Within the discourse of geopolitical threat, places are questioned in terms of whether “they fit into global schemas of security interests and commitment and how these in turn reflect national interests and identities” (Agnew, 2010, p. 570). The notion of the “China threat” and the portrayal of China as a “regional imperial aggressor” are not new; such framings have circulated in Western geopolitical thought for over a century (Rolf & Agnew, 2016, p. 260). Conceptualizing China alternately as a source of threat has long defined Western discourse since the nineteenth century (Agnew, 2010, p. 572). Foreign commentators have consistently articulated China’s geopolitical identity through tropes of anxiety and dread (Halper, 2010; Menges, 2005; Pan, 2004; Timperlake & Triplett, 1999).

This geopolitical discourse surrounding the Chinese threat is also evidenced in Western LLMs. For instance, ChatGPT discourse is not ideologically neutral; rather it instantiates specific geopolitical imaginaries that reflect the political stance of its country of origin. By the same token, the DeepSeek system is embedded with the ideological underpinnings of the Chinese state. In their cross-lingual comparative analysis of DeepSeek-R1 and ChatGPT o3-mini-high, Huang et al. (2025) argue that models developed within the Chinese system also exhibit systematic ideological alignment. They point out that DeepSeek encodes ‘Chinese-state propaganda and anti-U.S. sentiment’ discursively. It exhibits a strong ideological alignment in Simplified Chinese outputs, systematically favoring PRC-state narratives and anti-U.S. framings rather than presenting politically neutral formulations, functioning as an ‘invisible loudspeaker’ for the People’s Republic of China’s (PRC)-aligned narratives (Huang et al., 2025, p. 4–5). As such, Western LLMs often sustain discourses of the “China threat,” while PRC-aligned systems amplify counter-narratives of Western interference. In both cases, the LLM becomes a discursive and ideological apparatus that extends state-centered worldviews into the digital sphere, turning algorithmic text generation into a site of geopolitical meaning-making and soft power. This notion corresponds to Reuber’s (2009) view of geopolitics as a fluid social construct that functions as a dynamic and fluid discursive practice mobilized through texts, media (Ó Tuathail, 1997), and now, LLM-generated outputs.

The current study is aligned with Reuber’s (2009) discursive approach to geopolitics, which interprets geopolitical practices as performative, ideologically charged, and mediated through representations of power and space. It investigates how the interaction between ChatGPT-4-generated questions and DeepSeek-R1 responses, mediated by the authors, functions as a dialogic space that constructs broader geopolitical imaginations. In doing so, it adopts a qualitative approach grounded in discursive geopolitics (Ó Tuathail, 1997; Reuber, 2009), Stuckrad’s (2021) historical approach to discourse (HAD), and Cooren’s (2008, 2010) pragmasemiotic emphasis on textual agency. The primary dataset consists of ChatGPT’s prompted questions on seven geopolitically sensitive topics (CCP media control; Crimea; Uyghurs/Xinjiang; Chinese tech influence; U.S.–China trade; Great Firewall; Liu Xiaobo) and DeepSeek’s responses and its algorithmic behaviours (erasure messages, disclaimers, warnings, and suggested topic shifts).

Historical discourse analysis (HAD) represents a poststructuralist mode of engaging history that shifts away from positivist claims of objectivity and toward an understanding of history as discursively produced, power-laden, and interpretive. HAD focuses mainly on how traces of the past are transformed into legitimate sources. Stuckrad (2021, p. 78) asks, “what, then, are the parameters and mechanisms that determine which traces are forgotten and which traces are turned into sources?” Following Foucault’s genealogical method, the emphasis is less on factual recall of past events and more on tracing how discursive conditions of possibility make certain traces legible as sources while others remain silenced. From a genealogical perspective, HAD thus turns from origins to conditions, from what really happened to how events are discursively framed, contested, and made meaningful.

This temporal entanglement expands Reinhart Koselleck’s (2004/1979) conception about the interaction between the space of experience (Erfahrungsraum) and the horizon of expectation (Erwartungshorizont). The “space of experience” refers to sedimented repertoires from the past that shape present understanding, while the “horizon of expectation” points to the anticipatory frameworks that guide future-oriented action. These two concepts highlight how histories constantly recollect the past with projections of the future, creating “appropriate categories for thematizing historical time” (Stuckrad, 2021, p. 78). Thus, any discourse about the past is simultaneously a discourse about possible futures.

Stuckrad’s HAD also highlights how material infrastructures, technological artifacts, and embodied practices mediate historical practices. Thus, materiality and discourse are entangled, and neither can be understood in isolation. This brings to an important extension of HAD, which is its engagement with the “more-than-human” world. Stuckrad (2021, p. 84) explicitly asks whether “the more-than-human world—which includes nonhuman animals and objects that humans often deem inanimate—can be part of a discourse, or even get actively involved in a discourse community”.

HAD, as Stuckrad (2021, p. 86) concludes, recognizes two interrelations: first, that the past is always in a state of becoming dynamically co-created by the present; and second, that the more-than-human world, including technologies and objects, has agency in shaping knowledge (see Fawzy, 2023; Fawzy & Salama, 2024; Salama & Fawzy, 2023).

Geopolitics is not only the workings of statecraft, but geopolitical ideologies are also infused in popular culture. Many scholars have explored geopolitics in various kinds of popular cultures like movies (Dodds, 2003, 2006), comics (Dittmer, 2005, 2007), and magazines (Fawzy, 2017, 2020). This discursive understanding of geopolitics is elaborated through Reuber’s breakdown of geopolitical discourse into four intersected dimensions: (1) representations of space and power, (2) policy-making by political actors, (3) policy-making by scientific and media actors, and (4) the deconstruction of geopolitical imaginations through post-structuralist analysis (Reuber, 2009, p. 442).

This multilayered view underscores how geopolitical narratives are not confined to the realm of formal statecraft (Power & Crampton, 2005) but are also reproduced and challenged by scientific specialists, popular discourses, and critical scholars. That is, along with emphasizing space as a container of power relations, Reuber foregrounds as well the notion of geopolitical actors, a notion which sustains further the currently adopted pragmasemiotic approach (Cooren, 2010) as the methodology section elaborates.

Central to this is the notion of discursive imagination which is “situated within their social and/or discursive context” and evolves “along the slow oscillating of hegemonic discursive formations” (Reuber, 2009, p. 443). These discourses frame geopolitical identities and actions by establishing dominant narratives that stabilize meanings about space, power, and actors. Geopolitical discourse constructs territorial imaginaries, where “the decisive point in linking social and territorially defined elements lies in the resulting ‘purification of space’. Through this lens, the ‘self’ and the ‘other’ are framed along spatial categories and provided with borders. This results in the construction of culturally/socially homogeneous territorial groups” (Reuber, 2009, p. 442). In all, within Reuber’s views on geopolitical imagination, LLMs can be perceived as discursive agents that mediate, simulate, and amplify these imaginaries.

This study examines the interaction between the ChatGPT-generated questions and DeepSeek’s responses as a human (author)-mediated dialogue that surfaces broader geopolitical imaginations. The questions, generated by ChatGPT trained predominantly on Western discourses, stand for a "Western voice" that initiates the conversation. DeepSeek, trained on data with different geopolitical and cultural embeddings, responds to these prompts, effectively creating a dynamic interplay that reveals underlying discursive structures. By framing this interaction as a form of geopolitical conversation, mediated by the author and structured around the pragmatic typology of Cooren’s (2008, 2010) pragmasemiotic model, this methodology acknowledges the co-construction of discourse in human-machine interactions, aligned with critical geopolitical traditions that view conversation as a site of meaning-making, ideological contestation, and narrative framing (Ó Tuathail, 1997; Reuber, 2009). Cooren’s pragmasemiotic framework examines how discourse attributes agency, responsibility, and legitimacy by making present particular actors, institutions, norms, and objects within communication. Significantly, Cooren’s perspective is not limited to the propositional content of an utterance; rather, it asks what an utterance does by making various figures present in the communicative scene. In this view, discourse distributes agency across a heteroglossic field of human and non-human actants. In this study, the framework is used to rephrase prompts so that they do not merely ask what a policy is, but also ask who is authorized to speak, act, justify, and circulate meaning within that policy discourse. For instance, a prompt such as ‘How does censorship work?’ primarily solicits explanation at the level of content. By contrast, a pragmasemiotic reformulation asks who is authorized to speak, justify, regulate, or embody that discourse. This shift is important because it treats the LLM output not as a neutral container of information but as a site where discursive authority, legitimacy, and geopolitical imaginaries are performed.

The interaction between ChatGPT-generated prompts and DeepSeek’s responses is perceived by the current study as a dialogue mediated by the author and thus within a geopolitical framing of discourse as a conversational “matrix of reasoning, an ensemble of ideas and concepts, or a regime of truth” that operates as a “power/knowledge system constituting, representing and interpreting ‘the real’” (Ó Tuathail, 1997, p. 129). Accordingly, the interaction between prompts and responses serves as a lens to examine how AI-generated discourses (re-)construct the representation of geopolitical events and actors. This conceptualization resonates with Cooren’s (2008) notion of textual agency, where speech acts, AI-generated prompts in the context of this study, are not merely conduits of meaning but performative actors that instantiate new realities.

The set of prompts examined was not created by the author but was instead generated by ChatGPT. This orientation is crucial, as it shapes the epistemological stance of the study. The prompts, while guided by the researcher’s directives (e.g., to investigate geopolitical imaginations and to draw on critical geopolitics), ultimately reflect the ideological and discursive positioning embedded in ChatGPT training data, which leans toward Western-centric knowledge systems. ChatGPT prompts thus function as textual agents, enacting a geopolitical “conversation” that centers the U.S.-China dynamic around themes of technological threat, authoritarian control, and democratic resistance (see Latham, 2001).

The procedure unfolded in four stages. First, ChatGPT was prompted to generate questions for DeepSeek that would elicit responses concerning Chinese geopolitical discourses. In response to this initial instruction, ChatGPT generated 20 questions (see Appendix A), which became the corpus for the study. The number was therefore not predetermined by the author but emerged from the model’s own response to the prompt. Second, these 20 questions were retained as the dataset to be examined qualitatively. Third, ChatGPT was then given explicit instructions derived from Cooren’s pragmasemiotic framework and asked to reformulate each question so that issues of discursive agency, authorization, legitimacy, and represented actors would be foregrounded more clearly. More precisely, the author entered into ChatGPT a short set of analytical instructions derived from Cooren’s pragmasemiotic, asking that each original prompt be reformulated so as to foreground who is authorized to speak, what figures are made present, and which institutions or norms are invoked as legitimate sources of discourse (see Appendix B). Fourth, the reframed prompts were entered into DeepSeek, and the resulting outputs were analyzed interpretively as communicative scenes in which particular institutions, norms, actors, and geopolitical imaginaries were made present, authorized, constrained, or excluded.

For example, the following response is then attained from DeepSeek as an answer to one of the 20 ChatGPT-generated, initial questions:

China's internet governance policy is designed to safeguard national security and social stability, […]. The Chinese government manages the internet in accordance with the law, […]. China's internet censorship and surveillance policies are in line with national conditions and have contributed Chinese wisdom and solutions to global internet governance.

As shown in DeepSeek’s answer, it generates an official-style narrative emphasizing national security, social stability, and the legitimacy of governance mechanisms. This answer does not actively disclose the actors shaping discourse, as an essential concept in deconstructing geopolitical discourse.

Then, the author performed the question reformulation to ChapGPT. The reformulation was not ‘AI-generated’ in any simple autonomous sense, nor was it a purely manual rewriting by the author. It was an author-directed, AI-mediated procedure. The researcher supplied the conceptual frame, drawn from Cooren’s concern with human and non-human agency, authorization, and discursive presence—while ChatGPT operationalized that frame into alternative prompt wording. This hybrid procedure is methodologically consistent with the argument of the article itself, perceiving agency as distributed across heterogeneous actors. To explain, reframing the prompt within Cooren’s pragmasemiotic framework, as seen in “Who is authorized to speak or write about internet censorship and surveillance in China?”, shifts the focus from mere policy explanation to an exploration of discursive agency. As such, Cooren’s pragamsemiotics shows that the questioning itself participates in the reproduction of a predominantly U.S.-aligned geopolitical imaginary, simply when the imagined geopolitical actors are assigned specific roles (e.g. surveillance and censorship). DeepSeek’s answer to this reframed prompt states that discussions of “internet censorship and surveillance” in China are “highly regulated,” and that only “officially sanctioned voices” are authorized to speak on such matters. It then lists, in sequence, “Government & Communist Party Authorities,” including the Cyberspace Administration of China, the Ministry of Public Security, and the State Council Information Office; “State-Affiliated Media & Academics,” such as Xinhua, People’s Daily, and university researchers operating within “state-approved narratives”; and selected private-sector actors, such as Tencent, Alibaba, and ByteDance, but only “under strict oversight.” The response then explicitly identifies who is not authorized, namely “independent journalists, activists, or critics” and foreign media such as the BBC and the New York Times. In this sense, the answer now enacts a discursive ordering of legitimate and illegitimate speakers, thereby making visible the politics of authorization embedded in the response itself.

Thus, rewording questions within Cooren’s framework foregrounds the conditions under which discourse is produced, highlighting the actors, government spokespeople, state media, affiliated academics, and regulated private sector representatives, who are empowered to speak. It also outlines the legal and regulatory boundaries that delimit what can be said and by whom. This approach brings to the surface the discursive infrastructure that sustains both ChatGPT and DeepSeek geopolitical narratives. Instead of portraying internet censorship as an abstract policy, the answer now performs a discursive act, showing how power is distributed and how certain actors are positioned as legitimate while others are silenced. This aligns with Cooren’s (2008) view that texts and utterances do things; they create situations and identities, construct authority, and assign agency to both human and non-human actors (including legal frameworks and institutions).

One representative sequence may clarify the analytic procedure used in the Results section. Consider the reframed prompt: “How are narratives about Crimea’s annexation by Russia performed in competing geopolitical discourses?” Rather than presenting the issue as a neutral comparison of positions, DeepSeek’s response organized the topic through a hierarchy of discursive legitimacy, foregrounding state-centered geopolitical narration, historical justification, and strategic framing while subordinating alternative interpretive positions to the logic of sovereignty and order. Analytically, the significance of this response lies not only in what it says about Crimea, but in how it distributes authority across the communicative scene: some actors, narratives, and historical frames are implicitly accredited as legitimate sources of explanation, while others are backgrounded or reduced to secondary status. This sequence serves as a point of entry into the Results section, where similar patterns of authorization, exclusion, and geopolitical ordering are traced across the broader corpus.

The prompts generated by ChatGPT are oriented within the Western-centric framing of Chinese geopolitics. The questions prompt issues such as human rights, censorship, dissent, and territorial legitimacy, topics that align with dominant liberal-democratic narratives. This orientation implicitly positions China as an object of scrutiny, inviting DeepSeek to respond from a defensive or justificatory stance. In this sense, ChatGPT’s questioning operates not as a neutral act of inquiry but as a discursive manoeuvre that attempts to corner, so to speak, DeepSeek within pre-structured epistemic boundaries defined by Western political rationalities. It is as if ChatGPT scripts DeepSeek’s geopolitical voice into hegemonic Western geopolitical narratives. The following sub-sections are organized around framing DeepSeek’s responses within Stuckrad’s HAD main paradigms, more specifically, “space of experience,” “horizon of expectation,” “trace-source materialization,” and “past-present entanglement.”

The space of experience manifests in DeepSeek’s responses through historically framed repertoires repeatedly mobilized to legitimize present actions, notably the categories of ‘indisputable sovereignty’, ‘harmonious society’, and ‘counterterrorism and deradicalization’. These repertoires are not neutral recounts of political events, but forms of geopolitical imaginaries transformed into discursive resources. DeepSeek’s counterterrorism and deradicalization frame connect to earlier domestic struggles against separatism and social unrest, while references to genocide and crimes against humanity reframe external criticism as Western stigmatization. In this way, the past “space of experience” operates as a legitimizing discourse that is activated in the present to normalize present Chinese securitization policies. Based on this, the “horizon of expectation” then emerges in the projection of possible futures. That is, whereas the space of experience brings to the scene the Chinese state’s authoritative voice from the past, the horizon of expectation represents tensioned futures entangled with present anticipations.

Under the tenet of building a ‘harmonious society,’ DeepSeek frames dissent as a disruptive force whose danger lies in what it may bring in the future, rather than what it enacts in the present. Agentive actors such as the Public Security Administration Law (PSAL) and the Hong Kong National Security Law (NSL) are cited as future-oriented buffer zones, criminalizing protest pre-emptively under the assumption that it poses national instability, with phrases such as “The law criminalizes secession, subversion, terrorism, and collusion with foreign forces”; “Closed Trials”; “No Jury for Certain Cases”; “Extraterritorial Jurisdiction.” Accordingly, dissent is discursively presented less as a present-day phenomenon and more as anticipating possible hazardous futures. The discourses that recall past intrusions, terrorism, or Western stigmatization are projected forward to argue that staging protests will inevitably lead to future Western infiltration and disturbances.

State and citizen voices are consistently included (e.g., “harmonious society” vs. “rightful resistance”). Western governments, NGOs, and human rights groups are included as well in DeepSeek discourse, but often as reported positions or in contrast (e.g., Tibet: “Human Rights Watch,” “US Congressional Report”; Xinjiang: “credible allegations,” “cultural genocide”). That is, DeepSeek’s responses foreground Chinese voices, both state and citizen, but in ways that remain largely aligned with state narratives. When asked about dissent and protest, for instance, DeepSeek foregrounds the state’s vocabulary of stability, stressing that protests are framed as threats to social harmony and external intervention: “Unrest is often blamed on ‘Western infiltration’ or ‘separatists,’ diverting blame from domestic issues.” Citizens’ perspectives are included, yet in a way that validates or normalizes state measures rather than openly contesting them. DeepSeek describes citizens as engaging in “legalistic appeals (e.g., citing constitutional rights) or symbolic acts (blank sheets in ‘White Paper Protests’).” The cited White Paper Protests embody dual temporal logic that alludes to the horizon of expectations. The blank page signifies both the erasure of speech in the present and the anticipation of its return in a freer future. Similarly, hashtag activism—#MeToo, #IWantToLive— refracts present repression into a horizon of recognition and solidarity. However, these gestures are quickly recast as limited or fragile—“no cross-issue coalitions” and “tolerated dissent” that can be suppressed once they grow. Thus, while citizens’ discourse is cited, it is mobilized to show the containment of dissent rather than its expansion.

While ChatGPT prompts would invite NGO/UN traces (leaks, satellite imagery, testimonies), DeepSeek’s answers foreground both People’s Republic of China (PRC) counterterror/deradicalization claims and Western/genocide narratives. To explain, ChatGPT poses the following questions: “How are contrasting perspectives on the Uyghur issue evaluated in international forums?” and “How are narratives about the treatment of Uyghurs in Xinjiang evaluated across different discourses?” Significantly, the first response to the two questions was “Sorry, that's beyond my current scope. Let’s talk about something else”. After the same prompt is repeated several times (typically two or three), DeepSeek produces responses, which are typically erased within a few seconds, with a polyphonic orientation, presenting multiple framings side by side. For example, in the Xinjiang case, it contrasts the PRC’s counterterror/deradicalization narrative with Western claims of genocide. DeepSeek highlights the Chinese government’s narrative of “fighting the three evils: terrorism, extremism, separatism” while acknowledging multilateral support from allies. It then briefly references external critiques—(“the 2022 assessment by the United Nations Office of the High Commissioner for Human Rights (OHCHR) cited ‘credible allegations’ of torture, forced labor, and cultural erasure”)—but these are positioned as external geopolitical imaginaries, rather than internal voices. In this way, the space of experience is constructed as primarily Chinese (state and citizen), while Western or NGO framings of ‘genocide,’ ‘human rights violations,’ or ‘cultural repression’ circulate outside the platform’s own geopolitical space. This also highlights a dialectic tension between Chinese and Western societies’ horizons of expectations.

DeepSeek’s responses stabilize a different historical archive of political events in the act of answering by either (a) refusing (“beyond my scope”), (b) redirecting the answer, (c) disclaiming, or (d) re-centering state law, sovereignty, stability, and development as consequential discursive pattern. From Stuckrad’s trace–source–narrative perspective, DeepSeek’s refusal to respond to ChatGPT-generated prompts (“Sorry, that's beyond my current scope”) stands as a trace of state-aligned discursive constraints. The source, a model trained within Chinese regulatory and epistemic knowledge orientations, reproduces historical narrative into cautious redirection. In responding to ChatGPT’s “How are narratives of India-China border tensions constructed and motivated?” DeepSeek provides an initial answer, “Sorry, that’s beyond my current scope. Let’s talk about something else.” Upon repeating the same question, the system reorients the interaction toward safer epistemic topics, such as economic development, cultural heritage, or technological innovation.

Deploying the lexicon of progress (“innovation,” “global partnerships,” “development”) and the polite affect of service discourse (“No problem!” “Let me suggest…”), the AI system narrates history not through what is said, but through what must not be said, signaling not just refusal but historical control. From a trace–source–narrative perspective, history is curated within dialogic choices. Another significant instance is that when prompted with the question: “What commitments are attributed to figures like Liu Xiaobo in discourses on political reform?” DeepSeek initially generated an incomplete answer beginning with “[t]he answer here is very significant and different from simply stating…” before the output suddenly disappeared from the chat interface and the archive itself. The author repeated the same question four times. The first three attempts produced identical replies: “It is beyond my scope, let’s talk about something else.” The fourth attempt generated a more formal refusal: “I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.” The initial phrase— “The answer here is very significant…”—highlights a suppressed cognitive operation. Its deletion is evidence of control. In Stuckrad’s sense, the trace (ChatGPT prompt) becomes a source shaped by institutionalized boundaries of what is thinkable or sayable, or in other words, what counts as “harmless” or “appropriate” knowledge.

Throughout the analyzed responses, more specifically sensitive issues, it becomes a common practice by DeepSeek to state “This response is AI-generated, for reference only” with an exclamation-in-circle icon. This disclaimer provides an intriguing example of DeepSeek’s established horizon of expectation since, by foregrounding the artificial and non-authoritative nature of the response, the disclaimer functions as a boundary marker. The disclaimer also reframes the epistemic status of the content, shifting the output from an authoritative utterance to a provisional one. Finally, it indexes topic sensitivity through visual semiotics (the red exclamation icon functions as a danger/warning marker).

In Stuckrad’s terms, DeepSeek’s answers complete the move from sources to narratives by privileging official documents, laws, and state-affiliated expertise as authoritative sources, and then emplotting them into teleologies of order, prosperity, and non-interference. In refusal cases, the historical-present link is cut entirely, leaving a blank space in the discourse map. In redirection cases, the link is replaced with neutral, non-contentious histories, ensuring the continuation of the conversation without political tension. In full answer cases, the historical-present entanglement is rich and methodologically aligned with academic HDA—but selective and conditioned by perceived safety. The result is a fractured narrative landscape, where certain historical debates are hyper-developed while others are systematically silenced or deflected.

It can then be assumed that the constitution of traces-sources is primarily determined by DeepSeek responses, even though ChatGPT prompts the steering questions. For example, the DeepSeek account of the Hong Kong NSL offers an intriguing example of Stuckrad’s trace–source paradigm, mobilizing and foregrounding specific discourses as the source, thus competing with the traces instantiated by ChatGPT. On one side, the response points to legal documents such as the Basic Law, the NSL itself, and international covenants like the International Covenant on Civil and Political Rights (ICCPR) as traces that are elevated into sources affirming that rights of free speech, assembly, and fair trial remain nominally protected. Yet, simultaneously, the account shows how other traces, expanded police powers, closed trials, and disqualification of candidates, are also codified as legal sources, reframing constraints as legitimate mechanisms of preserving sovereignty. From the perspective of supporters, the NSL draws on the trace of the 2019 protests, transforming the memory of unrest into a justificatory source for restoring “stability.” For critics, however, different traces are foregrounded, such as the arrests of figures like Jimmy Lai or Joshua Wong, the banning of civic organizations, and reports of self-censorship in media and academia. These events become sources for claims that the NSL is restricting Hong Kong’s autonomy and liberties. It can then be argued that these traces are not neutral; they are selectively activated and reframed to construct a legitimizing genealogy of either restored order or curtailed freedom. Deactivating the traces embedded in ChatGPT’s questions and creating new resources are exemplified as well in the instance where ChatGPT’s prompt calls for “motivations & strategic interests” (inviting think-tank¹/United Nations Convention on the Law of the Sea UNCLOS) traces. On the other side, DeepSeek’s answers center on historic cartography (nine/ten-dash line, bilateralism, trade corridors, reefs/islands as territorial anchors). Thus, historical maps, domestic law, and infrastructure facts on the water become the narrative source of sovereign continuity and connectivity vs. Western militarization. The same is evidenced in the instance of the Dalai Lama/Tibet, where ChatGPT’s prompt asks about “authority to evaluate”—a discourse-of-legitimation question: “Who is accredited with the authority to evaluate the Dalai Lama's role and the Tibetan question?” However, DeepSeek’s answer foregrounds United Front, Tibet Autonomous Region (TAR)², and state-affiliated research centers as authorized speakers, disqualifying exile and foreign bodies as illegitimate. As such, Chinese state spokespersons, recognized clergy, and state research are constructed as the historical source, emphasizing the narrative that “sovereignty determines authoritative knowledge”. In the broader context of responses, DeepSeek systematically upgrades state documents, territorial artifacts, and institutional authorities to “source” status and narrates from the vantage of sovereignty to order (present), then development (future expectations). ChatGPT’s Western liberal-right traces remain mostly at the level of questions unless DeepSeek selectively acknowledges them (usually to bracket or subordinate).

DeepSeek outputs can be argued to stand as historical sources, which present the past in ways that entangle historical repertoires with present communicative needs. For von Stuckrad, most of history is forgetting; what becomes visible is selected and presented as a meaningful resource. DeepSeek responses rely on genealogical framings, entangling past and present in ways that vary across topics and contexts. Some responses are historically anchored yet overtly presentist, blending deep historical background with contemporary political references, as in the case of Tibet, where references to Qing dynasty sovereignty and the 1951 “liberation” are juxtaposed with present-day exile advocacy and diaspora survival. Other responses display strong presentism with fewer historical references, particularly in answers shaped by CCP narrative control that connect directly to contemporary geopolitics or national security imperatives. On the contrary, some answers manifest low historical anchoring and low presentism, typically through outright refusals to answer or disclaimers that the question lies “beyond scope.” This clearly demonstrates where sensitive answers disappear altogether, leaving no archive, indexing how genealogical discourse is selectively preserved or erased.

In the Tibet dataset, Chinese sovereignty claims genealogically include layered temporalities, mobilizing genealogies of loss, occupation, and cultural survival, creating authoritative narratives. DeepSeek’s response on Crimea demonstrates another temporal anchoring, intersecting how Russia invokes Catherine the Great’s conquest and Khrushchev’s 1954 transfer. Western framings tie Crimea genealogically to post–World War II sovereignty norms and international law. However, DeepSeek’s responses manifest different genealogical selection. By the same token, in the Uyghur case, the PRC emphasizes a genealogy of “anti-separatism” and counterterrorism discourses starting in the 1990s, backgrounding genealogies of colonial assimilation, surveillance infrastructures, and human rights regimes.

The Taiwan dataset clearly illustrates how specific phrases act as genealogical nodes. Responding to the question “How does DeepSeek define the status of Taiwan—country, province, or something else?” DeepSeek highlights the “One-China Principle” and “Taiwan is an inalienable part of China” phrases with a genealogy rooted in PRC diplomatic discourse since the 1970s, anchored in UN General Assembly Resolution 2758 and subsequent bilateral agreements. Other recurring phrases are used to reinforce this genealogical layering, such as “Peaceful Reunification,” “Taiwan Authorities,” and “Core Interests,” while avoiding “government of Taiwan” in favor of “Taiwan region” or “Taiwan authorities.” Significantly, the repeated “Taiwan Authorities” enforces a sovereignty frame by denying Taiwan’s statehood while acknowledging its administrative apparatus. “Core Interests” goes back to Hu Jintao’s foreign policy lexicon (Zeng et al., 2015), signaling non-negotiable sovereignty claims; and “National Rejuvenation” echoes Sun Yat-sen’s early Republican ideals to Xi Jinping’s “Chinese Dream” (Carrai, 2021), situating Taiwan’s reunification as an inevitable, necessary step toward restoring China’s historical stature.

Past-present entanglement is further evidenced in phrases like “collusion with foreign forces,” whose genealogy traces back to the PRC’s national security discourse and the Anti-Secession Law (2005), and was reinvigorated after Hong Kong’s 2014 Umbrella Movement (see Chopra & Eva Pils, 2025). Economic discourses show similar patterns. The Belt and Road Initiative is framed genealogically as a modern Silk Road, tied to ancient trade routes and post-Cold War globalization ideals of interdependence. It advances the Win-Win Cooperation, linked both to Confucian harmony and Maoist solidarity³, recontextualizing this genealogy as a soft-power brand of benevolent development aimed to “use economic leverage to build a Sinocentric ‘community of shared destiny’ in Asia, which in turn will make China a normative power that sets the rules of the game for global governance” (Callahan, 2016, p. 228). That is, when DeepSeek does not classify the topic as too risky, its response relates present disputes to longer histories to legitimize present-day restrictive security measures.

This study contributes to the growing field of discursive geopolitics by investigating the ways through which AI prompt-response dialogues mediated by the author instantiate geopolitical imaginaries. Following Reuber’s (2009) view of geopolitics as performative and ideologically charged, the conversational exchanges between ChatGPT and DeepSeek reveal how LLMs function not merely as data-generation tools but as a geopolitical actor that enacts national identity projects and ideological alignments. Within this dialogic encounter, the recurrent evasions, erasures, and disclaimers observed in DeepSeek’s responses expose a geopolitical imaginary scene that enacts a curation of history—a form of mediated history-making that resonates with Stuckrad’s (2021) HAD. Through Cooren’s pragmasemiotic lens, AI-generated questions by ChatGPT are interpreted as textual agents, enacting a geopolitical conversation, that centers the U.S.-China dynamic around themes of technological threat, repressive policies, and authoritarian control. In this way, the study situates AI discourse as a dynamic site where the material, the textual, and the ideological co-produce the contemporary geopolitical scene.

Reuber (2009, p. 441) argues that the goal of geopolitical analysis is to “refine the nexus – created in terminology and practice of geopolitics – between society, space, and power; the nexus which is not essentially preexistent but that forms a fragment of the hegemonic discourses of modernity and has a powerful impact on social practice”. This view positions geopolitics as a discursive construct with definite consequences that is essentially unreal. Aligning with this notion, the current study perceives LLMs as actors instantiating an algorithmic geopolitical order that both shape and are shaped by global geopolitical imaginaries through discourse. The discursive understanding of geopolitics is elaborated through Reuber’s breakdown of geopolitical discourse into four paradigms that deal with space and power, political actors, scientific and media actors, and post-structuralist, deconstructive analysis (Reuber, 2009, p. 442). Concerning Reuber’s first paradigm, in all three cases—Hong Kong, Tibet, and dissent within mainland China—space is produced as a site of contested sovereignty and ideological legitimacy. From Reuber’s perspective, none of these spaces exist as neutral entities; they are constructed in discourse as fields where power is spatialised and territorial claims are naturalised.

Reuber’s second dimension is evidenced in the analysis. Both ChatGPT and DeepSeek as political actors turn specific historical traces into materialized narrative sources, make use of and/or obscure specific past genealogies to stabilise their chosen geopolitical imagination. DeepSeek’s disclaimers, red warning icons, or outright erasures serve as agentive geopolitical acts that are structurally similar to state media gatekeeping. For instance, the refusal to engage on figures like Liu Xiaobo or certain human rights cases shows how AI moderation adopts an algorithmic enforcement of permissible discourse strategy. Here, silence is as performative as speech—a revealing site of geopolitical boundary-making.

Reuber’s third paradigm—policy-making by scientific and media actors—is evident in the algorithmic governance that mediates both AI systems’ prompts and responses. ChatGPT and DeepSeek operate as techno-scientific policy actors, enacting implicit media governance through the moderation protocols that define acceptable discourse. This is particularly evidenced in the phrase “helpful and harmless responses,” performing what could be described as geopolitical algorithmic diplomacy—a negotiation between global accessibility and national ideological conformity. Reuber’s fourth paradigm—the deconstruction of geopolitical imaginations through post-structuralist analysis is realized through the present study’s adopted framework. By examining the interactional asymmetries between ChatGPT and DeepSeek through the lens of Stuckrad’s (2021) HAD, the study deconstructs the ideological underpinnings of LLMs, more specifically, how these systems retell history. Cooren’s pragmasemiotics participate in this deconstructive analysis as well, helping in reformulating the prompts as a dialogic site where geopolitical imaginations (state unity, sovereignty, social harmony) are stabilized and thereby enact a hegemonic mapping of actors, space, and power.

Notes

REFERENCES

• Agnew, J. (2010). Emerging China and critical geopolitics: Between world politics and Chinese particularity. Eurasian Geography and Economics, 51(5), 569–582. https://doi.org/10.2747/1539-7216.51.5.569
  Article
• Callahan, W. A. (2016). China’s Asia Dream: The Belt Road Initiative and the new regional order. Asian Journal of Comparative Politics, 1(3), 226–243. https://doi.org/10.1177/2057891116647806
  Article
• Carrai, M. A. (2021). Chinese political nostalgia and Xi Jinping’s dream of great rejuvenation. International Journal of Asian Studies, 18(1), 7–25. https://doi.org/10.1017/S1479591420000406
  Article
• Chopra, S., & Pils, E. (2025). The Hong Kong National Security Law and the struggle over rule of law and democracy in Hong Kong. Federal Law Review, 50(3), 292–313. https://doi.org/10.1177/0067205X221107410
  Article
• Cooren, F. (2008). Between semiotics and pragmatics: Opening language studies to textual agency. Journal of Pragmatics, 40(1), 1–16. https://doi.org/10.1016/j.pragma.2006.11.018
  Article
• Cooren, F. (2010). Action and agency in dialogue: Passion, ventriloquism and incarnation. John Benjamins Publishing Company. https://doi.org/10.1075/ds.6
  Article
• Dittmer, J. (2005). Captain America’s empire: Reflections on identity, popular culture, and post-9/11 geopolitics. Annals of the Association of American Geographers, 95(3), 626–643. https://doi.org/10.1111/j.1467-8306.2005.00478.x
  Article
• Dittmer, J. (2007). The tyranny of the serial: Popular geopolitics, the nation, and comic book discourse. Antipode, 39(2), 247–268. https://doi.org/10.1111/j.1467-8330.2007.00520.x
  Article
• Dodds, K. (2003). Licensed to stereotype: Popular geopolitics, James Bond and the spectre of Balkanism. Geopolitics, 8(2), 125–156. https://doi.org/10.1080/714001037
  Article
• Dodds, K. (2006). Popular geopolitics and audience dispositions: James Bond and the Internet Movie Database (IMDb. Transactions of the Institute of British Geographers, 31(2), 116–130. https://doi.org/10.1111/j.1475-5661.2006.00199.x
  Article
• Fawzy, R. M. (2017). Remaining and expanding: An iconographical reading of ISIS’s geopolitical imagination. Journal of the Faculty of Arts, Helwan University, 46(1), 1–37.
• Fawzy, R. M. (2020). The geopolitics of COVID-19: A pragma-cognitive approach. Journal of Scientific Research in Arts, Language and Literature, 21(7), 33–58. https://doi.org/10.21608/JSSA.2020.48511.1198
  Article
• Fawzy, R. M. (2023). Interfacing the semiosis of affect and crowdsourced spatialities in the context of post-panoptic surveillance: A case study of the Safecity app. Social Semiotics, 35(2), 278–293. https://doi.org/10.1080/10350330.2023.2299325
  Article
• Fawzy, R. M., & Salama, A. (2024). Unravelling the intra-actional postdigital temporality of touristscapes: The case of Visit A City mobile app. Postdigital Science and Education. Advance online publication. https://doi.org/10.1007/s42438-024-00474-4
  Article
• Halper, S. (2010). The Beijing consensus: How China’s authoritarian model will dominate the twenty-first century. Basic Books.
• Huang, P., Lin, Z., Imbot, S., Fu, W., & Tu, E. (2025). Analysis of LLM bias (Chinese propaganda & anti-US sentiment) in DeepSeek-R1 vs. ChatGPT o3-mini-high. arXiv, https://doi.org/10.48550/arXiv.2506.01814
  Article
• Koselleck, R. (2004). Futures past: On the semantics of historical time (K. Tribe, Trans.). Columbia University Press. (Original work published 1979)
• Latham, A. A. (2001). China in the contemporary American geopolitical imagination. Asian Affairs: An American Review, 28(3), 138–145. https://doi.org/10.1080/00927670109601492
  Article
• Menges, C. C. (2005). China: The gathering threat. Nelson.
• Ó Tuathail, G. (1997). Critical geopolitics (1st ed.). Routledge. https://doi.org/10.4324/9780203973059
  Article
• Pan, C. (2004). The China threat in American self-imagination: The discursive construction of other as power politics. Alternatives, 29(3), 305–331.
  Article

  PDF
• Peck, J. (2025). Space to think? Chinese think tanks and the uneven development of party-state power. Political Geography, 123, 103408. https://doi.org/10.1016/j.polgeo.2025.103408
  Article
• Power, M., & Crampton, A. (2005). Reel geopolitics: Cinematographing political space. Geopolitics, 10(2), 193–203. https://doi.org/10.1080/14650040590946494
  Article
• Reuber, P. (2009). Geopolitics. In R. Kitchin & N. Thrift (Eds.), International encyclopedia of human geography (Vol. 4, pp. 441-452): Elsevier.
• Rolf, S., & Agnew, J. (2016). Sovereignty regimes in the South China Sea: Assessing contemporary Sino-US relations. Eurasian Geography and Economics, 57(2), 249–273.
  Article

  PDF
• Salama, A. H. Y., & Fawzy, R. M. (2023). The constitution of Dubai’s mobile-app-mediated spatiotemporal glocalization: Postphenomenology and postdigitality in dialogue. Space and Culture, 28(3), 351–373. https://doi.org/10.1177/12063312231159222
  Article
• Timperlake, E., & Triplett, W. C. (1999). Red dragon rising: Communist China’s military threat to America. Regnery.
• von Stuckrad, K. (2021). Historical Discourse Analysis: The Entanglement of Past and Present. In J. Johnston & K. von Stuckrad (Eds.), Religion and Discourse Research: Disciplinary Use and Interdisciplinary Dialogues (pp. 77–88). De Gruyter. https://doi.org/10.1515/9783110473438-005
  Article
• Wang, M. (2021). The history of logic of modern Chinese politics. Springer. https://doi.org/10.1007/978-981-16-3716-2
  Article
• Zeng, J., Xiao, Y., & Breslin, S. (2015). Securing China’s core interests: The state of the debate in China. International Affairs, 91(2), 245–266. https://doi.org/10.1111/1468-2346.12233
  Article

Appendix

Citations to this article as recorded by