From narrative competition to model confrontation:

Generative AI: From tool adoption to infrastructural reordering

1.1

Over the past decade, international politics has been shaped by successive waves of digital transformation, from social media platforms to algorithmic governance. The recent diffusion of generative artificial intelligence represents a qualitatively different moment within this trajectory. Models such as ChatGPT, DALL-E 3, Midjourney, and Sora are no longer experimental novelties; they are increasingly embedded in the routine production, circulation, and verification of information across global communication systems. According to Stanford University’s Artificial Intelligence Index Report 2024, global investment in generative AI reached $25.2 billion in 2023, representing a nearly ninefold increase compared to the previous year; OpenAI’s ChatGPT surpassed 100 million users within two months of its release, establishing itself as the fastest-growing consumer application in history.¹

What distinguishes generative AI from earlier digital technologies is not merely its technical performance, but its infrastructural role. Rather than functioning as a discrete instrument that assists human actors, generative AI reorganizes how communicative content is produced, scaled, translated, and recombined. As Crawford (2021) observes, contemporary AI systems operate as extractive infrastructures, embedding asymmetries of power into the material foundations of knowledge production and social coordination. In international politics, these infrastructures increasingly mediate how events are narrated, interpreted, and rendered credible.

Three interrelated transformations are particularly salient for strategic communication in conflict settings. First, generative AI dramatically lowers the skill and labour thresholds associated with content creation. Activities that previously required specialized professional expertise – such as multilingual text production, visual design, and video editing – can now be performed through natural-language prompting and automated pipelines. Empirical evidence suggests that such systems not only accelerate production but also compress costs, enabling a scale of output that would be unattainable through human labour alone.²

Second, generative AI destabilizes long-standing assumptions about authenticity and evidentiary authority. Audio-visual materials have historically derived credibility from their indexical relationship to reality; synthetic media technologies weaken this relationship by making realistic fabrication routine rather than exceptional. Survey data indicate that public confidence in distinguishing authentic from Ai-generated content has declined sharply in recent years,³ suggesting that verification practices struggle to keep pace with automated generation.

Third, the deployment of generative AI reconfigures power relations through technological concentration and dependency. Advanced model development depends on access to large-scale computational infrastructure, proprietary datasets, and platform ecosystems that remain concentrated among a small number of corporate and state actors. Control over these resources enables forms of systemic influence that extend beyond individual messages, aligning with what Farrell and Newman (2019) describe as networked power embedded in infrastructural chokepoints.

Taken together, these developments signal not simply a technological upgrade within existing communication practices, but a reordering of the conditions under which strategic communication is conducted in international conflict.

Strategic communication in conflict: from supporting function to contested domain

1.2

Strategic communication has long been intertwined with the conduct of international conflict. Defined broadly as the coordination of actions, messages, and symbolic representations in pursuit of political objectives,⁴ it has historically operated alongside military and economic instruments of power. From wartime propaganda in the early twentieth century to Cold War ideological confrontation and post-9/11 public diplomacy, communicative power has consistently stood alongside military force and economic strength as a pillar of national power.⁵

In the twenty-first century, this communicative dimension has moved from the periphery of conflict management toward its centre. Military doctrines increasingly recognize information environments as operational domains rather than auxiliary arenas. NATO explicitly positioned strategic communication as “an operational capability equivalent to military operations” in its 2009 Strategic Communications Policy.⁶ In the concept of “Hybrid Warfare” proposed by Russian military theorist Gerasimov in 2013, “information psychological” operations were placed at the forefront, with their effects considered potentially four times greater than military means.⁷ These reflect a growing consensus that conflicts are fought not only over territory, but over perception, legitimacy, and cognition.

Yet the rise of generative AI marks a turning point within this longer trajectory. If earlier digital platforms functioned primarily as amplifiers – accelerating dissemination while relying on human-crafted content -, generative AI introduces a qualitatively different dynamic. Content can now be produced automatically, in large volumes, and in forms that adapt continuously to audience feedback. As a result, strategic communication increasingly involves competition over the systems that generate and circulate representations of reality, rather than solely over the narratives those representations convey.

This shift complicates existing theoretical frameworks that treat communication as a matter of message design and persuasion. When content generation itself becomes automated and infrastructural, the locus of strategic advantage moves upstream, toward control over models, data, and distribution architectures.

Symptoms of Strain: Limits of the Narrative-Centred Framework

1.3

The diffusion of generative AI exposes growing tensions within traditional approaches to strategic communication. This technological transformation challenges the core assumptions of traditional strategic communication across multiple dimensions, revealing distinct signs of what Kuhn (1962) terms a “paradigmatic crisis” – a growing accumulation of anomalies that existing theories cannot explain. Existing frameworks, particularly those centred on narrative competition, presuppose relatively stable distinctions between truth and falsehood, identifiable communicative actors, and audiences capable of evaluating competing frames. In AI-mediated environments, these assumptions are increasingly difficult to sustain.

First, verification mechanisms face structural limits. Generative systems frequently produce content that recombines factual elements within misleading or speculative contexts, blurring binary distinctions between authentic and false information. The 2023 fake image incident depicting a Pentagon “explosion” demonstrates this challenge: a realistic AI-generated image triggered significant financial market volatility before being debunked, revealing that verification speed cannot keep pace with generation and dissemination speed.⁸

Second, the boundaries of agency become less clear. Communication processes increasingly involve algorithmic systems that autonomously generate content, recommend dissemination pathways, or simulate human personas. These systems complicate conventional distinctions between sender, medium, and audience, challenging analytical models that locate intent and responsibility exclusively in human actors. While Woolley & Howard’s (2016) concept of “computational propaganda” initially captured this trend, it still viewed bots as “manipulated tools.” Generative AI further blurs boundaries between “manipulator,” “tool”, and “audience”.

Third, generative AI dissolves the long-standing trade-offbetween scale and personalization. Whereas mass communication historically privileged reach at the expense of specificity, AI-driven systems can generate tailored variants of messages for different audiences while maintaining strategic coherence. This capability extends earlier forms of micro-targeting into a continuous and automated process. The Cambridge Analytica scandal preliminarily revealed the power of micro-targeting;⁹ generative AI pushes this capability to its extreme.

Finally, speed and volume increasingly outweigh coherence and quality. In algorithmically curated environments, persistent exposure and informational saturation can shape perceptions more effectively than carefully constructed arguments. In algorithm-dominated information environments, however, speed, volume, and persistent exposure often shape perception more effectively than quality – this is the basis of the strategies of “information flooding” and “trash talk”.¹⁰ Generative AI lowers the cost of such saturation strategies, altering the cost-benefit calculus of strategic communication.

These developments do not render narrative irrelevant, but they strain frameworks that treat persuasion and framing as the primary mechanisms of communicative power.

Research Questions and Argument

1.4

Against this backdrop, this article addresses two core questions:

RQ1: How is generative AI reshaping the mechanisms of strategic communication in international conflicts?

RQ2: What changes does this entail for the units of communication, modes of operation, and sources of communicative power?

The central argument advanced here is that strategic communication is undergoing a shift from narrative competition to what this article terms model confrontation. In earlier configurations, competition centred on the construction and circulation of coherent narratives that sought to define meaning, legitimacy, and identity. In AI-mediated environments, competition increasingly extends to the technological systems that generate, distribute, and authenticate information at scale.

This shift reconfigures communicative power along several dimensions. Symbolic and discursive resources remain relevant, but they are increasingly supplemented by technological capabilities such as computational power, data access, and control over algorithmic infrastructures. Operationally, strategies focused on persuasion coexist with practices aimed at shaping the broader information environment through saturation, fragmentation, and cognitive overload.

Importantly, this transformation does not entail the replacement of one paradigm by another. Narrative competition and model confrontation coexist in an asymmetric and dynamic relationship, with older practices persisting even as new logics permeate and restructure them. Understanding this layered configuration is essential for analysing contemporary conflicts in which communicative power is exercised not only through meaning-making, but through infrastructural control over how meaning is produced and encountered.

Narrative competition and model confrontation

2.1

This study is grounded in two analytical constructs – narrative competition and model confrontation – which are not treated as mutually exclusive categories, but as ideal-typical logics that foreground different sources of communicative power.

Narrative competition refers to communicative struggles in which actors seek to establish dominant interpretations of events by constructing coherent frameworks of meaning. Such frameworks typically organize causal attribution, moral evaluation, and identity positioning, enabling audiences to make sense of complex political realities.¹¹ Within this logic, strategic advantage derives from discursive authority, media access, and the capacity to mobilize symbolic resources. This concept draws theoretical insights from soft power theory,¹² which emphasizes attraction and identity shaping; framing theory,¹³ which reveals the mechanisms of issue selection and meaning construction; and public diplomacy studies,¹⁴ which explores the role of strategic narratives in shaping international public opinion.

Model confrontation, by contrast, designates a mode of competition in which strategic advantage increasingly hinges on control over the technological systems that generate, distribute, and authenticate information. Its theoretical roots can be traced to algorithmic governance studies,¹⁵ infrastructure studies,¹⁶ computational propaganda theory,¹⁷ cognitive security framework.¹⁸ Here, communicative power is exercised not only through persuasive content, but through algorithmic infrastructures that shape exposure, visibility, and credibility at scale. While this study introduces model confrontation as a novel conceptual lens, it builds on existing scholarship in algorithmic governance, infrastructure studies, and computational propaganda.

Framing these concepts as competing logics rather than discrete stages allows for a more nuanced analysis of coexistence and overlap. In contemporary conflicts, narrative-based persuasion and model-driven environment shaping often operate simultaneously, albeit with shifting relative importance.

Evolution of strategic communication

2.2

Research on strategic communication has developed at the intersection of international relations, communication studies, and critical political theory. Within international relations, early attention focused primarily on material power, with communication gaining prominence through the articulation of soft power and public diplomacy. Nye’s (1990) formulation emphasized attraction and persuasion as alternatives to coercion, prompting extensive research on image management, cultural diplomacy, and international broadcasting. Afterwards, scholars began to study how countries shape their international image through cultural exchanges, educational programs and media communication.¹⁹

However, soft power theory’s limitations have become apparent. Critics argue it overly relies on the assumption of “attraction”, neglecting the reality of coercive, manipulative communication;²⁰ it is also difficult to quantify and verify causally, with fuzzy conceptualization.²¹ In the early 21^st century, scholars began focusing on “sharp power”: non-attractive influence exercised by authoritarian states through manipulation of information environments,²² and disinformation campaigns in the context of sharp power.²³ The emergence of concepts such as sharp power reflects a growing recognition that communicative influence often operates through distortion, disruption, and exploitation of open information environments rather than mutual attraction.

Communication studies have contributed a different set of analytical tools, particularly through agenda-setting²⁴ and framing theories.²⁵ These approaches clarify how issue salience and interpretive structures shape audience perceptions, and have become essential tools for analysing strategic communication.²⁶ Applied to international conflict, they have generated valuable insights into how governments and non-state actors attempt to legitimize violence, assign blame, and mobilize support. Entman (2008) analysed how the U.S. government-controlled framing of post-9/11 wars using a cascading activation model; Archetti (2013) examined strategic narrative construction by terrorist organizations. Yet these models largely presuppose identifiable human communicators and relatively stable media institutions.

Critical and discursive traditions extend this critique by foregrounding the constitutive role of discourse in producing social reality. From this perspective, power operates not merely through persuasion, but through the normalization of particular ways of seeing and speaking. This perspective has spawned critical analyses of “media imperialism”²⁷, “news globalization”²⁸, and “discursive hegemony”.²⁹ However, much of this work remains centred on symbolic representation, devoting less attention to the material infrastructures through which contemporary communication is organized.

Taken together, these traditions provide essential foundations for understanding strategic communication. At the same time, their shared emphasis on narratives, discourse, and framing leaves key questions unresolved in environments increasingly shaped by automation and algorithmic mediation.

Generative AI, Misinformation, and Algorithmic Power

2.3

Since the widespread adoption of large language models in late 2022, scholarly attention to generative AI has expanded rapidly. Much of this research has focused on labour displacement,³⁰ creativity,³¹ and ethical risk.³² While these studies are essential for understanding the broader societal implications of AI, they rarely address the specific dynamics of international conflict.

A growing body of work examines the relationship between generative AI and misinformation. Experimental studies demonstrate that AI-generated political content can match human-written texts in perceived credibility while being produced at a fraction of the cost and time.³³ From a security perspective,³⁴ they warn that such capabilities lower barriers to large-scale manipulation, enabling what has been described as “industrialized disinformation”.

However, existing research often treats generative AI as an isolated tool rather than as an integrated system. Most studies focus on individual models, content types, or platforms, paying limited attention to how production, distribution, amplification, and verification interact within a single technological stack. As a result, the transformation of strategic communication tends to be described in additive terms – AI as an accelerant – rather than as a reconfiguration of underlying mechanisms.

This study departs from that tendency by foregrounding generative AI as infrastructural power. Rather than asking whether Ai-generated content is persuasive, it examines how algorithmic systems reshape the conditions under which persuasion, confusion, and legitimacy operate.

Addressed theoretical gaps

2.4

Despite rapid growth in research on AI and political communication, three interrelated gaps remain. First, international conflict is often treated as one application context among many, rather than as a distinctive environment characterized by high stakes, strategic rationality, and asymmetric power. Analytical frameworks developed for commercial platforms or domestic politics may not travel easily to wartime conditions.

Second, existing studies tend to emphasize observable effects – misinformation, polarization, erosion of trust – while offering limited accounts of how communicative mechanisms themselves are reorganized. Questions concerning shifts in core units of communication, sources of power, and rules of competition remain under-theorized.

Third, disciplinary fragmentation continues to constrain cumulative knowledge. International relations scholarship frequently abstracts away from technical detail, while studies of AI governance and algorithmic systems often lack sensitivity to geopolitical dynamics. Communication research, positioned between these domains, is uniquely suited to bridge this divide, but has yet to fully exploit this potential.

This study positions itself at this intersection. By conceptualizing the shift from narrative competition to model confrontation, it offers an analytical framework capable of capturing both continuity and transformation. In doing so, it seeks to contribute not only to debates on AI and misinformation, but to broader discussions about power, communication, and conflict in the digital age.

Case selection and analytical approach

3.1

A theory-driven comparison between the 2014 Crimea crisis and the 2022 Russia-Ukraine conflict is developed to examine how strategic communication mechanisms evolve under different technological conditions. The comparison is not intended to establish linear causality or universal generalization. Instead, itserves as a structured contrast that highlights how shifts in communicative infrastructure interact with strategic objectives to produce distinct modes of influence.

The two cases are analytically valuable for three reasons. First, they involve largely overlapping state actors – most notably Russia and Ukraine – allowing changes in communication practices to be examined without attributing differences solely to national culture or political identity. Second, both conflicts unfolded within digitally mediated information environments, yet under markedly different technological configurations. While social media platforms were already influential in 2014, the large-scale deployment of generative AI systems had not yet occurred. By contrast, the post-2022 conflict coincides with the widespread availability of large language models, automated content generation tools, and algorithmically optimized distribution systems. Third, both cases were accompanied by extensive documentation from journalists, researchers, and international organizations, enabling triangulation across multiple sources (Table 3-1).

The analysis proceeds by tracing how strategic communication operated across five interrelated dimensions: (1) the basic unit of communication, (2) modes of content production, (3) dissemination logics, (4) key strategic resources, and (5) dominant operational objectives. Rather than treating these dimensions as static variables, the chapter emphasizes their interaction and co-evolution, showing how changes in one dimension reshape the others.

Importantly, the comparison does not treat technological change as a linear background variable. While digital platforms were already influential in 2014, the analytical focus here is on the emergence of automated generative systems that reorganize production and distribution at scale. The shift identified in this chapter is therefore not attributed to digitization per se, but to the introduction of model-based infrastructures capable of continuous, low-cost variation and algorithmic amplification, which were capabilities largely absent during the Crimea crisis.

Dimension one: units of communication

3.2

From Coherent Narratives to Modular Information Fragments

2014 Crimea: During the Crimea crisis, the dominant unit of strategic communication was the coherent narrative. Russian state actors and affiliated media outlets advanced storylines that linked historical memory, moral evaluation, and political justification into relatively stable interpretive frameworks (Table 3-2). These narratives functioned as sense-making devices, allowing audiences to situate unfolding events within familiar symbolic repertoires.³⁵

Crucially, narratives in this period retained internal consistency and temporal continuity. They unfolded over weeks and months, enabling repetition and reinforcement without substantial variation. Competing interpretations were identifiable and contestable, reinforcing the centrality of narrative competition as the primary mechanism of influence.

2022 Russia-Ukraine conflict: By contrast, the post-2022 conflict is characterized by the fragmentation of communicative units. Short videos, memes, screenshots, translated excerpts, and decontextualized images circulate independently of overarching storylines. Empirical studies of TikTok demonstrate that highly visible content often lacks narrative completeness, relying instead on affective intensity and immediacy.³⁶

Generative AI accelerates this fragmentation by enabling rapid recombination and stylistic variation. The same informational claim can appear simultaneously as text, image, video, or meme across platforms and languages, undermining the stabilizing function traditionally performed by narrative coherence.³⁷

Furthermore, the 2022 Russia-Ukraine conflict has revealed a distinct dual content ecosystem, a phenomenon that validates Chadwick’s (2017) theory of the “hybrid media system”: the logic of old media (such as grand narrative and framing) coexists with the logic of new media (such as viral spread, memetic expression, and algorithm-driven dynamics) in a relationship marked by significant tension. Specifically, the upper layer of this content ecosystem is centred on coherent grand narratives, primarily crafted manually through traditional media and official statements, aimed at constructing strategic frameworks and establishing legitimacy – embodying the classic logic of old media. In contrast, the lower layer is saturated with fragmented micro-content, predominantly disseminated across social media and instant messaging platforms and produced on a large scale through automated or semi-automated means (including AI generation). Its functions focus on emotional mobilization, pollution of the information environment, and the creation of false consensus, which are characteristic of new media logic. The interplay and contention between these two logics within the same communicative space constitute the complex informational dynamics of this conflict.

Comparative Implication: The shift from narratives to fragments alters how meaning is produced. Under narrative competition, influence depends on interpretive dominance; under model confrontation, it depends on exposure patterns shaped by algorithmic systems. Meaning emerges from accumulation rather than coherence. Comparative Implication: The shift from narratives to fragments alters how meaning is produced. Under narrative competition, influence depends on interpretive dominance; under model confrontation, it depends on exposure patterns shaped by algorithmic systems. Meaning emerges from accumulation rather than coherence.

Dimension two: modes of content production

3.3

From Human-Centred Creation to Automated Pipelines

2014 Crimea: Content production during the Crimea crisis remained largely human-centred and institutionally organized. State media relied on professional journalists, editors, and producers working within established editorial hierarchies. Just as Yablokov’s (2015) research shows, RT significantly increased its investment during the Crimea crisis, producing dozens of video reports, articles and social media content every day. However, all of these contents were created by human teams and followed the traditional news production process. While output intensified during the crisis, it remained constrained by labour, time, and organizational routines.³⁸

In addition, social media was used during 2014 Crimea, but mainly as a channel for the dissemination of traditional content rather than an independent content production platform. According to the research of Freelon & Lokot (2020), RT’s YouTube channel grew rapidly during the crisis, but the content was still reprints of TV programs. Twitter and Facebook are used to share news links and quotes, but long-form and in-depth content is still found in traditional media. Discussions on VKontakte (the Russian version of Facebook) are active, but organized dissemination still relies on media organizations.

Although early forms of digital manipulation and trolling existed, they did not fundamentally alter production logics. Human judgment remained central to content creation, framing decisions, and narrative maintenance.

2022 Russia-Ukraine conflict: In the later conflict, content production increasingly involves automated or semi-automated systems. Large language models can facilitate text generation and translation, while image and video synthesis tools can achieve rapid visual production and have greater dissemination power. Epstein et al.’s research (2023) shows that political propaganda images generated by Midjourney often surpass human designs in terms of visual appeal, because AI has learned a large number of successful design patterns. The automated application of generative AI has also led to an exponential increase in the production speed of dissemination content. Human actors remain involved, particularly in strategic planning and prompt design, but the marginal cost of producing additional content has declined sharply.³⁹ Graphika (2023) compared the out put of suspected state-supported information manipulation networks in 2014 and 2022 (Table 3-3).

Research on hybrid human-AI production systems suggests that even partial automation significantly increases output volume and speed, reshaping strategic expectations about scale and persistence.⁴⁰

Comparative Implication: This shift transforms production from a bottleneck into a strategic advantage. Under model confrontation, communicative power derives less from editorial capacity than from the ability to sustain automated output over time.

Dimension three: dissemination and amplification

3.4

From Broadcast Logic to Algorithmic Curation

2014 Crimea: Dissemination during the Crimea crisis largely followed a broadcast logic: diffusion from the centre to the periphery. Content originated from centralized producers and circulated outward through television and online platforms. Social media amplified messages, but did not fundamentally shape their selection or visibility.⁴¹

Agenda-setting effects were observable and traceable, according to the comparative analysis of BBC, CNN and RT by Hutchings & Szostek (2015), RT successfully forced Western media to respond to its framework (even if it was rebuttal), thereby partially setting the agenda. Strategic communication operated through repetition and framing rather than through personalization.

2022 Russia-Ukraine conflict: In contrast, dissemination in the post-2022 conflict is governed by algorithmic curation. Recommendation systems on platforms such as YouTube personalize information streams based on user behaviour, producing fragmented and opaque exposure patterns.⁴²

Algorithmic amplification privileges engagement over coherence, favouring emotionally charged and rapidly circulating content regardless of narrative alignment.⁴³ Moreover, Hern’s (2022) investigation report further reveals that the war-related content users are exposed to is highly dependent on their initial interaction choices, thus forming two almost parallel information universes of Pro-Russian and pro-Ukrainian. During this process, algorithms, oriented towards optimizing engagement, tend to recommend more emotional and extreme content, thereby intensifying the information cocoon and cognitive differentiation.

Comparative Implication: Under algorithmic curation, communicative power shifts from message salience to system compatibility. Visibility depends on alignment with platform logics rather than narrative persuasiveness.

Dimension four: strategic resources and power bases

3.5

From Discursive Authority to Infrastructural Control

2014 Crimea: In 2014, strategic resources primarily consisted of discursive authority, media reach, and symbolic capital. For instance, the Russian traditional media RT played a significant role in 2014, Yablokov (2015) attributed the success of RT to its imitation of the professionalism of Western media while infusing a Russian perspective, which required a large amount of human capital and financial investment. State actors leveraged institutional credibility and historical narratives to shape interpretation. Although infrastructural asymmetries existed, they did not constitute the primary locus of competition.

2022 Russia-Ukraine conflict: In the later conflict, access to computational resources, training data, model availability, and platform integration emerges as a critical source of power. Control over these infrastructures enables actors to shape information environments at scale.⁴⁴

This infrastructural dimension aligns with broader accounts of algorithmic power, which emphasize how control over technical systems structures social and political outcomes beyond individual acts of persuasion.⁴⁵

Comparative Implication: Strategic communication increasingly resembles infrastructural competition, where advantage derives from system-level positioning rather than discursive skill alone.

Dimension Five: Operational Objectives

3.6

From Persuasion to Environment Shaping

2014 Crimea: The dominant operational objective in 2014 was persuasion. Strategic communication sought to legitimize territorial change, mobilize support, and counter external criticism by stabilizing meaning. Although manipulation occurred, it remained oriented toward belief formation.

The core objective of Russia’s communication strategy in Crimea still follows the classic logic of persuasion. At the domestic level, the strategy aimed to persuade the Russian people to support the intervention operation, successfully pushing Putin’s approval rating from 65% to 86% (according to data from the Levada Center), reaching a record high. At the international level, efforts have been made to persuade the international community to accept the so-called “legitimacy of the referendum”. Although it has not been widely recognized, it has successfully sparked extensive debates in the international public opinion field. At the same time, efforts have been made to persuade the sceptical forces within the West (such as the far right in Europe and some anti-war left wings) to weaken the internal unity of the sanctions alliance.

2022 Russia-Ukraine conflict: In the post-2022 environment, many information operations prioritize disruption over persuasion. Analysts document strategies aimed at generating uncertainty, fatigue, and interpretive overload rather than consensus.⁴⁶

The RAND study by Helmus et al. (2023) points out that in 2022, Russia shifted its strategic focus more toward creating uncertainty, sowing doubt, and deploying contradictory narratives and false evidence to induce cognitive dissonance. This change in strategy is succinctly captured by Pomerantsev’s (2019) observation: “The goal is not to persuade you of his narrative, but to make you doubt all narratives.” This argument is further supported by Bateman’s (2023) public opinion tracking, which shows that while 85-90% of Western audiences believe Ukraine’s version of events, the proportion who are “completely certain” has fallen from an initial 75% to 60%. In Global South countries, 40-50% of respondents report being “uncertain or needing more evidence.” This suggests that although the strategy of confusion has not succeeded in persuading audiences, it has effectively generated widespread doubt. As a result, credible information is being drowned in a sea of low-quality content, forcing audiences to invest significantly more time and effort to locate reliable information.

Generative AI lowers the cost of producing ambiguity at scale, enabling strategies that shape the conditions of interpretation rather than its content.⁴⁷

Comparative Implication: Under model confrontation, success is measured less by belief alignment than by the ability to destabilize shared reference points.

Comparison: the shift in mechanisms

3.7

A comparison of the two cases reveals a clear transformation in the mechanisms of strategic communication. In the Crimea crisis, narratives structured the information environment, and strategic competition centred on meaning stabilization and legitimation. In the post-2022 conflict, narratives operate within environments increasingly shaped by automated systems that privilege scale, speed, and fragmentation.

Across the five analytical dimensions, the shift is consistent. Communicative units move from coherent stories to modular fragments. Production shifts from labour-intensive human processes to automated pipelines. Dissemination shifts from broadcast logics to algorithmic curation. Strategic resources shift from discursive authority to infrastructural control. Operational objectives shift from persuasion toward environment shaping.

Crucially, model confrontation cannot be reduced to an intensified form of narrative competition. While narratives operate at the level of meaning and interpretation, model confrontation targets the upstream conditions of communicative possibility by shaping how content is generated, circulated, and rendered visible. The two logics therefore operate on analytically distinct planes: one competes over interpretation, the other over infrastructural control. Narrative competition and model confrontation coexist in an asymmetric relationship, with older practices persisting even as new technological conditions restructure their effectiveness. Recognizing this layered configuration is essential for understanding how strategic communication operates in contemporary international conflict.

Limitations and future research

This study has several limitations. The comparative analysis focuses on two cases involving a specific set of geopolitical actors, which constrains the generalizability of its findings. Future research could extend the framework to conflicts involving different media systems, levels of technological capacity, or forms of political organization.

In addition, the rapid evolution of generative AI technologies presents a moving target for empirical analysis. As models, platforms, and regulatory regimes continue to change, the forms of model confrontation identified here may themselves evolve. Longitudinal and mixed-methods approaches will be particularly valuable for capturing these dynamics.

Finally, further research is needed to examine how audiences experience and navigate AI-mediated information environments in conflict settings. Understanding how individuals interpret, resist, or adapt to automated saturation remains crucial for assessing the broader societal consequences of generative infrastructures.

Conclusion

4.5

This research begins with an observation: generative artificial intelligence is profoundly transforming strategic communication in international conflicts. Through systematic theoretical construction and empirical analysis, we have confirmed the paradigm shift from “narrative competition” to “model confrontation”, revealing the technical foundation, operational mechanism, and profound implications of this transformation.

This discovery not only satisfies academic curiosity, but also holds urgent practical significance. We are standing at a historical turning point: technology is redefining the boundaries of power, truth, and human cognition. If the major challenge of the 20^th century was how to deal with nuclear weapons, totalitarianism, and ecological crises, one of the core challenges of the 21^st century will be how to protect human cognitive autonomy, democratic space, and common reality in an era enabled by technology.

The paradigm shift from narrative competition to model confrontation is still ongoing and far from complete. New technologies (such as multimodal AI, embodied AI, and general artificial intelligence) will continue to emerge. New tactics will be invented. New governance challenges will emerge. Therefore, this study is not a conclusion but a starting point.