War on News: Struggles Over Visibility
- War on News is a concept defining struggles over news visibility, legitimacy, framing, and circulation, shaped by algorithmic suppression and censorship amid conflict.
- Recent studies show that algorithmic interventions like Meta’s feed modifications depressed news engagement by about 78%, affecting all news quality tiers.
- Research reveals that news circulation and narrative framing reconfigure during conflicts, challenging credibility and enabling diverse forms of information warfare.
“War on News” denotes, in its narrowest documented usage, the multi-year period in which Meta “algorithmically deprioritized news and political content” on Facebook, but the research record also supports a broader understanding of the term as a struggle over news visibility, legitimacy, framing, and circulation under conditions of armed conflict, censorship, polarization, and platform governance (Talaga et al., 25 Jul 2025). In that broader sense, news is not merely a channel through which war, crisis, or political contest is described; it is itself a site of intervention. Platforms can suppress it, states and aligned media can weaponize it, audiences can re-route it through alternative infrastructures, and computational systems can measure how competing actors construct incompatible accounts of the same events (Mohanty et al., 22 Jan 2026).
1. Conceptual range and historical extension
In the Facebook case, “War on News” refers to a temporally bounded policy trajectory beginning around Facebook’s February 10, 2021 announcement that it would reduce political and civic content in News Feed, followed by further expansions in 2021, 2022, 2023, and 2024, and ending with Meta’s January 2025 reversal (Talaga et al., 25 Jul 2025). That is the most literal and operationalized use of the phrase in the materials.
By extension, the term also captures a wider class of struggles in which war or postwar politics reshape the conditions under which news is produced and trusted. Research on the Russia–Ukraine war describes a “second battlefield” in the online space, where social media is used both to garner support and to wage information warfare (Chen et al., 2022). Work on postwar Nigeria argues that propaganda techniques and spins are not limited to wartime but remain active in peace time, so that postwar journalism can function as an extension of civil-war politics by other means (Folayan et al., 2021). Studies of censorship circumvention during the 2022 Russian invasion show that when major platforms and news sites were blocked, users migrated war-related communication into Google Maps and Tripadvisor reviews, turning non-news affordances into improvised channels for wartime information (Moreno et al., 2023).
This suggests that “War on News” is not exhausted by any single mechanism. It can name direct suppression of visibility, but it can also name a recurrent condition in which news becomes a contested infrastructure: a field of ranking, access, narrative authority, and strategic substitution.
2. Algorithmic suppression and agenda compression
The most explicit empirical account of a war on news as platform governance comes from the analysis of Facebook between 2016 and early 2025. Using data from 40 national U.S. news organizations, 21 non-news pages, 5,243,302 Facebook posts from news outlets, 396,468 posts from non-news pages, and billions of user reactions, the study concludes that Meta’s feed algorithms substantially depressed news visibility from 2021 through 2024 (Talaga et al., 25 Jul 2025). The headline result is a decline of about in reactions to news posts during that period, with , while reactions to non-news posts increased symmetrically by about , with . The paper treats reactions as a proxy for visibility because the intervention targeted ranking rather than publishers’ ability to post.
Methodologically, the analysis separates supply from visibility. News outlets continued to post at stable or increasing levels, but reactions fell sharply. After accounting for strong temporal autocorrelation with an AR(1) process and an estimated , the correlations between weekly post counts and reaction counts were essentially nil across news quality tiers and for non-news. The study also combines BEAST changepoint detection, generalized linear mixed models, and multiplicative difference-in-differences contrasts. Its central comparison is formalized as
where non-news serves as the counterfactual baseline (Talaga et al., 25 Jul 2025).
A common misconception is that such deprioritization targeted only low-quality sources. The paper rejects that narrower interpretation. Low-quality sources were especially suppressed, but the effects extended across the news ecosystem: during most of the War on News period, low-, medium-, and high-quality sources converged to similarly depressed levels of reactions. The authors therefore interpret the intervention as broad suppression rather than tightly precision-targeted ranking (Talaga et al., 25 Jul 2025).
A related but distinct form of agenda compression appears in wartime coverage in emerging democracies. In Pakistani English-language newspapers during the May 2025 Indo-Pak conflict, war-related reporting increasingly overshadowed coverage of political opposition and dissent, with the sharpest shift during the war phase and persistence into the post-war phase (Bajwa, 24 Sep 2025). In this usage, the war on news is not a recommender-system intervention but a reordering of editorial priorities in which conflict crowds out democratic contestation.
3. Reconfigured circulation under censorship and conflict
Conflict frequently reorganizes where audiences get news and how news-like information is circulated. During the May 2021 Palestinian–Israel war, analysis of more than eight thousand Twitter users and over 29 million tweets found that users may consume more topically related content from foreign and less popular sources, because such outlets may reaffirm their views, offer more extreme, hyper-partisan, or sensational content, or provide more in depth coverage of the event (Darwish, 2021). The shift was strongest for pro-Palestinian users, among whom Al Jazeera, Middle East Eye, and Jewish Currents gained conflict-related attention at the expense of mainstream U.S. outlets perceived as misaligned with user stance.
Where conventional journalism is intimidated or delayed, citizens can become de facto war correspondents. In the Mexican Drug War, analysis of 609,744 tweets across four city hashtags showed that ordinary users used Twitter for real-time danger reporting and that a small number of highly active “civic media curators” emerged to aggregate and disseminate information to large urban audiences (Monroy-Hernández et al., 2015). The paper explicitly argues that these curators take on roles akin to public information officers or war correspondents under conditions of institutional weakness.
Under direct information controls, circulation may move into infrastructures not designed for news at all. During the early months of the 2022 Russian invasion, war-related posts surged in Tripadvisor forums and Google Maps reviews as users sought to bypass Russian censorship, share anti-war appeals, exchange humanitarian and travel advice, and disseminate polarized or false messages (Moreno et al., 2023). The result was not a stable substitute for journalism but a hybrid communication ecology combining information, advocacy, rumor, and moderation conflict.
Telegram occupies a related position in the post-Soviet sphere. A study of 95 public Telegram channels covering politics and war news assembled a corpus of over 2.3 million posts and an annotated sentence-level dataset for human-rights-violation detection, arguing that Telegram had become the leading social media platform for reading independent news in post-Soviet regions and an indispensable lifeline where freedom of expression is constrained and information warfare is pervasive (Nemkova et al., 2023). In parallel, research on event-themed malicious websites shows that attackers tailor war-themed campaigns around fundraising, aid, essential supplies, and “seeking updated news,” which means wartime demand for updates becomes an exploitable trust channel rather than a purely informational one (Mia et al., 29 Sep 2025).
4. Narrative warfare, propaganda, and multilingual framing
Comparative media analysis of the Russo-Ukrainian war shows that Western, Russian, and Chinese ecosystems turned the same conflict into different informational objects. Western coverage emphasized military and humanitarian dimensions; Russian coverage emphasized justifications for the “special military operation,” including “bio-weapons” and “neo-nazis”; and Chinese coverage emphasized diplomatic and economic consequences (Hanley et al., 2022). The paper detects Russian disinformation narratives in Chinese outlets and reports that Chinese state media increasingly cited Russian outlets and spread Russian narratives after the invasion (Hanley et al., 2022).
DNIPRO generalizes this insight by introducing a longitudinal corpus of 246,229 articles across eleven outlets from Russia, Ukraine, the U.S., the U.K., and China, in English, Russian, and Mandarin Chinese, covering February 2022 through August 2024 (Mohanty et al., 22 Jan 2026). Its exploratory analyses find that outlets construct “competing realities” and that media ecosystems build “parallel realities through attribution choices, selective emphasis, and strategic framing, rarely engaging with competing narratives.” Across the corpus, Security is the dominant topical frame and Morality is extremely rare, which is a striking finding for a war corpus because it indicates that explicit moralization is often displaced by security, political, legal, and policy vocabularies (Mohanty et al., 22 Jan 2026).
At the level of conflict-journalism theory, a computational study of about 22,000 articles on the Israel–Palestine war operationalizes Galtung’s distinction between war journalism and peace journalism through LLM-based extraction and frame-semantic analysis (Kaur et al., 12 Jun 2025). It reports a higher focus on war based reporting than peace based reporting, substantial regional differences in who is framed as assailant and victim, and a heavy emphasis on visible effects of war rather than invisible effects such as trauma, healing, and social disruption. The same paper shows that all regions are elite-heavy, though Middle Eastern coverage appears somewhat more attentive to named civilians and social suffering (Kaur et al., 12 Jun 2025).
Annotation studies reinforce how contested such categories are. The FIGNEWS shared task on multilingual news media narratives collected 15,000 Facebook posts in English, French, Arabic, Hebrew, and Hindi about the early phase of the Israel War on Gaza and produced 129,800 data points across bias and propaganda annotation subtasks (Zaghouani et al., 2024). The task found that within-team agreement exceeded across-team agreement by large margins, which implies that even structured annotation frameworks do not collapse wartime framing into a single stable truth condition. Bias and propaganda appear as negotiated judgments rather than straightforward extraction targets (Zaghouani et al., 2024).
5. Victimhood, sourcing, and epistemic authority
One major dimension of the war on news concerns not whether suffering is covered, but how it is made visible and credible. A study of 14,280 articles from The New York Times, BBC, CNN, and Al Jazeera English during the first year of the Israel–Gaza war identifies three systematic biases in Western outlets: Israeli victims were more likely to be portrayed as identifiable individuals, Western coverage created a false balance between Israeli and Palestinian suffering despite radical asymmetry, and reports about Palestinian casualty numbers more often used language that cast doubt on the credibility of the source or the numbers themselves (AlShebli et al., 7 Oct 2025). The paper reports that Palestinian mentions had 13.6% lower odds of being individualized than Israeli mentions in the BBC, 27.0% lower odds in CNN, and 38.9% lower odds in the New York Times, while Al Jazeera English showed no statistically significant difference (AlShebli et al., 7 Oct 2025).
The same study argues that these patterns operate through humanization, salience, and credibility. Israeli victims were more often narrated as named persons with biographies and futures; Palestinians were more often represented as collectives or numbers. Western outlets also repeatedly returned to October 7 narratives, with 92% of Israeli casualty-related stories tied to October 7, thereby maintaining Israeli victim salience even during later periods dominated by new Palestinian mass-casualty events (AlShebli et al., 7 Oct 2025). Source-doubting formulations such as “Hamas-run health ministry” were overwhelmingly attached to Palestinian civilian victim numbers, whereas equivalent source-doubting language was not found for Israeli casualty numbers (AlShebli et al., 7 Oct 2025).
A related but distinct line of work studies “epistemic strategies” in partisan reporting. Comparing CNN and Fox News on matched events across the COVID-19 period and the Israel–Hamas war, a computational analysis of 476,000 articles and 16.5 million sentences finds that CNN’s reporting contains more factual statements and is more likely to ground them in external sources, whereas Fox News favors News Reports and direct quotations (Mor-Lan et al., 12 Oct 2025). CNN is described as constructing an appeal to formal authority through Experts and Expert Documents; Fox News constructs a more media-centric and quotation-heavy authority structure. This reframes bias as more than topic selection or lexical slant. It is also a contest over what counts as a fact, what kinds of sources can authorize it, and how war is made epistemically legible (Mor-Lan et al., 12 Oct 2025).
6. Computational instruments, multimodal pipelines, and analytical limits
The recent literature treats the war on news as measurable through increasingly specialized computational instruments. NewsLens, for example, recasts media-bias analysis as adversarial navigation rather than left/right classification by using a five-agent pipeline consisting of a Fact Verifier, Progressive Framing Analyst, Conservative Framing Analyst, Propaganda Detector, and Neutral Summarizer (Bose, 17 May 2026). Its central metric is the Perspective Divergence Score,
where and are token sets extracted from progressive and conservative ideological framing fields (Bose, 17 May 2026). In the reported sample of 15 articles across Kashmir, Gaza, Climate Policy, and Ukraine clusters, PDS values were generally high, and the framework was especially consistent on strongly propagandistic content such as the Republic World Kashmir case (Bose, 17 May 2026).
Other work addresses the fact that new crises produce new misinformation domains before labels exist. DAUD, an LLM-based domain-aware framework for cross-domain fake news detection on unseen domains, explicitly motivates the problem with cases such as the COVID-19 outbreak and the Russia–Ukraine war (Yang et al., 2 Feb 2026). The framework combines LLM-based semantic enrichment of news, domain-aware user modeling, and relation-aware alignment between original and LLM-derived features. In unseen-domain experiments, it outperformed prior baselines across Politics, Entertainment, and COVID-19 target domains, with the largest gains appearing where domain shift was strongest (Yang et al., 2 Feb 2026). The implication is that war-related fake-news detection cannot rely only on static in-domain classifiers.
Short-form video introduces a further layer of compression. A multimodal study of 2,371 YouTube Shorts and 94,042 visual frames from Al Jazeera, BBC, Deutsche Welle, and TRT World on the Israel–Hamas war shows that war on short-form platforms is represented less through continuous combat footage than through studio/interview mediation, humanitarian destruction, protest, and diplomatic spectacle (Miehling et al., 1 Apr 2026). The paper reports that smaller domain-adapted models outperformed larger transformers and even fine-tuned LLMs for aspect-based sentiment analysis, which underscores a recurrent methodological point in this literature: scale alone does not guarantee better conflict analysis (Miehling et al., 1 Apr 2026).
These infrastructures are analytically powerful but not epistemically self-sufficient. The Facebook study infers visibility from reactions rather than directly observing reach (Talaga et al., 25 Jul 2025). FIGNEWS shows that cross-team agreement on bias and propaganda remains substantially lower than within-team agreement (Zaghouani et al., 2024). NewsLens reports no statistically significant between-group differences at and treats its findings as feasibility rather than population-level proof (Bose, 17 May 2026). Telegram-based HRV detection reaches 0, but it detects mentions of possible violations, not verified atrocities (Nemkova et al., 2023). Taken together, these results indicate that the war on news is measurable, but only through methods that remain sensitive to proxy choice, annotation subjectivity, platform-specific affordances, and the continuing need for human review.