The Discursive Construction of Generative AI
The papers filed here converge on a single methodological wager: that generative AI does not simply exist as a technology to be described, but is made through the accumulated work of framing—by journalists, industry actors, platforms, and communities of practice. Taken together, they trace three concentric circles of imaginative work: news media as a site of public epistemology, the industry and platform actors who seed and manage those narratives, and finally a media-theoretical register that asks what generative AI is doing as a medium once the hype is bracketed.
News media as an imaginary engine
Several papers treat journalism itself as an object of study rather than a neutral conduit. Nguyen2026-vm offers the most granular empirical account, showing that mentalistic-agentic framings of LLMs are real but far from dominant—only 8.8% of LLM-centric articles anthropomorphize explicitly, and this anthropomorphism serves heterogeneous functions, from uncritical amplification of corporate narratives to pointed critique (“bullshitters”). This complicates any simple story of media-driven AI hype, suggesting instead that a techno-capitalist master frame organizes coverage globally while regional inflections (risk/ethics vs. economic utility) and editorial idiosyncrasies do the finer work. Wang2025-zy extends this comparative logic across the UK, US, China, and India, operationalizing Cave and Dihal’s utopia/dystopia framework to show that national press systems recombine hopes and fears according to their own political-economic stakes—confirming that “the AI imaginary” is not singular but pluralized along national lines, a point Nguyen2026-vm arrives at from a different angle by finding regional variation in secondary rather than master frames.
Dodds2026-df reframes this entire literature by insisting that AI hype in journalism is not merely a discursive bubble but a structuring infrastructure—one embedded in newsroom communities of practice, sourcing routines, and resource allocation. This is a productive friction with Nguyen2026-vm and Wang2025-zy: where those papers map framing as a textual/thematic phenomenon, Dodds2026-df insists on tracing how framing choices (especially repetitive sourcing of tech executives) materially redistribute legitimacy and render alternatives—labor, environmental cost—invisible. The special issue’s call for “breakdown and repair” as an alternative to mere debunking supplies a normative horizon that the more descriptive framing-analysis papers lack.
Industry discourse and the politics of futuring
A second cluster shifts the lens from journalism to the industry actors who supply journalism’s raw material. Stanusch2026-ec examines the November 2023 Sam Altman controversy as a moment when latent industry imaginaries—Longtermism, Regulatory Ambivalence, Techno-Hagiography—became visible, and shows how the industry manages critique through “premediation” (displacing concerns into speculative futures) and “preclusion” (positioning itself as the only legitimate solver of problems it creates). This dovetails with Dodds2026-df’s account of “open-ended technological inevitability” as a temporal narrative strategy, and both papers identify absences—environmental cost, labor, historical comparison—as constitutive rather than incidental to hype work.
Hepp2026-oi complicates the industry-discourse picture further by distinguishing the “loud futuring” of CEOs and investors (the register Stanusch2026-ec largely studies) from a “quiet futuring” among Bay Area labs, Effective Altruist and Rationalist communities, and pioneer groups. This is a genuinely structural intervention: it argues that critical AI studies has overfocused on spectacular pronouncements and underexamined the spatially anchored, community-based reservoir of ideas (especially long-termist and x-risk framings) that loud futuring later condenses and amplifies. Read alongside Stanusch2026-ec, whose “Longtermism” imaginary is traced explicitly to EA/Rationalist roots, Hepp2026-oi supplies the ethnographic and geographic texture behind that imaginary’s production—while insisting, importantly, that this landscape is internally fragmented and only apparently apolitical.
Richter2026-bt’s longitudinal Twitter analysis provides a historical backbone for these industry-discourse papers, showing that the current dominance of tech-industry and influencer voices in US AI discourse is not a timeless feature but a trajectory that consolidated across 2012–2021, in sharp contrast to Germany’s institutionalization toward academic, advocacy, and governmental actors. This paper usefully historicizes Hepp2026-oi’s Silicon Valley present and Stanusch2026-ec’s 2023 snapshot: the imaginaries mapped by those two papers are the settled residue of a decade-long stakeholder consolidation, and the German/US divergence Richter identifies plausibly explains why Wang2025-zy and Nguyen2026-vm find such different secondary emphases across national presses.
Platforms as a distinct site of framing
Weinbrand2026-sf narrows the aperture to a single contested artifact—Google’s AI Overviews—showing how platform, press, and commercial (SEO) actors compete to legitimize or contest a specific generative-AI feature. Its findings that user perspectives and environmental/democratic implications are conspicuously absent from all three discourses directly echoes the absences flagged by Dodds2026-df and Stanusch2026-ec, suggesting a recurring structural blind spot across scales of AI discourse—from industry controversy to platform rollout to routine news coverage. The paper’s attention to Google’s hedging (“experimental” labels) as a rhetorical-legal strategy also extends Stanusch2026-ec’s “preclusion” concept to a mundane, non-crisis setting.
Publics on the receiving end
Fattorini2026-bo and Suk2026-ai pivot from producers of framing to its uptake. Fattorini’s Italian survey data empirically grounds the “critical ambivalence” that the framing-analysis papers largely infer discursively: Italians combine rising AI use with rising perceptions of threat, and symbolic imagery (humanoid robots) persists even as direct experience with chatbots reshapes association patterns—an uptake-side confirmation that the utopia/dystopia oscillation Wang2025-zy finds in newspapers has a real correlate in audience attitudes. Suk2026-ai, more programmatic than empirical, proposes folding generative AI into the lineage of media-trust research, implicitly arguing that all the framing work traced above—mentalistic language, national imaginaries, industry hype, platform discourse—ultimately funnels into questions of trust that communication scholars already have theoretical tools to address, if adapted across individual, institutional, and societal levels.
From imaginaries to medium theory
The final two papers step outside the frame/imaginary vocabulary altogether, asking not how AI is talked about but what kind of thing it is as a cultural artifact. Manovich2026-ih brackets ethical and social critique to theorize generative AI as a medium with its own affordances—probabilistic variability, built-in “media cognition,” structural conservatism, and style-content entanglement—treating it as continuous with photography and painting rather than as a sociotechnical controversy. This is a deliberate counterpoint to the rest of the topic cluster: where Dodds2026-df, Stanusch2026-ec, and Hepp2026-oi insist that AI discourse is inseparable from power, labor, and politics, Manovich insists on the analytic value of setting that aside to ask what the medium formally does and resists.
Galip2026-ix occupies a fascinating middle position, taking seriously precisely the “degenerate” outputs (slop, brainrot) that Manovich’s medium-theoretical account and the industry imaginaries alike tend to exclude or dismiss. Via Ngai’s theory of the gimmick, the paper reads slop’s simultaneous excess and deficiency—working “too little” and “too hard”—as exposing the same contradictions that Dodds2026-df locates in hype’s obscured labor and Stanusch2026-ec locates in preclusion’s absorption of critique, but does so at the level of aesthetic form and affect rather than institutional discourse. In this sense it closes the loop of the topic: if the earlier papers ask how generative AI is imagined and legitimated, Galip2026-ix and Manovich2026-ih ask what remains to be theorized once the imaginaries are stripped away—the medium’s formal properties on one hand, and its disavowed, degraded output on the other.