To ensure a comprehensive and exhaustive capture of the relevant academic literature, our search strategy was systematically implemented across two primary, high-impact academic databases: Scopus and Web of Science (WoS Core Collection). This strategic selection was based on their extensive coverage of communication and media studies journals. The initial search yielded a total of 1,084 records: 893 from Web of Science and 191 from Scopus. The final search was conducted on 03 September 2025.

The search queries were meticulously constructed using Boolean operators to combine controlled vocabulary and free-text terms related to “digital journalism,” “news production,” and “media legitimacy,” with platform- and algorithm-specific terms including: “algorithm,” “recommendation,” “ranking,” “news feed,” “Facebook,” “X/Twitter,” “YouTube,” “TikTok,” and “Instagram” (Bramer et al., 2018; Spencer and Eldredge, 2018). and expert consultation. The full search strings are available in Table 1 and the public data repository.

To enhance the precision of the search results, filters were applied within each database to exclude non-article document types (e.g., book chapters, conference proceedings), non-English publications, and outdated records outside the 2015–2025 window. Subject areas unrelated to journalism and communication were excluded based on predefined Web of Science categories, and decisions regarding Open Access status were recorded. These filters were set a priori and documented transparently (Table 1). While exclusions based on subject and access type are non-standard, they were justified to focus the review on relevant empirical literature and reduce disciplinary noise.

The screening and selection of studies followed PRISMA 2020 guidelines. From 1,084 initial records, 9 duplicates, 48 auto-screened, and 9 other ineligible items were removed. Of 392 screened titles/abstracts, 214 were excluded. 178 full-text reports were retrieved and assessed, yielding 78 included studies. Disagreements between the two reviewers were resolved via consensus discussions. Inter-rater reliability was calculated as Cohen’s κ = 0.82, indicating strong agreement. Study counts were harmonized across all manuscript sections and the PRISMA diagram (Figure 1).

Data were extracted independently by two reviewers using a predefined template covering bibliographic info, platform(s), methods, sample, geography, findings, and limitations. A 10% pilot ensured clarity. Coding used a hybrid thematic approach: deductive themes based on theory (Section 3) and inductive codes emergent from data. The final codebook is in Appendix Findings Review. A summary of all 78 included studies is provided in Supplementary Findings Review.

A 10% pilot extraction was conducted on a subset of the included studies prior to the full data extraction phase. This pilot aimed to refine the template fields and ensure the clarity and consistency of code definitions. The coding process itself employed a hybrid thematic analysis approach. This involved starting with deductive themes that were pre-specified based on the research questions and the theoretical framework (outlined in Section 3). These deductive themes were then complemented by an inductive process of identifying new, emergent sub-codes and patterns directly from the extracted data. The finalized codebook, complete with definitions and examples, is provided in Appendix A (Rodriguez et al., 2022; Tam et al., 2017), ensuring the transparency and reproducibility of the coding process. A comprehensive table summarizing the characteristics of all 78 included studies is also provided separately in Supplementary Table Findings Review.

Each study was appraised using appropriate tools: CASP for qualitative/mixed-methods, and Risk-of-Bias frameworks for quantitative/observational designs (Juniardi and Putra, 2024; Shea et al., 2017). Two reviewers conducted this independently. Inter-rater reliability was high (κ = 0.82). Per-study ratings appear in Supplementary Table S1. Appraisal scores informed the synthesis process via evidence weighting and sensitivity analysis (e.g., excluding low-quality studies to test robustness).

Due to high heterogeneity (platforms, regions, methods), a narrative thematic synthesis was employed. Four themes guided analysis: (1) algorithmic influence on editorial autonomy, (2) commercialization of news production, (3) platform legitimacy, and (4) amplification of polarization and misinformation. To assess evidence distribution and claim strength, an evidence map was generated, cross-tabulating methods, platforms, regions, and outcomes. Access to the synthesis scripts and coded data is provided in the repository.

All materials—search strings, PRISMA logs, screening sheets, extractions, codebooks, quality ratings, and synthesis scripts—are publicly available in the linked data repository. This ensures full reproducibility and auditability of the review. Harmonized counts from all sections are included.

As a systematic review of published literature, this study did not require ethical approval. This review did not require ethical approval. However, principles of transparency and reflexivity guided all decisions. Potential biases (e.g., language restriction, regional gaps, exclusion rationale) are addressed in Section 5: Limitations.

The concept of platformization offers a critical lens through which to examine how digital platforms have become central to journalistic production, distribution, and audience engagement. In the context of a “platform society,” platforms are increasingly understood as overarching infrastructures that shape communication norms and practices (Poell et al., 2020). For journalism, this is acutely evident in the escalating reliance on platforms such as Facebook, X (formerly Twitter), Instagram, TikTok, YouTube, and Twitch for essential functions like content dissemination and audience reach (Al-Zoubi et al., 2023; Burgess and Hurcombe, 2019; Chua and Westlund, 2022; McGregor and Molyneux, 2020; Swart, 2021). This increasing integration signifies a fundamental shift in the journalistic ecosystem, moving from more traditional, structured news flows towards dynamic, interactive, and often real-time environments dictated by platform affordances.

Platformization inherently integrates distinct economic, technological, and social logics into newsroom routines and practices. The prevailing commercial imperatives within this model often compel news organizations to drive alignment with platform-specific visibility and engagement metrics (Poell et al., 2020). This necessitates a reframing of professional autonomy, wherein editorial judgments become increasingly calibrated to algorithmic performance indicators rather than solely relying on traditional normative news values (Carlson, 2019; Chiridza and Mare, 2025). These insights are instrumental in the development of the “Commercialization/Platformization” code family used to structure comparisons across different organizational types, audience demographics, and geographical regions throughout this review.

Furthermore, platformization carries significant potential to contribute to what has been termed “data colonialism,” a phenomenon wherein journalistic activities become increasingly embedded within extractive datafication economies (Couldry and Mejias, 2019). This concept was operationalized in our coding structure through the “Platformization/Commercialization” category and shaped our interpretation of regional asymmetries (Sections 4.2 and 4.3). It also informed how platform logic was evaluated during the quality appraisal stage (Section 2.5), particularly regarding commercial influences on editorial practices.

The concept of gatekeeping, traditionally understood as the process by which editors and journalists filter information flows, is undergoing a significant evolution in the contemporary platformed news ecosystem. In our analysis, “Gatekeeping/Algorithmic Gatekeeping” was applied as a key deductive code to classify how algorithmic systems mediate visibility, news values, and editorial control, especially in relation to metric-based decision-making.

Algorithmic curation in digital journalism is thus conceptualized through the lens of recommender systems and their underlying visibility logics (Kaiser and Puschmann, 2017). Complementing this understanding, Actor-Network Theory (ANT) offers a valuable framework. These theoretical concepts were translated into the analytical framework through the construction of dedicated code families (e.g., “ANT/Assemblages,” “Metrics/Dashboards”), facilitating a granular examination of empirical variations in newsroom agency and adaptation strategies.

The Social Shaping of Technology (SST) theory further enriches this perspective by highlighting how cultural, economic, and political values become embedded within algorithmic designs. In our coding, SST-informed analysis helped to reveal how algorithmic affordances reflect deeper structural biases. This was particularly salient in analyzing commercialization pressures, coded under “SST/Platformization,” and subsequently integrated into the synthesis of Theme 2 (Section 4.2).

Traditionally, journalism’s legitimacy has been predicated on foundational principles such as objectivity and public trust. In this review, media legitimacy was not only examined conceptually but also operationalized through a dedicated code family (“Legitimacy/Trust”) used during thematic synthesis. This enabled systematic tracking of how algorithmic systems influence perceived trustworthiness, across both audience and journalistic perspectives.

Historically, media legitimacy was anchored in institutional norms. In the current era, algorithmic mediation has reshaped those conditions. Our coding captured both trust-eroding dynamics (e.g., opacity, personalization concerns) and mitigation mechanisms (e.g., transparency features, user controls). These variations were mapped in our evidence synthesis and visualized in the evidence map (Section 4.4).

Furthermore, scholarly debates increasingly foreground issues of bias and embedded incentives. In our synthesis, we differentiated between trust erosion due to algorithmic opacity and trust reinforcement due to transparency-oriented innovations, treating each as distinct sub-codes. These distinctions shaped both the interpretive framing of our conclusions and the weighting of evidence in the synthesis (see Table 2).

This section provides the theoretical scaffolding necessary to understand how algorithmic influence on news production impacts media legitimacy. Crucially, these theories were not merely reviewed conceptually but were actively operationalized within our analytical framework through code development, theme refinement, and synthesis structuring. This integration ensures that empirical patterns are interpreted through well-established theoretical lenses, enhancing the validity and coherence of the review’s conclusions.

This section synthesizes the evidence summarized in Table 3 (Theme 1) and explicitly maps the findings to the theoretical scaffolding and coding structure introduced in Section 3. Each pattern is connected to specific code families—“Algorithmic Gatekeeping,” “Platformization/Commercialization,” “ANT/Assemblages,” and “SST/Platformization”—ensuring traceability between theory, empirical data, and interpretation. The synthesis was conducted via hybrid thematic analysis, blending deductive themes with inductively surfaced sub-codes. Across the 30 studies inventoried in Table 1, the core pattern is consistent: algorithmic curation and metricization do not merely “pressure” editorial decision-making; they reconfigure it. This reconfiguration is visible in routinized metric work, accelerated temporalities, and recalibrated notions of newsworthiness toward platform-compatible “shareworthiness,” while leaving bounded spaces for professional judgment and strategic resistance.

First, the studies converge on a redistribution of gatekeeping authority, which was consistently coded under “Algorithmic Gatekeeping” and “Assemblages.” Ethnographic and survey-based work shows that real-time analytics are operationalized as boundary objects in newsrooms, aligning daily choices with performance signals (Conyers, 2025; D’Amico et al., 2023; Sehl et al., 2024). Experimental and platform-analytic evidence reinforces that feed ranking and personalization narrow the editorial “window,” biasing selection toward items expected to perform under algorithmic logics (Dodds et al., 2023; McGregor and Molyneux, 2020). Studies of newspay models and micro-segmentation add that revenue instrumentation can tilt calendars and formats toward calculable, low-risk outputs (Myllylahti, 2020, 2024). Taken together, these results empirically instantiate algorithmic gatekeeping and ANT’s distributed agency: human editors, metrics, interfaces, and business rules co-produce editorial outcomes rather than technology simply “overriding” journalists. Within SST, this co-production reflects embedded commercial values that privilege calculability and control (Çifçi and Ayhan, 2024; Cohen, 2019; Creech and Nadler, 2018).

Second, algorithmic visibility logics compress verification windows and accelerate newsroom temporalities—a pattern captured under the “Temporal Compression” sub-code. Evidence from X/Twitter shows wire-like reliance on trending signals that favors speed over depth (Boling and Walsh, 2025; McGregor and Molyneux, 2020). Computational diffusion analyses indicate that negative news and personality-driven stories spread faster, incentivizing timeliness and viral frames (Buhl et al., 2019). Platform-specific studies of YouTube document optimization toward monetization/discovery, with attendant impacts on packaging and cadence (Cheng and Tandoc, 2021). Stimulus-based interviews further reveal how affordances across TikTok, Instagram, and Facebook shape coordination and selection decisions on the desk (Anter, 2025). Regionally, newsroom practices in the Global South incorporate informal metrics and WhatsApp circuits This confirms that algorithmic influence is contextually mediated rather than universally deterministic (Omanga et al., 2023). These patterns substantiate the hypothesis (H1) that algorithmic curation materially reshapes editorial choices by structuring attention, timing, and visibility.

Third, the empirical corpus links metricization to normative and epistemic tensions in media legitimacy, categorized under the “Trust/Transparency” code family. Cross-national survey evidence shows that higher perceived use of news recommender systems (NRS) is associated with lower trust in outlets, moderated by perceived benefits/concerns (Blassnig et al., 2024). This dovetails with the review’s broader claim that opacity depresses legitimacy while communicative transparency can mitigate skepticism (see Section 3’s legitimacy discussion). Studies warn against technological determinism, urging nuanced, context-aware explanations of platform effects (Appelgren, 2023; Carlson, 2023). Crowdsourcing research illuminates a concrete trade-off: while open calls enhance knowledge discovery and tip flows, volume can erode verification, yielding blended responsibility between journalists and publics (Aitamurto, 2016). These findings reinforce H2’s moderation logic: transparency, explicability, and user control can soften but not eliminate trust risks arising from opaque curation.

Fourth, Table 3 documents organizational adaptation strategies, mapped to the codes “Professional Autonomy,” “Resistance,” and “Coping Mechanisms.” Ethnographies distinguish “metric confirmation” work (low-cost, high-gain) from riskier “journalistic discovery,” indicating how temporalities and incentives sort labor inside the desk (Conyers, 2025). Labour-process and intersectional accounts show intensification, commodification, and precarity, with differentiated burdens for women of color (Cohen, 2015; Cohen, 2019; Cohen and Clarke, 2024). Comparative work underscores newsroom strategies—diversifying content, advocating editorial independence, or selective resistance—to preserve judgment under platform dependence (Chua and Westlund, 2022; Eldridge et al., 2019). Studies also report “strategic ignorance” as a coping practice to manage the opacity and volatility of algorithmic systems (Christin et al., 2024). These results support a “bounded agency” reading that is compatible with ANT and consistent with the framework’s expectation that socio-technical contexts shape, but do not erase, professional autonomy.

Fifth, the studies identify design-level and pedagogical responses to algorithmic influence, classified under the “Reconfiguration/Design,” “Identity/Audience,” and “Reskilling” codes. Analyses of identity in news sharing show that political self-presentation structures dissemination practices even when mainstream outlets dominate link sources (Baas et al., 2025). Research quantifying journalistic values via textual indices finds measurable associations between linguistic features and perceived balance, diversity, importance, and factuality—suggesting feasible pathways for auditable quality signals compatible with recommender design (Choi, 2019). Case-based education and skills work indicate that cultivating cognitive flexibility and data-visualization literacy may buffer against the deskilling risks of automation and metricization (Breit, 2020). These strands connect directly to Section 3’s call for governance mechanisms that reward public-interest quality rather than pure engagement.

The cumulative implications of Theme 1 reinforce the theoretical coherence and methodological robustness of the review. Conceptually, the studies corroborate the framework Empirically, the synthesis privileges findings from higher-quality studies (as weighted via appraisal scores in Supplementary Table S1), and incorporates variation across geographies, platforms, and journalistic roles. Practically, the results justify three governance levers referenced in the overall framework: (i) routine exposure audits of recommender outcomes; (ii) auditable transparency and communication about NRS use; and (iii) multi-metric portfolios that elevate accuracy, diversity, and civic value alongside reach. Concretely, verification safeguards for crowdsourcing (Aitamurto, 2016), platform-affordance literacy for desk editors(Anter, 2025), and institutional protections for discovery work (Conyers, 2025), are prudent organizational responses.

Finally, scope conditions significantly shape the manifestation of algorithmic influence. As documented in Table 3 and visualized in the evidence map, the sample is skewed toward Western/English-language contexts and certain platforms (e.g., Twitter, Facebook), with limited representation of Reddit, LinkedIn, and Oceania (Badr, 2022; Chiridza and Mare, 2025). These disparities were acknowledged in the “Limitations” section and inform the interpretation of generalizability. The review therefore treats enabling effects—audience growth, novel formats—as real but context-bound and typically offset by trade-offs in depth, verification, and trust. Aligning editorial autonomy with legitimacy in a platformed environment requires moving beyond engagement-maximization toward transparent, auditable, and quality-sensitive systems, as theorized in Section 3 and operationalized via the evidence map.

Commercialization operates as a constitutive force in the platformized news ecology, shaping editorial judgment through embedded economic incentives. Reading Table 4—This synthesis follows a hybrid thematic coding strategy (deductive + inductive) and aligns with theory-informed code families—“Platformization/Commercialization,” “Metrics Governance,” and “Organizational Form.” market logics are embedded in interfaces, dashboards, and platform partnerships, thereby co-producing editorial outcomes with journalists and managers. In this framework, platforms do not merely host content; they mediate value by aligning visibility with monetizable engagement, narrowing the space for public-interest work unless counterbalanced by institutional safeguards and diversified revenue.

This section synthesizes evidence—coded under “Trust,” “Credibility Signals,” and “Platform Affordances”—to analyze how platform interfaces and ranking systems condition public trust, credibility, and authority claims in journalism. Reading Table 5—The synthesis builds on theory-driven coding and was triangulated across journalist and audience perspectives using a quality-weighted comparative approach. Platforms act as legitimacy infrastructures: their interfaces, metrics, and recommender hooks create cues that audiences use to infer credibility, while also re-framing what counts as legitimate performance inside newsrooms.

This theme synthesizes findings from 26 studies coded under Theme 4—categorized into “Polarization,” “Misinformation Amplification,” and “Editorial Risk Management”—to examine how algorithmic engagement logics intensify division, spread falsehoods, and constrain autonomy. The synthesis follows a theory-informed coding scheme and comparative appraisal method (see Table 6). Consistent with Section 3’s scaffolding. The evidence confirms that algorithms function as socio-technical agents—not neutral intermediaries—reconfiguring the visibility and legitimacy of journalistic content.

The corpus shows a clear Western and English-language bias. This reflects broader structural inequalities in academic publishing, where Global South perspectives often face linguistic, financial, or infrastructural barriers to inclusion. In addition, much of the literature treats algorithms as opaque “black boxes,” with limited technical characterization of recommender systems. This gap highlights the need for interdisciplinary collaboration with computer scientists to advance transparency in algorithmic auditing and methodological rigor.

Finally, limited attention is given to workforce diversity and equity, despite evidence that metricization disproportionately affects precarious and marginalized journalists. This blind spot curtails intersectional analysis of how algorithmic pressures disproportionately shape newsroom labor conditions for marginalized groups. The review process itself also presents constraints. Future systematic reviews should explicitly incorporate multilingual searches, grey literature databases, and region-specific repositories to mitigate these limitations.

Findings underscore three implications. In practice, this means embedding algorithmic transparency not only into content presentation but also into the internal governance of editorial tools and newsroom dashboards.

Policy interventions should prioritize co-regulatory models granting researchers access to platform data, coupled with obligations for transparency and accountability. Such models should be grounded in international standards for data access, algorithmic explainability, and ethical design, especially in regions with fragile media ecosystems. For news organizations, experimenting with alternative business models—membership schemes, philanthropy, or mixed revenue streams—may reduce dependence on engagement-driven metrics. Diversifying revenue can help mitigate the “platform trap,” where content decisions become subordinated to algorithmic distribution incentives.

This review identifies urgent priorities for the field. Participatory action research involving journalists, platform engineers, and civic actors could further enrich this knowledge base. Rigorous mixed-method designs that combine algorithmic audits, log data, and newsroom experiments are needed to establish causal mechanisms (McGregor and Molyneux, 2020). Longitudinal consortia tracking algorithmic changes and newsroom responses across diverse regions would provide broader generalizability (Zheng et al., 2021). Under-studied platforms and geographies require systematic inclusion. For example, platforms such as TikTok, Reddit, and encrypted messaging apps remain poorly represented despite their growing informational relevance. Equity-focused research must evaluate how algorithmic pressures intersect with gender, race, and precarity in journalism.

Finally, shared taxonomies and open data corpora are essential for cumulative research and replication. This requires cross-institutional coordination, robust metadata standards, and accessible repositories that support long-term research infrastructure.