilize compelling storytelling and real-world scenarios to deepen user reflection and encourage critical engagement with misinformation. Additionally, educational media productions, such as TED-Ed’s animated lessons on misinformation, demonstrate how integrating affective storytelling with cognitive instruction significantly boosts learner engagement and critical evaluation capabilities.

In summary, these practical implications offer a multi-layered approach grounded in robust empirical findings. Successful interventions will require collaborative efforts among educational institutions, regulatory bodies, digital platform providers, and content creators, emphasizing that meaningful digital literacy enhancement is contingent on addressing both individual competencies and the socio-technical environments in which media interactions occur.

Conclusion and future research directions

This study provides an empirically grounded and theoretically integrative contribution to media literacy scholarship by demonstrating that higher media literacy, contrary to conventional expectations, does not necessarily diminish perceptions of information fragmentation. Instead, it sharpens users’ critical awareness of content silos, algorithmic biases, and selective exposure patterns. This outcome refines Media System Dependency Theory (MSDT) by reframing media dependency not merely as reliance on digital infrastructures, but as a cognitively heightened sensitivity to the systemic and structural constraints that shape contemporary information ecosystems.

By incorporating cognitive, emotional, and social pathways—evidenced through the mediating roles of cognitive engagement and emotional responses—this research extends the explanatory scope of Information Processing Theory and Dual-Processing Theory. The findings indicate that media literacy is not solely a rational or analytical competence but also profoundly shaped by affective resonance and social reinforcement. Consequently, interventions must address the emotional and communal dimensions of digital engagement alongside cognitive training.

Equally significant are the systemic and contextual barriers identified in the study. Platform characteristics—such as algorithmic curation, personalization bias, and interface design—were found to restrict the positive influence of media literacy on exposure diversity. Meanwhile, technological proficiency and usage frequency alone did not enhance critical engagement, suggesting that functional digital skills are insufficient without reflective, intentional use. These outcomes collectively challenge prevailing assumptions that increased screen time or digital fluency automatically strengthen media discernment. Instead, effective literacy requires socially supported, critically conscious, and structurally enabled engagement environments.

From a practical perspective, this research highlights the need for multi-dimensional literacy initiatives that integrate cognitive, emotional, and social reinforcement strategies. Educators should develop curricula that foster both analytical and affective competencies; policymakers should promote transparency and accountability in algorithmic governance; and digital platforms should design user-centered systems that empower individuals to diversify their media exposure. The interplay between individual agency and systemic design underscores the necessity of shared responsibility among users, institutions, and platforms in cultivating informed, reflective digital citizens.

Limitations and avenues for future research

While the study provides robust theoretical and empirical insights, several limitations warrant acknowledgment. First, the research is geographically confined to Generation Z users in Guangdong Province, China, which may limit the generalizability of findings across different socio-cultural or regulatory environments. Second, the use of cross-sectional survey data precludes causal inference, suggesting that future studies should adopt longitudinal or experimental designs to examine how literacy competencies evolve over time. Third, all measures were self-reported, which may introduce social desirability or common-method bias despite procedural controls. Finally, cultural and linguistic adaptations of the instrument, although rigorously validated, may still have influenced item interpretation.

Building on these limitations, future research should pursue cross-cultural comparative studies that examine how differing cultural orientations (e.g., collectivist vs. individualist societies), regulatory frameworks (e.g., EU’s Digital Services Act vs. China’s content moderation systems), and platform ecologies influence the relationships observed in this model. Longitudinal research could explore how sustained exposure to algorithmically curated environments shapes users’ literacy trajectories and emotional resilience. Moreover, the accelerating influence of emerging technologies—including artificial intelligence, augmented and virtual reality, and generative content systems—should be incorporated into future frameworks to assess how new media ecologies transform users’ evaluative, ethical, and affective literacies.

In addition, micro-level variables such as cognitive flexibility, skepticism, trust calibration, and emotional regulation merit deeper investigation as potential mediators or moderators of literacy outcomes. These factors could help explain individual heterogeneity in responses to similar media environments. Future studies may also benefit from adopting mixed-methods designs, integrating qualitative insights to contextualize quantitative patterns, and employing behavioral tracking or eye-movement analytics to complement self-report data.

In summary, this study reconceptualizes media literacy as a dynamic, multi-layered competency—one that operates at the intersection of cognitive processing, emotional engagement, social influence, and technological structure. By bridging theoretical traditions and empirical evidence, it provides a more realistic and contextually grounded foundation for advancing future research, policy development, and educational innovation. Ultimately, fostering critical, emotionally attuned, and socially supported media literacy represents a pivotal step toward building resilient, ethically informed, and critically engaged digital citizens in an era of pervasive algorithmic mediation.