Agenda-Setting and Gatekeeping Theories in the Digital Age of Algorithmic Media: A Critical Synthesis Review and Theoretical Extension (2000–2025)

1. Introduction

The capacity of media systems to determine which issues, events, and interpretations achieve public visibility has long been recognized as among the most consequential dimensions of communicative power in democratic societies. Two theoretical traditions have supplied the dominant analytical frameworks for understanding this capacity. Gatekeeping theory, originating in Kurt Lewin's wartime studies of food distribution channels [24] and adapted to communication research by David Manning White [50], examines the processes, actors, and institutional structures through which information is selected, filtered, and channeled toward publics. Agenda-setting theory, inaugurated by McCombs and Shaw's landmark study of the 1968 United States presidential campaign [25], investigates the mechanisms through which media attention to particular issues shapes public perceptions of their importance, establishing a correspondence between the media agenda and the public agenda that has since been replicated across hundreds of empirical studies in diverse national and cultural contexts.

Despite their evident functional relationship — gatekeeping determines what information passes through institutional filters, while agenda-setting describes the consequences of that filtering for public cognition — these traditions have developed along largely separate intellectual trajectories. Gatekeeping research has concentrated on the internal sociology of media organizations, examining the decision-making routines, professional norms, and institutional pressures that shape editorial selection [40]. Agenda-setting research has concentrated on the external effects of media coverage on public opinion, treating the media agenda as a largely exogenous variable whose formation is analytically prior to the question of its influence [26]. The result is what this paper characterizes as a disciplinary bifurcation: a division of labor that has generated important insights within each tradition while obscuring the systemic connections between the processes of information selection and their consequences for public salience.

The digital transformation of the media landscape between 2000 and 2025 has rendered this bifurcation untenable. The rise of social media platforms, algorithmic content curation, and participatory digital culture has fundamentally altered the structures through which information achieves public visibility, collapsing the institutional boundaries that once separated the gatekeeping function from its agenda-setting consequences. In the algorithmic media environment, the act of filtering information and the act of conferring salience upon it are no longer sequential processes performed by distinct actors in distinct institutional settings; they are simultaneous, recursive, and frequently indistinguishable operations performed by computational systems whose decision logic is opaque both to the publics they serve and to the scholars who study them [7,40]. Recent large-scale field experiments conducted on Facebook and Instagram during the 2020 United States election cycle have confirmed empirically what theorists had argued conceptually: the feed-ranking algorithm is now a first-order determinant of what citizens see, substantially restructuring exposure to political information relative to chronological alternatives [16,18].

This paper undertakes a critical, integrative review of gatekeeping and agenda-setting theory across this transformative period, pursuing three interconnected objectives. The first is descriptive: to map the evolution of both traditions from their pre-digital foundations through their successive adaptations to the realities of digital and algorithmic media. The second is analytical: to identify and examine the points of theoretical convergence that the digital transformation has made visible, including the shared mechanisms of algorithmic filtering, platform governance, and networked information flow that now govern both processes simultaneously. The third is constructive: to propose the concept of salience agency — defined as the capacity of any actor, whether human, institutional, or algorithmic, to influence the visibility, prominence, and perceived importance of information within the public communication system — as an integrative meta-theoretical framework that bridges the analytical gap between gatekeeping and agenda-setting, accommodates the empirical complexity of the contemporary media landscape, and generates productive directions for future research.

The significance of this undertaking extends beyond academic communication theory. The processes through which information achieves public visibility are foundational to democratic governance, the formation of public opinion, and the capacity of citizens to make informed judgments about matters of collective concern. When these processes are governed by opaque algorithmic systems optimized for commercial engagement rather than democratic deliberation, the implications for the quality of public discourse and the health of democratic institutions are profound [44,52]. Understanding the convergence of gatekeeping and agenda-setting in the algorithmic age is therefore not merely a scholarly exercise but a civic imperative.

2. Review Methodology

This paper employs a structured integrative review methodology designed to synthesize theoretical and empirical contributions from communication studies, supplemented by relevant work in information science, political communication, computational social science, and platform governance research. The integrative review method was selected for its capacity to accommodate both quantitative and qualitative research traditions, to identify theoretical convergences across heterogeneous bodies of literature, and to generate new conceptual frameworks through systematic synthesis rather than mere aggregation [47].

The review encompasses peer-reviewed journal articles, scholarly monographs, edited volumes, and substantive institutional reports published between 2000 and 2025, with selective engagement with foundational texts from earlier periods where necessary to establish theoretical lineage. Literature was identified through systematic searches of major academic databases — Scopus, Web of Science, Communication Abstracts, and Google Scholar — using search terms encompassing "gatekeeping theory," "agenda-setting," "algorithmic curation," "platform governance," "digital media," "networked gatekeeping," and related constructs. Searches were supplemented by backward citation tracking from key recent publications and forward citation tracking from foundational texts in both traditions. Particular weight was given to the most recent empirical wave (2020–2025), including the large-scale platform experiments conducted in collaboration with Meta during the 2020 United States election [16,18,35], algorithm-audit studies [1], and cross-national surveys of news consumption [33].

The scope of the review is deliberately broad, encompassing work from diverse national and cultural contexts in recognition that the digital transformation of media systems is a global phenomenon whose dynamics vary significantly across geopolitical settings. Particular attention has been paid to scholarship examining platform governance beyond Western liberal democracies — including work on China [21], the Arab world [22], Ukraine [15], and Palestine [41] — to resist the universalization of theoretical models derived primarily from Western media systems.

The analytical framework is organized around three interconnected dimensions. The first is temporal, tracing the evolution of both traditions across the twenty-five-year period and identifying inflection points at which technological, institutional, or political developments necessitated theoretical revision. The second is structural, examining the architectural features of digital platforms that shape information filtering and salience conferral, including recommendation systems, content moderation policies, and platform-specific affordances. The third is normative, assessing the democratic implications of the observed transformations and evaluating the adequacy of existing regulatory frameworks for addressing them.

3. Theoretical Foundations: Gatekeeping and Agenda-Setting in the Pre-Digital Era

The study of how media systems shape the public visibility of issues has been central to understanding communicative power in democratic societies, and the gatekeeping and agenda-setting traditions supplied its primary analytical instruments. In the pre-digital era the two theories developed along distinct paths: gatekeeping illuminated the internal mechanics of media organizations, while agenda-setting measured the external influence of media coverage on public opinion. Together they laid the groundwork for understanding the flow and salience of information in traditional media environments — groundwork whose assumptions the algorithmic turn would later unsettle. Table 1 summarizes the principal milestones in the development of both traditions, from their mid-century origins to the threshold of the algorithmic era [2,5,6,19,25,26,39,40,42,50].

3.1. The Gatekeeping Tradition

Gatekeeping theory emerged from a fortuitous interdisciplinary transfer. Lewin's observation that food items pass through a series of "channels" governed by "gates" — decision points at which items are admitted or rejected [24] — was adapted by White to the study of news selection in his celebrated case study of a wire editor pseudonymously named "Mr. Gates" [50]. White demonstrated that the selection of news was governed not by objective criteria of newsworthiness alone but by the subjective judgments, personal biases, and professional routines of the individual gatekeeper. This finding — that the construction of "the news" is an inherently selective and interpretive process — supplied the foundational insight on which subsequent gatekeeping research was built.

Over the following decades the theory underwent progressive elaboration. Shoemaker developed a multi-level model situating the individual gatekeeper within concentric analytical levels — individual, routine, organizational, extra-media, and ideological — each exerting constraining influence on the selection process [39]. This model advanced decisively beyond White's individualistic framework by demonstrating that gatekeeping is not reducible to the preferences of individual decision-makers but is systematically shaped by institutional routines, organizational hierarchies, economic pressures, and broader ideological formations. Shoemaker and Vos later refined the framework, introducing the concept of audience gatekeeping to acknowledge the growing capacity of media consumers to participate in the selection and dissemination of information — an insight that anticipated the participatory dynamics of the digital environment [40].

A critical turning point was Barzilai-Nahon's reconceptualization of gatekeeping as a networked phenomenon [2]. Observing that the internet had introduced a communicative architecture fundamentally different from the one-to-many broadcast model assumed by classical theory, she proposed a network gatekeeping theory foregrounding network position, information control mechanisms, and the political relationship between gatekeepers and the "gated" — those whose access to information is controlled by gatekeeping processes. The reconceptualization was significant not only analytically but normatively: by identifying the gate as a distinct category with interests potentially opposed to those of the gatekeeper, it introduced questions of power, accountability, and legitimacy that earlier formulations had left underdeveloped. Subsequent work by Singer on "secondary gatekeeping" [42] and by Bruns on "gate watching" [5] extended the theory to the participatory dynamics of social media, in which users select, share, and amplify content in ways that complement and contest institutional editorial authority, while Bucher's analysis of algorithmic power marked the tradition's decisive confrontation with computational curation [6].

3.2. The Agenda-Setting Tradition

Agenda-setting theory was established by McCombs and Shaw's study of undecided voters in Chapel Hill, North Carolina, during the 1968 presidential election [25]. The study demonstrated a strong correlation between the issues emphasized in news coverage and the issues voters identified as most important, supporting the hypothesis that media do not tell people what to think but what to think about. This foundational insight — that media influence operates primarily through the conferral of salience rather than direct persuasion — has become one of the most robust findings in communication research.

The theory's evolution has proceeded through identifiable phases. The first level, established at Chapel Hill, concerns the transfer of object salience: the capacity of media to make certain issues more prominent in public consciousness. The second level, developed in the 1990s, extends the analysis to attribute salience — the capacity of media to shape not only which objects the public thinks about but which attributes of those objects are perceived as most important [26]. A newspaper that covers a candidate primarily in terms of leadership, for example, transfers salience not only to the candidate but to leadership as the evaluative frame through which the candidate is judged.

The emergence of digital media prompted a third level network agenda setting (NAS), which examines how media construct associative networks linking multiple objects and attributes, shaping not only what the public thinks about and in what terms but the cognitive associations through which discrete issues are connected into coherent interpretive frameworks [19]. NAS moved agenda-setting from a dyadic model of salience transfer to a network model of cognitive architecture in which the relationships among agenda items are as consequential as the items themselves. Vargo and Guo subsequently extended the NAS model to intermedia agenda-setting across digital platforms, demonstrating that the associative networks through which issues are cognitively linked are increasingly shaped by the algorithmic logic of platform-mediated information flows [48]. In the digital environment, the bundling of issues — the association of immigration with crime, or climate change with economic loss — is not merely a function of editorial framing but of algorithmic co-presentation, in which recommendation engines systematically present certain issues in conjunction with others on the basis of engagement optimization rather than journalistic judgment.

3.3. Points of Pre-Digital Convergence

Even before the digital transformation, perceptive scholars identified structural parallels between the traditions that pointed toward their eventual convergence. Both address the fundamental process through which certain information achieves public visibility while other information remains obscure. Both recognize that this process is not neutral but systematically shaped by institutional structures, professional norms, and power relations. And both implicitly acknowledge that the selection of information and the conferral of importance upon it are functionally inseparable — an item that passes through the gate simultaneously acquires a measure of public salience by virtue of its selection.

What prevented explicit integration in the pre-digital era was the institutional architecture of the legacy media system. In an environment characterized by a relatively small number of broadcasting and publishing institutions, each employing professional journalists governed by shared normative frameworks, the gatekeeping function was institutionally localized: it occurred within identifiable organizations whose decision-making could be studied through ethnographic, interview, and content-analytic methods. The agenda-setting function, by contrast, was diffused across the media system and manifested in its cumulative effects on public opinion. The institutional separation of these functions — one inside the newsroom, the other in the domain of public cognition — provided a pragmatic basis for their analytical separation, even as the theoretical grounds for that separation were always questionable.

The digital transformation dissolved this pragmatic basis by dissolving the institutional boundaries that sustained it. When algorithms replace editors, when users become distributors, and when platform architecture determines the conditions of information visibility, the gatekeeping function is no longer institutionally localized, and its agenda-setting consequences are no longer analytically separable from the filtering processes that produce them. The remainder of this paper examines that dissolution and its implications.

4. The Digital Media Landscape: Key Transformations (2000–2025)

The early twenty-first century ushered in a period of rapid and profound transformation in the media environment, reshaping how information is produced, disseminated, and consumed. As digital technologies matured and platforms emerged to challenge the dominance of traditional media, the foundations of public communication were fundamentally altered. This section traces the pivotal changes that redefined the media landscape between 2000 and 2025, highlighting the shift from institutionally controlled channels to algorithmically mediated networks.

4.1. From Institutional to Algorithmic Mediation

The period under examination witnessed a transformation in the architecture of public communication that is without precedent in the history of media. At the turn of the millennium, the dominant intermediaries between information producers and publics were institutional media organizations — newspapers, television networks, radio stations — staffed by professional journalists whose gatekeeping decisions were governed by established norms of newsworthiness, editorial standards, and institutional accountability. By 2025, these organizations, while still significant participants in the information ecosystem, had been substantially displaced as primary intermediaries by a small number of technology companies whose platforms — Meta, X, Google and YouTube, TikTok, and a handful of others — governed the visibility of information for billions of users through proprietary algorithmic systems [30,31]. The displacement is now measurable at the level of national news diets: by 2025, social media and video networks had for the first time overtaken television as the leading news source in the United States [33]. As shown in Table 2, this transformation unfolded across three analytically distinguishable phases [7,30].

The first phase, spanning roughly 2000 to 2008, was characterized by the digitization of legacy media and the emergence of early participatory platforms, including blogs, forums, and early social networking sites. During this period the internet was frequently conceptualized as a supplement to existing media structures rather than a qualitatively new communicative environment. Legacy organizations established online presences that largely reproduced their editorial hierarchies in digital form, and audience participation, while technically enabled, remained marginal to the primary circuits of information production [42].

The second phase, from approximately 2008 to 2016, witnessed the rapid ascendancy of social media platforms as major nodes in the information ecosystem. The proliferation of smartphones, the expansion of mobile internet access, and the development of algorithmically curated news feeds transformed platforms from social networking tools into primary channels of news consumption for large segments of the global population [12]. During this period the gatekeeping function began its migration from human editorial decision-makers to algorithmic systems, as platform companies discovered that personalized content feeds generated substantially higher engagement — and therefore advertising revenue — than chronologically ordered or editorially curated alternatives [36]. News organizations, in turn, became structurally dependent on platform distribution, reorganizing editorial workflows around the referral economies of a small number of dominant intermediaries [27,34].

The third phase, from 2016 onward, has been defined by the recognition — among scholars, policymakers, and publics — that the algorithmic mediation of information carries profound consequences for democratic governance, social cohesion, and epistemic integrity. The role of social media in the 2016 United States presidential election and the Brexit referendum catalyzed critical scrutiny of the mechanisms through which algorithmic systems shape public discourse [30], culminating in landmark regulatory responses including the European Union's Digital Services Act [10] and Artificial Intelligence Act [11]. This scrutiny generated a rapidly expanding body of scholarship on platform governance and algorithmic accountability — scholarship that provides the empirical foundation for the theoretical synthesis undertaken in this paper.

4.2. Hybrid Media Dynamics and the Dissolution of Categorical Boundaries

The concept of the hybrid media system, introduced by Chadwick [7], provides an essential analytical framework for understanding the contemporary information landscape. Chadwick argues that the media environment is characterized not by the simple displacement of "old" media by "new" media but by the ongoing interaction, competition, and mutual adaptation of diverse media logics — broadcast and networked, professional and participatory, institutional and algorithmic — within a single complex ecosystem. In this hybrid system, the boundaries between media production and consumption, between editorial selection and algorithmic curation, and between institutional authority and participatory influence are not fixed but fluid, contested, and continuously renegotiated.

The hybrid framework has profound implications for both traditions under review. For gatekeeping, it implies that the relevant unit of analysis is no longer the individual gatekeeper or media organization but the entire assemblage of human and non-human actors — journalists, editors, algorithms, platform policies, user behaviors, regulatory frameworks — that collectively determine the conditions of information visibility. For agenda-setting, it implies that the media agenda is no longer a singular construct measurable through content analysis of a finite set of outlets but a fragmented, platform-specific, and algorithmically personalized phenomenon that varies across audiences, platforms, and temporal contexts [46].

The dissolution of categorical boundaries has also generated new forms of intermedia agenda-setting. Harder and colleagues demonstrate that the directional flow of agenda influence between legacy media and digital platforms is reciprocal and dynamic, with issues originating on social media increasingly setting the agenda for legacy coverage and vice versa — even as traditional journalistic actors retain disproportionate influence over the composite news agenda during election periods [20]. The speed of this reciprocal influence has accelerated dramatically, with social media capable of elevating issues to national prominence within hours — a temporal compression that has fundamentally altered the rhythms of public discourse and the capacity of media professionals to exercise deliberative editorial judgment [48].

4.3. Platform Governance as Structural Gatekeeping

A critical analytical development of the post-2016 period has been the recognition that the governance structures of digital platforms — their terms of service, content moderation policies, and algorithmic design choices — function as structural gatekeeping mechanisms with consequences as far-reaching as any editorial decision [13]. Platform governance determines not only what content is visible but the conditions under which visibility is possible: the criteria by which content is recommended or suppressed, the standards by which it is evaluated for compliance with community guidelines, and the mechanisms through which users can contest moderation decisions.

Empirical research has increasingly documented the consequential nature of these governance structures. Roberts demonstrates that content moderation on major platforms functions not merely as a technical mechanism for removing harmful material but as a labor-intensive and politically significant process that shapes the informational environment according to platform priorities frequently opaque to external observers [37]. Zeng and Kaye's study of TikTok extends this insight, showing that governance on algorithm-intensive platforms increasingly operates through what they term visibility moderation — the calibrated amplification and suppression of content through recommendation infrastructure rather than through binary removal decisions [51]. Gillespie similarly documents the rise of "reduction" as a moderation strategy in which platforms quietly demote borderline content without removing it, exercising consequential gatekeeping power in a register that is nearly invisible to users and regulators alike [14]. Case-based scholarship reveals the political stakes of these arrangements: Siapera documents how Palestinian communicators navigating platform community standards have developed adaptive strategies for maintaining informational presence despite moderation policies that restrict conflict-related content [41], while Golovchenko and colleagues show how the information war over Ukraine was shaped by state actors, media institutions, and citizen curators operating within and against platform governance structures [15].

These cases underscore a broader analytical point: platform governance constitutes a form of gatekeeping structurally distinct from both traditional editorial gatekeeping and emergent algorithmic gatekeeping. Where editorial gatekeeping operates through the professional judgment of identifiable individuals and algorithmic gatekeeping through computational processes optimized for engagement, platform governance operates through the establishment of systemic rules and norms that define the parameters within which both of the former occur. It is, in this sense, a form of meta-gatekeeping — gatekeeping over the conditions of gatekeeping itself [13].

5. Empirical Analysis and Contemporary Data Synthesis (2000–2025)

This section synthesizes the empirical record on the evolving mechanisms of information filtering, gatekeeping, and agenda-setting across the review period. Drawing on recent scholarship, industry data, and comparative trends across platform architectures, it examines the transition from the digitization of legacy media to the rise of algorithmically curated platforms and the emergence of platform governance as a pivotal force in shaping public discourse. Three empirical domains are examined in turn: comparative filtering logics across platform types, algorithmic bias and salience distortion, and the counter-hegemonic potential of citizen journalism and decentralized networks.

5.1. Comparative Trends in Information Filtering Across Platform Architectures

Empirical investigation across the period reveals a fundamental and accelerating transformation in how information is prioritized, filtered, and disseminated. The shift was not merely quantitative — more content, more channels, more users — but qualitative, involving a structural reconfiguration of the relationships among producers, intermediaries, and audiences. By the early 2010s, platforms such as Facebook and Twitter had established themselves as significant nodes in the information ecosystem, capable of shaping content visibility in ways that paralleled and increasingly rivaled the editorial authority of legacy institutions [31]. By 2025, the algorithmic systems governing these platforms had assumed a gatekeeping function of historically unprecedented scale and complexity. Platform architecture constrains the logic of information filtering in distinctive and consequential ways: professional norms developed under conditions of institutional autonomy are compelled to coexist with platform-based algorithmic constraints operating according to fundamentally different logics of value [34]. Table 3 compares the principal gatekeeping mechanisms across the major platform types that now structure the information ecosystem [9,13].

Comparative evidence indicates that each architecture produces a distinctive information ecology. In the case of Meta and X, gatekeeping is enacted primarily through content moderation policies and the selective algorithmic weighting of content according to engagement signals and policy compliance [37]. Algorithm-audit research provides direct evidence of the material consequences of this weighting: Bandy and Dimakopoulos, comparing algorithmically curated with chronological Twitter timelines, found that the curation algorithm substantially altered users' media exposure, surfacing more accounts but fewer external news links and shifting the source composition of what users encountered [1]. For algorithm-intensive platforms such as TikTok, the gatekeeping mechanism is embedded within the recommendation engine itself, which curates content aligned with personal identities, preferences, and behavioral histories through what Bucher analyzes as programmed sociality — the algorithmic shaping of social relations and information exposure according to computational logics largely invisible to users [6]. This represents a qualitative departure from legacy gatekeeping logic. Where traditional editors selected content against professionally defined criteria of newsworthiness — timeliness, proximity, impact, prominence — the recommendation engine selects content on predicted individual relevance, substituting individualized pertinence for collective newsworthiness at the foundational level of information filtering. Critically, the inference of relevance is itself an act of gatekeeping: Thorson and colleagues demonstrate that Facebook's algorithmic classification of users as politically interested or uninterested becomes a self-fulfilling filter that determines whether political content reaches them at all [45].

The cumulative behavioral consequences of these architectural shifts are now visible in cross-national audience data. The Reuters Institute's forty-eight-market survey documents that social media and video networks reached 54% of Americans as a news source in 2025, overtaking both television news (50%) and news websites and applications (48%) for the first time, while personality-driven creators and podcasters increasingly rival institutional newsrooms as gateways to public affairs and roughly 7% of respondents globally — 15% among those under twenty-five — already use artificial intelligence chatbots to access news weekly [33]. Figure 1 visualizes this reconfiguration of the principal gateways through which publics now encounter news [33].

The agenda-setting implications of this substitution are profound. When the primary criterion of information visibility is individual relevance rather than collective importance, the structural conditions for the formation of a shared public agenda are undermined [44]. The classical agenda-setting model presupposed a common informational environment in which diverse publics encountered a broadly similar set of issues, enabling collective judgment about matters of shared concern. Algorithmic personalization fragments this common environment into millions of individualized micro-environments, each reflecting the inferred preferences, behaviors, and demographic characteristics of its user [36,46]. The public agenda, under these conditions, is no longer a collective construct but an aggregation of individual agendas whose points of convergence are incidental rather than structurally produced. Incidental exposure — encountering news while using platforms for other purposes — partially offsets this fragmentation, but such exposure is itself algorithmically conditioned and unevenly distributed across platforms and populations [12].

5.2. Algorithmic Bias, Salience Distortion, and Democratic Cognition

Algorithmic bias — the systematic and inequitable distortion of information visibility resulting from the design, training, or deployment of automated systems — carries direct and measurable consequences for the construction of public salience hierarchies. The empirical literature demonstrates that algorithms do not filter information neutrally; they prioritize on the basis of features and optimization objectives that may diverge substantially from criteria of public interest, democratic relevance, or informational accuracy. Guess and colleagues demonstrate that algorithmically mediated exposure to untrustworthy websites during the 2016 United States election was concentrated among specific demographic groups, with older and more ideologically extreme users disproportionately exposed to low-quality information [17]. The landmark 2020 election collaboration between academic researchers and Meta subsequently provided experimental evidence at unprecedented scale: González-Bailón and colleagues documented pronounced asymmetric ideological segregation in political news exposure on Facebook, with a large share of misinformation-labeled content circulating within homogeneously conservative audience segments [16], while Guess and colleagues found that replacing the ranking algorithm with a reverse-chronological feed substantially changed what users saw — increasing exposure to untrustworthy content while decreasing overall platform use — even though the three-month intervention did not measurably alter political attitudes [18]. Table 4 summarizes selected empirical evidence of algorithmic salience distortion across contexts and methodologies [1,3,4,8,16,17,18,43].

Two features of this evidence merit emphasis. First, the technical mechanisms of algorithmic bias are increasingly visible through methodological innovation. Dimakopoulos documents how algorithm-auditing techniques — sock-puppet audits, crowdsourced audits, reverse engineering — have enabled researchers to identify the input variables that determine visibility outcomes in news recommendation systems [9]. Large-scale analyses of language models reveal that these systems amplify biases embedded in training data, including biases of gender, race, institutional prestige, and political orientation, in ways that are systematic yet difficult to detect through conventional quality assurance [3]. As generative systems are increasingly deployed to summarize and curate news, the resulting distortion of salience acquires a structural character that transcends individual instances of bias. Second, the relationship between algorithmic mechanism and attitudinal outcome is more complex than early filter-bubble accounts assumed. The 2020 election experiments found that reducing exposure to like-minded sources did not measurably reduce affective polarization over the study window [35], and Metzler and Garcia argue on this basis that algorithmic effects must be understood as interacting with — rather than overriding — the social and psychological drivers of selective exposure [29]. The theoretical implication is not that algorithmic gatekeeping is inconsequential but that its consequences operate primarily at the level of exposure architecture, cumulative over time and unevenly across subpopulations, rather than through immediate attitudinal conversion.

Evidence from national media systems corroborates the structural reading. Analysis of United Kingdom election coverage shows that ownership structures and platform-driven logics of content optimization materially influence the salience of political candidates and their attributes [8]. Benkler and colleagues demonstrate that the networked architecture of American political communication, combined with platform amplification dynamics, facilitated an asymmetrically polarized media ecosystem in which disinformation achieved national salience [4]. Starbird and colleagues document how participatory disinformation campaigns exploit algorithmic amplification to elevate fringe interpretations of mass-casualty events into publicly salient counter-narratives [43]. Collectively, these findings indicate that in the algorithmic environment the classical transfer of salience identified by McCombs and Shaw [25] is simultaneously a transfer of bias, wherein the optimization objectives of the algorithmic gatekeeper become, through recursive feedback, the cognitive priorities of the public.

5.3. Citizen Journalism, Decentralized Networks, and Counter-Hegemonic Agenda-Setting

The emergence of citizen journalism and distributed news networks offers a counter-narrative to the concentration of gatekeeping power in algorithmic platforms, demonstrating that the digital environment retains meaningful — if structurally constrained — possibilities for participatory and counter-hegemonic agenda-setting. In contexts where institutional media are constrained by state censorship, commercial capture, or platform governance policies, distributed gatekeeping has enabled insurgent agendas to surface and achieve public visibility.

The empirical record provides numerous illustrations. In China during the early 2000s, netizens used blogs to construct an alternative information ecosystem that challenged the narrative monopoly of state media, demonstrating the capacity of decentralized communication to open spaces for political discourse even within authoritarian environments [21]. The Arab Spring uprisings provide a particularly instructive case: civil society actors deployed decentralized communication technologies to resist autocratic control over the information environment and construct political agendas independent of state channels [22]. Palestinian communicators have developed adaptive strategies for maintaining informational presence despite restrictive platform moderation [41], and citizen curators shaped the informational contest surrounding the Ukraine conflict [15]. Collaborative platforms exhibit a further variant: Wikipedia interacts with global news media to generate what Mesgari and colleagues describe as a collaborative knowledge ecosystem, in which topic salience is determined through the aggregated judgments of thousands of contributors rather than the decisions of a small number of institutional gatekeepers [28].

The evidence also reveals significant limitations of citizen journalism as an alternative to institutional gatekeeping. During the COVID-19 pandemic, audiences demonstrated continued and often increased reliance on traditional news media for verification and editorial safeguarding, suggesting that professional gatekeeping retains a form of public legitimacy that citizen journalism has not achieved [32]. Moreover, even the most structurally distributed networks are subject to what might be termed shadow gatekeeping. Shaw's study of an open online collective shows that while participation is actively encouraged through platform design, the salience of contributions is ultimately determined by structural features including administrator visibility decisions and algorithmic content ordering [38]. Kreiss and McGregor demonstrate that technology firms shape the parameters of political communication through infrastructure provision, data analytics, and partnership arrangements that concentrate consequential authority within platform companies even as surface dynamics appear decentralized [23]. The decentralized surface of participatory platforms frequently conceals concentrated decision-making authority, reproducing in new technological forms the gatekeeping hierarchies it ostensibly displaces [52].

These findings suggest that the digital environment presents not a binary between centralized institutional gatekeeping and decentralized participatory openness but a complex ecology in which multiple modes of gatekeeping and agenda-setting coexist, interact, and compete. The challenge for theory is to capture this complexity without either romanticizing the democratizing potential of decentralized networks or dismissing them as epiphenomenal to algorithmic platform power.

6. Synthesizing Gatekeeping and Agenda-Setting in the Algorithmic Ecosystem

In the algorithmic media landscape the boundaries between gatekeeping and agenda-setting have become progressively blurred. Traditional models treated the processes as distinct: gatekeeping involved the selection and filtering of information by journalists and editors, while agenda-setting referred to the prioritization of issues for public attention. The rise of algorithmic platforms has reconfigured these roles, embedding both selection and salience within complex automated systems. This section synthesizes the two traditions, arguing that the recursive, distributed, and technically mediated mechanisms of information visibility require integrated theoretical vocabulary.

6.1. The Convergence Thesis

The empirical patterns examined above collectively substantiate the central argument of this paper: the digital transformation has produced a functional convergence of gatekeeping and agenda-setting that demands theoretical integration. In the algorithmic environment, filtering information and conferring salience upon it are no longer sequential processes performed by distinct actors in distinct institutional settings. They are simultaneous, recursive, and frequently indistinguishable operations performed by computational systems whose design and optimization objectives embed specific — if often unacknowledged — assumptions about what information should be visible and to whom [6].

This convergence is manifest across multiple analytical dimensions. At the level of mechanism, recommendation systems simultaneously perform the gatekeeping function of selecting which content is presented and the agenda-setting function of determining its relative prominence and perceived importance. At the level of structure, platform architecture constitutes the institutional framework for both filtering and salience conferral, displacing the newsroom as the primary site at which the conditions of public visibility are determined [31]. At the level of outcome, personalized information environments simultaneously reflect the algorithm's filtering decisions and shape the user's salience perceptions in a recursive loop that defies the linear logic of classical models.

The implications for existing frameworks are substantial. Classical gatekeeping theory assumed the gate was a discrete decision point at which information was admitted or rejected by an identifiable human agent. In the algorithmic environment, the gate is a continuous probabilistic process in which information is assigned a visibility score determining its position within a personalized feed — a position that changes dynamically with real-time engagement signals [9]. The moderation repertoire itself has shifted from exclusion to modulation: platforms increasingly govern by adjusting visibility rather than admitting or removing content [14,51]. Classical agenda-setting theory assumed salience was transferred from a unified media agenda to a mass public through repeated common exposure. In the algorithmic environment there is no unified media agenda, no mass public, and no common content — only a multiplicity of algorithmically constituted micro-agendas tailored to individual users on predicted relevance and engagement potential [46].

6.2. Salience Agency: An Integrative Framework

The foregoing analysis motivates the introduction of salience agency as a unifying conceptual framework. Salience agency is defined as the capacity of any actor — human, institutional, or algorithmic — to influence the visibility, prominence, and perceived importance of information within the public communication system. Unlike traditional conceptualizations of gatekeeping authority, which locate agencies in identifiable decision-makers within institutional hierarchies, salience agency is distributed, recursive, and emergent. It encompasses the micro-gatekeeping acts of individual users (sharing, liking, commenting), the meso-level curation performed by platform algorithms (recommendation, moderation, ranking), and the macro-level influence of institutional media, regulatory bodies, and civil society organizations. Table 5 sets out the five analytical dimensions of the framework [2,6,40].

The concept addresses three analytical gaps in existing literature. First, it provides a vocabulary for describing the functional equivalence of diverse gatekeeping acts that operate at different scales and through different mechanisms but produce converging effects on public salience hierarchies. A journalist's editorial decision, an algorithm's ranking calculation, and a user's decision to share a post are ontologically distinct acts, yet each contributes to the same aggregate outcome: the construction of what the public perceives as important. Second, salience agency captures the recursive character of contemporary information flows, in which the outputs of gatekeeping and agenda-setting feed back into the inputs that drive subsequent iterations. Third, the framework accommodates the normative heterogeneity of the contemporary ecosystem, in which professional journalistic ethics, platform terms of service, community norms, and individual preferences simultaneously govern — and compete over — the criteria by which information achieves visibility.

Operationally, salience agency can be analyzed across the five dimensions set out in Table 5. By mapping empirical cases across these dimensions, researchers can generate systematic comparative analyses of salience agency across platforms, political contexts, and historical periods, identifying patterns of concentration, contestation, and change that remain invisible within the separate frameworks of gatekeeping and agenda-setting theory alone.

6.3. Recursive Dynamics: The Salience Loop

The recursive character of information flows in the algorithmic environment deserves particular theoretical emphasis, as it represents perhaps the most fundamental departure from the linear, sequential logic of classical models. In the legacy environment, the flow of influence was predominantly unidirectional: media organizations selected and presented information, audiences formed salience perceptions on that basis, and audience feedback — ratings, circulation figures, letters to the editor — influenced subsequent editorial decisions only indirectly and with significant delay. The information cycle was measured in days and weeks, and each cycle was largely independent of its predecessors.

In the algorithmic environment this cycle has been compressed to seconds and made continuously recursive. An algorithm presents content to a user; the user's engagement or non-engagement generates data that feeds back into the algorithm, modifying the criteria by which subsequent content is selected and ranked; the modified algorithm presents new content reflecting the user's revealed preferences as interpreted by the system; and the cycle repeats, continuously and in real time [6,36]. Each iteration simultaneously performs gatekeeping and agenda-setting, with the output of each cycle serving as the input to the next. Figure 2 depicts this recursive architecture, which this paper designates the salience loop [6,36,46].

The recursive dynamic has three consequential properties. The first is path dependence: because each iteration builds on the outputs of previous iterations, initial conditions — the user's first platform interactions, the content available at the moment of initial engagement, the algorithmic parameters then in effect — exert disproportionate influence on subsequent exposure, a mechanism empirically visible in the self-reinforcing classification dynamics documented on Facebook [45]. The second is convergence: optimization objectives designed to maximize engagement tend to drive the loop toward content that is emotionally arousing, ideologically confirming, or socially validating, regardless of informational accuracy or democratic relevance [44,52]. The third is opacity: because the loop operates within proprietary systems inaccessible to external scrutiny, the processes through which public salience is constructed are effectively invisible to the publics whose cognitive priorities they shape [9]. Table 6 contrasts the classical and algorithmic models of communication across their principal structural features [6,25,40,46].

The contrast summarized in Table 6 should not be read as a claim that the classical model has been wholly superseded. The hybrid media system ensures that editorial and algorithmic logics coexist and interpenetrate [7], and the experimental literature cautions against technologically determinist readings in which the algorithm unilaterally manufactures public opinion [29,35]. The claim, rather, is architectural: the infrastructure within which all salience-seeking actors — journalists, politicians, activists, ordinary users — must now operate is one whose default dynamics are recursive, personalized, and computationally governed.

7. Discussion: Implications for Theory, Practice, and Policy

The findings and conceptual advances presented above carry implications across three domains: communication theory, media management and professional journalism, and regulatory policy. As gatekeeping and agenda-setting converge within algorithmic environments, the recursive dynamics of salience agency reshape how information is filtered, prioritized, and made visible — and therefore how the construction of public knowledge must be understood, managed, and governed.

7.1. Theoretical Implications

The convergence documented in this paper necessitates a fundamental reorientation of communication theory. The salience agency framework represents one such reorientation, but its implications extend beyond the specific concepts it introduces. The analysis indicates that scholars should shift from examining discrete media effects toward studying systemic dynamics: how information ecosystems are constituted, and how public knowledge and public ignorance are collectively produced. The 2020 election experiments illustrate the stakes of this shift. Their finding that feed interventions changed exposure without changing short-run attitudes [18,35] is anomalous within a linear effects paradigm but intelligible within a systemic one: the power of algorithmic gatekeeping lies less in immediate persuasion than in the long-run architecture of visibility — path-dependent, cumulative, and unevenly distributed across the population [16,29].

This shift requires methodological as well as theoretical innovation. The recursive dynamics of the salience loop cannot be adequately captured by the cross-sectional surveys and content analyses that have dominated agenda-setting research, nor by the ethnographic and interview methods characteristic of gatekeeping research. Computational methods — agent-based modeling, network analysis, algorithm auditing, large-scale digital trace analysis — are essential for tracking the real-time processes through which salience is constructed [1,9,48]. At the same time, computational methods must be integrated with qualitative approaches capable of capturing the interpretive, cultural, and political dimensions of information filtering that purely quantitative analyses risk obscuring. The future of research in this domain lies in mixed-method designs combining computational rigor with interpretive depth.

The analysis also highlights the need for greater attention to the normative dimensions of communication theory. Classical gatekeeping and agenda-setting research was predominantly descriptive, identifying the mechanisms of filtering and salience conferral without evaluating their democratic adequacy. In an environment where the processes governing public information are opaque, commercially motivated, and demonstrably capable of distorting deliberation, purely descriptive approaches are insufficient. Communication theory must develop normative frameworks capable of evaluating the democratic legitimacy of different configurations of salience agency — frameworks that specify not only how salience is constructed but how it ought to be constructed in a society committed to informed self-governance [44].

7.2. Implications for Media Management and Professional Practice

For media managers and professional journalists, the convergence presents both challenges and opportunities. The challenge is institutional adaptation: organizations developed for an environment of concentrated editorial authority must now operate within an ecosystem in which that authority is distributed across a complex assemblage of human and algorithmic actors, and in which their own distribution depends on intermediaries they do not control [27,34]. This requires new technical competencies — platform analytics, algorithmic optimization, digital audience engagement — and the negotiation of profound tensions between journalistic norms of public service and the engagement-driven logics of platform distribution.

The opportunity lies in the distinctive value professional journalism can offer within the algorithmic ecosystem. In an environment saturated with unverified, algorithmically amplified content, the capacity for independent verification, contextual analysis, and accountable editorial judgment constitutes a form of institutional legitimacy that algorithmic systems cannot replicate. The COVID-19 pandemic demonstrated that audiences value this capacity and return to professional institutions when the stakes of accuracy are perceived as high [32]. Yet the 2025 audience data counsel against complacency: as personality-driven creators and AI chatbots claim growing shares of news attention [33], institutional journalism's gateway position is eroding even where its epistemic authority persists. The concept of entrepreneurial journalism [49] provides a useful framework for adaptation: by developing platform-literate editorial strategies that leverage algorithmic dynamics while maintaining commitments to accuracy, fairness, and public service, media organizations can position themselves as salience intermediaries — actors that mediate between the engagement logic of platforms and the deliberative requirements of democratic publics.

7.3. Policy and Regulatory Implications

The concentration of salience agency in a small number of technology companies whose algorithmic systems govern information visibility for billions of users represents a historically unprecedented concentration of communicative power [52]. This concentration is not adequately addressed by regulatory frameworks developed for a media environment of institutional pluralism, editorial accountability, and relatively transparent decision-making.

The European Union's Digital Services Act [10] and Artificial Intelligence Act [11] represent significant regulatory innovations, establishing requirements for algorithmic transparency, content moderation accountability, researcher data access, and the assessment of systemic risks posed by very large online platforms. Their effectiveness, however, depends on enforcement mechanisms capable of monitoring algorithmic behavior at scale and assessing the democratic consequences of governance decisions that are technically complex, commercially sensitive, and continuously evolving. Comparative research suggests that regulatory effectiveness depends critically on technical expertise within agencies, platform cooperation with external auditing, and independent research infrastructure for ongoing assessment of algorithmic impact [30,31]. The salience agency framework may be of practical value here by providing a common analytical vocabulary for assessing how platform architectures, algorithmic designs, and governance policies distribute the capacity to influence visibility — and whether the resulting distribution is compatible with democratic norms of communicative pluralism.

The cultivation of algorithmic literacy among citizens is a complementary priority. Just as media literacy education sought to equip citizens to evaluate media messages critically, algorithmic literacy would equip them to understand how their information environments are constructed, the commercial and political interests shaping those mechanisms, and the strategies available for diversifying exposure. Research on personalization effects suggests that without such literacy, the recursive dynamics of the salience loop will continue to narrow information exposure in ways that undermine the epistemic foundations of deliberation [17,36,45]. Because the loop is distributed and technically mediated, interventions at any single point — technical, professional, regulatory, or educational — are unlikely to suffice alone; sustained progress requires coordinated, cross-sectoral effort integrating expertise from computer science, social science, law, and media practice.

8. Conclusion

The trajectory identified in this paper — from human-controlled gates to algorithmically modulated visibility, from collective public agendas to personalized micro-agendas, from institutional accountability to platform opacity — represents not the end of gatekeeping and agenda-setting as communicative phenomena but their transformation into forms that demand new theoretical tools, new empirical methodologies, and new normative commitments. The quarter-century examined here has witnessed a structural reconfiguration of the processes through which democratic publics come to perceive certain issues as important and others as invisible, and this reconfiguration carries consequences extending far beyond academic communication theory.

The salience agency framework offers one integrative vocabulary for analyzing the distributed, recursive, and technically mediated processes that characterize the algorithmic environment. By reconceptualizing gatekeeping and agenda-setting as dimensions of a single unified process — the construction of information visibility — the framework enables a more comprehensive understanding of how public salience is produced, contested, and transformed. Its five analytical dimensions — agent identity, mechanism of influence, scale of effect, degree of transparency, and normative orientation — provide a structured basis for empirical investigation and comparative analysis across platforms, political contexts, and historical periods.

The evidence reviewed demonstrates that the algorithmic mediation of information visibility is neither politically neutral nor democratically benign. Algorithmic gatekeeping systematically favors engagement over accuracy and commercial profitability over public interest [4,52]; its recursive dynamics create self-reinforcing patterns of exposure that narrow rather than broaden the epistemic horizons of democratic citizens; and the concentration of salience agency in a small number of opaque technology companies poses a structural challenge to the communicative foundations of self-governance. At the same time, the most rigorous recent evidence warns against fatalism: algorithmic architectures shape exposure powerfully but do not mechanically dictate belief [18,35], citizen networks retain real if constrained counter-hegemonic capacity [21,22], and regulatory instruments capable of mandating transparency now exist [10,11].

Addressing the challenge will require sustained effort across multiple domains: scholars must build frameworks and methods adequate to the complexity of the algorithmic environment; media professionals must negotiate platform logics without subordinating journalistic values to algorithmic imperatives; policymakers must ensure transparency, accountability, and pluralism in the distribution of salience agency; and citizens must acquire the algorithmic literacy necessary to understand and shape the information environments in which their political judgments are formed. The convergence of gatekeeping and agenda-setting in the algorithmic age is, at its core, a question about the conditions under which democratic publics can form the informed, shared judgments on which self-governance depends. The adequacy of the salience agency framework will be determined not by its theoretical elegance but by its capacity to generate empirical insight, inform practical intervention, and contribute to the construction of information environments worthy of the democratic publics they serve.

9. Recommendations for Future Research

The analysis presented in this paper suggests five priorities for the next phase of scholarship. First, longitudinal designs are needed to test the cumulative-exposure hypothesis that emerges from the reconciliation of the experimental and observational literatures: if algorithmic gatekeeping operates through path-dependent exposure architecture rather than short-run persuasion [18,29,35], then studies spanning years rather than months — tracking cohorts across platform environments — are required to capture its effects on public agendas. Second, the salience agency framework should be operationalized and validated: comparative studies mapping empirical cases across its five dimensions could establish whether the framework discriminates meaningfully among platforms, national contexts, and historical periods. Third, research must extend to the generative-AI layer of the ecosystem, as chatbots and AI-mediated summaries become news gateways for younger users [33]; the gatekeeping and agenda-setting properties of conversational systems — which compress the entire salience loop into a single opaque interlocutor — remain almost entirely unstudied [3]. Fourth, the field requires systematic non-Western and comparative research, since the platform experiments that anchor current knowledge were conducted almost exclusively in the United States [16,18,35], while the governance dynamics documented in China, the Arab world, Ukraine, and Palestine [15,21,22,41] indicate that salience agency is configured differently under different political regimes. Fifth, regulatory scholarships should evaluate the Digital Services Act's researcher-access and risk-assessment provisions [10] as a natural experiment in mandated transparency, generating evidence on whether disclosure regimes materially alter the distribution of salience agency.