In June 2024, reflecting on his experiment using Generative Artificial Intelligence (GenAI), journalist Emanuel Maiberg wrote that "flooding the internet with an infinite amount of what could pass for journalism is cheap and even easier than I imagined, as long as I didn't respect the craft, my audience, or myself." (Maiberg 2024, np) Maiberg captures well three intertwined challenges facing contemporary digital journalism.
The first is journalism's ongoing economic uncertainties; it costs virtually nothing and requires little paid labor to publish content that looks like news. The second is the dizzying speed with which GenAI (using large language models and off-the-shelf tools to synthetically create news-like content) has appeared as a topic of concern and excitement for journalists, publishers, and audiences alike. And the third prompts existential questions about why journalism exists, whether journalistic work is unique, and how journalists and audiences value themselves and each other when content is cheap, attention is fragmented (Klimeš 2022), and people often confuse audiences, users, and publics.
Put differently, there is simultaneous panic: business models are collapsing, data-driven newsrooms dominate, and a growing number of journalists are in unions, or wish they were (Greenhouse 2022;Liedke 2022). As this special issue traces, some of these hopes and fears are hype-what Communication scholar Devon Powers calls a "state of anticipation generated through the circulation of promotion, resulting in a crisis of value." (Powers 2012, 863) Though hype can be easily dismissed as temporary and vapid, it can dramatically shape how people understand technologies and values. Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 4 Here we examine hype through the joint lens of organized labor and GenAI-two intertwined ways that journalists discover, articulate, and debate the value of what they do as they grapple with technology that has appeared quickly and largely from outside their newsrooms and professional cultures. We trace this intersection through scholarship on hype as promotional communication, expectation setting, and attempts to stabilize technologies. We argue that unions' responses to GenAI are not simply about organized labor defending news and jobs-protecting against hallucinations and automation-but they are also about media workers' trying to stabilize a new, opaque, and rapidly changing technology. In doing so, they show what they think their work is, why it has value, and what working conditions their success requires.
We trace this through a discourse analysis of trade press stories, union statements, labor leader testimonies, calls for action, and proposed and successful contracts. We find patterns within and across these stories, documents, statements, and contracts, and analyze how the discourse (a) defines GenAI as a journalistic problem, (b) articulates journalistic work in GenAI terms, and (c) suggests and celebrates solutions to GenAI's challenges to journalism. We end by reflecting on what these patterns suggest about the press's relationship to new technologies and note concerns that unions do not seem to have about GenAI, topics that are largely absent from the discourse but that could animate its future.
Though boosters and critics alike often simplistically see "hype" as excitement or fatigue, the word suggests a rich and often dismissed way to understand technological promises, anticipations, fears, and stabilizations. Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 Eager to predict, control, and commodify technological change, management consultants have long mapped how novel technologies reflect and shape emerging economies and cultures.
One of the more popular maps, Gartner's "Hype Cycle," traces how a technological "trigger" creates a burst of visibility and prompts an "expectations peak," and then settles into a "trough of disillusionment," a "slope of enlightenment" and, eventually, a "plateau of productivity" (cited in Borup et al. (2006, 291)). Though this and similar cycles promise a "highly linear understanding of a technology's path dependency," they fail to account for how "technologies actually change over time in a continual and practical process of reconfiguring and being reconfigured in use." (Borup et al. 2006, 292) Indeed, hype cycle studies show technologies following different paths, hype talk having little connection to technological capabilities, and hype bringing fields of practice together and hastening their fragmentation (van Lente, Spitters, and Peine 2013). The promise of predicting and controlling technological hype fails as both technologies and discourses around them change-as people adopt, reshape, and reject tools and talk.
Rather than representing contained or consistent innovation, hype is more correctly seen as intertwined social, technical, economic, cultural, and political processes of (a) popular communication, (b) expectation setting, and (c) technology stabilization.
As popular communication, Devon Powers argues that hype is a "state of anticipation generated through the circulation of promotion, resulting in a crisis of value." (Powers 2012, 863) Tracing hype across digital music infrastructures, finance systems, and trade press publications, Powers shows people celebrating, amplifying, hoping for and fearing futures that seemed unimaginable. Hype's promotional forces include advertising and public relations industries (Davis 2013), influencer cultures (Hund 2023), charismatic innovators (Ames 2019), and commodified authenticity (Banet- Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 6 Weiser 2012)-all of which drive today's AI hype (Bourne 2024). Such hype is relational, cultural, and normative; it is not to be dismissed or naively celebrated but read as discursive evidence of what people think should or could be valued. Technological innovations may trigger and frame promotional discourse, but they emerge from deeper genealogies of excitement, hope, cynicism, backlash, fear, fatigue, and valuation (Powers 2012). As we see in this inquiry, closely tracing GenAI as an emerging technological force is a way to see how journalists and publishers variously feed, endure, resist, or contextualize hype cultures that the press cannot ignore or insulate itself from.
Technological hype is also how people understand sociotechnical worlds-how they think humans and systems work together to structure beliefs and actions (Suchman et al. 1999).
Expectations reflect existing sociotechnical worlds-how people think they work-but also hopes, fears, and imaginations about what future sociotechnical arrangements could or should bring (Jasanoff 2015). People's willingness to endure celebratory hype and failed innovations depends upon how new they judge a technology to be, how close they are to its use (Brown and Michael 2003), how clearly they see its benefits, and how well its potentially positive outcomes align with their values (Brown 2003). Hype is thus "situated" (Brown and Michael 2003, 3) in the "networks and activities, certainties and uncertainties" that people have before hype starts. Though hopes and fears may be felt anew, hype always reflects what people already know and value. Indeed, when hyped technologies fail-sinking into the "trough of disillusionment"-people draw on different resources (existing knowledge, successful experiences, clear identities) to explain the failure. Past disappointments "tend to be rationalized" (Borup et al. 2006, 290) as people revert to earlier thinking, revise and renew hopes, search for alternatives, and resign themselves to ongoing uncertainty (Geels and Smit 2000). People who see themselves as most impacted by hyped technologies cope, in part, by seeing a "prospective structure" (Lente and Rip 1998, 203) of action-tactics and strategies to contain, endure, or adapt to a technology that they expect to impact them. Hype's real power is its capacity to influence people's expectations of how sociotechnical systems could or should work, and their agency to reshape them. People's experiences, memories, and confidence to act help them to understand hype, contextualize its promises, and prepare for its futures.
Finally, as scholars of technology stabilization find, hype can be a way to understand how systems seem to change. While "hype cycles" like Gartner's model place technological stability in a "plateau of productivity," Science and Technology Studies (STS) scholars see nuanced dynamics driving stability. A classic, oft-cited model of stabilization in STS's Social Construction of Technology (SCOT) tradition traces new technologies from "interpretive flexibility" (when people debate technologies' meanings and technological forms change) through "closure and stabilization" (when "relevant social groups" agree that a technology is a desired and acceptable solution to an agreed upon problem) (Pinch and Bijker 1984).
Though the SCOT model continues to inform studies of technological stabilization, critics find it simplistic. It too narrowly defines social groups, leaving little room for different practices and interpretations; it fails to define technologies precisely, ignoring how similarities drive stabilizations; and it pays little attention to the role of language in articulating change and stabilization (Humphreys 2005). Newer scholarship finds people stabilizing technologies locally and temporarily, agreeing to define hyped and controversial technologies in ways that let them make problems tractable (Venturini and Munk 2022;Marres et al. 2024). Stabilizing a technology's social and political meaning is not one-off or uniform; it requires compromise, tolerating ambiguities, and seeing technologies as political objects (Marres and Lezaun 2011) The SCOT model-and critiques of it-shows how technological hype may have a popularized and promoted trigger, but it is an endless, relational, and deeply political process with people debating a technology's scope, power, and future, as they try to accelerate, slow, freeze, or redirect technologies in ways that help them address shared problems.
Hype is thus an analytic-a way to see people interpreting technologies that have appeared quickly, from outside their experiences and expectations, and with potentials to disrupt their work and values in ways that are hard to predict amidst competing interests. As sensemaking work, hype is: communication about practices, power, and value; expectations of sociotechnical systems; and stabilizations of knowledge and problems.
As journalists grapple with and try to stabilize GenAI hype they use a mix of public-facing communication, professional self-regulation, and experimentation.
A spate of journalism about GenAI has appeared since Fall 2022, tracing its power to disrupt industries (Nielsen 2024), spur debate amongst charismatic figures (Ananny 2024), and pose new legal and regulatory questions (Nelson 2024). Within journalism, there is an active "metajournalistic discourse" (Carlson 2015) about GenAI's impact on reporting, editing, publishing, and the press as a producer of original, timely, fact-based information crucial to democracy. Some publishers have publicly stated how GenAI would or would not appear in newsrooms with many creating guidelines governing its use (de-Lima-Santos, Yeung, and Dodds 2024). Surveys find that people are sometimes comfortable with reporters using GenAI (Fletcher and Nielsen 2024) and that "the news industry in the US seems to be slowly shifting its labor pool in Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 ways that may help it realize the potential efficiency gains from generative AI." (Diakopoulos 2024, np) Journalism scholars and computational social scientists find myriad intersections of news work and AI: in fact-checking and misinformation generation; automated reporting, writing, and investigation; unauthorized appropriation of news data by GenAI models; and potential legal liabilities (Lopez et al. 2023;Cools, Van Gorp, and Opgenhaffen 2023;Maiden et al. 2023;Diakopoulos and Johnson 2021;Lewis, Sanders, and Carmody 2019;Thomson, Thomas, and Matich 2024;Mahony and Chen 2024;Broussard 2015;Broussard et al. 2019). AI and GenAI are increasingly inseparable from journalistic work. Journalism is in the throes of GenAI hype and searching for ways forward. It is showing all the signs of industry and profession interpreting novelty, celebrating and panicking, setting and managing expectations, using familiar language to contextualize upheaval, and working to control the technology's speed and scope of change (Hepp et al. 2023;Slota et al. 2020;Cools, Van Gorp, and Opgenhaffen 2022;Sloane, Danks, and Moss 2024). Journalism is managing GenAI hype as scholarship finds that practitioners often do, through a mix of communication, expectation, and technological stabilization.
One additional and largely understudied way that journalism is stabilizing GenAI hype is through organized labor. Variously called media guilds, news unions, or reporter collectives, there is a long and rich history of journalists organizing to resist and reshape publishers' power, creating for themselves the working conditions that they see as essential for a vibrant press.
Appearing within studies of journalistic professionalization, media capitalism, and print and broadcasting technologies, there is no single beginning to scholarship on organized journalistic labor. In 1925, after World War One and during heightened concern about the profession's sophistication (Hallin 1985), the International Association of Journalists commissioned a study of the "professional and economic situation to help aid journalists in their self-understanding" (Hardt 1995, 4). Media and labor historians largely ignored organized journalism during the 20 th century, with media workers' "fate and the development of the profession in the face of subsequent technological advancements" remaining largely understudied (Hardt 1995, 4).
Recently, though, journalism practitioners and scholars have centered organized labor. In the UK, the National Union of Journalists drove conversations about media ethics and the press's democratic role (Harcup 2002;Gall 2005). In Canada and the US, media worker power consolidated in the early 2000s with "integration of labor unions and worker associations across the converging communication industries of Canada and the United States, bringing together, for Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 example, journalists, broadcast workers, telecommunications and information specialists, and other knowledge workers in one large organization." (Mosco 2008, 109) More recently, amidst media workers' ambivalence about their fetishizations of digital technologies and promised flexibility (Comor and Compton 2015), scholars find freelance journalists and media workers being entrepreneurially flexible while resisting fragmentations of platformized labor. They find and organize themselves, reorganize after corporate consolidations, and publicly defend the value of their work to audiences (Cohen 2019;Proffitt 2021;Salamon 2018Salamon , 2020Salamon , 2022)), all while organizing and bargaining to protect their happiness and wellbeing (Salamon 2023b;Örnebring 2024). Indeed, there has been a rapid uptick in organizing and collective bargaining among journalists and media workers (Cohen and Peuter 2020), and roughly one in six journalists in the United States are now unionized (Liedke 2022). Journalists report that unionized publications have better organizational structures and opportunities, though still failing on "pay equity, diversity, and inclusion" (Assmann 2024, 84).
Today, in the context of GenAI, journalists must organize and bargain while not necessarily agreeing on what GenAI is, how it impacts media workers, which grievances to prioritize, and whose jobs risk automation (Salamon 2024;Woodruff et al. 2024). Especially amidst rapidly changing GenAI infrastructures and newsroom roles-product managers, digital designers, software engineers, reporters, editors, illustrators, and more awkwardly fall under the umbrella of "media worker" (Cools, Van Gorp, and Opgenhaffen 2022; Deuze and Witschge 2017)-organizers may understandably struggle to know exactly who or what they are organizing for and against.
This challenge to interpret and respond to GenAI motivates this paper. More specifically, we are interested in how media unions stabilize GenAI hype-how they understand, frame, and make tractable the rapidly unfolding, hard to understand, and rhetorically sensationalized power of GenAI to shape journalistic work and working conditions.
We ask three types of questions. Inspired by Salamon's studies of media unions' resistance rhetoric and organized labor language (Salamon 2023a), we answer our questions by analyzing three sites of discourse: (a) metajournalistic, trade press discourse about organized media labor's relationship to GenAI; (b) media guilds' own statements (press releases, legislative testimonies, survey findings, contract proposals) about why they are organizing around GenAI; and (c) contract language in successfully adopted collective bargaining agreements focused on GenAI. Discourse analysis (Gee 2014) shows how media guilds use language, articulate analyses, imagine futures, juxtapose claims, offer rationales, challenge interpretations, signal solidarities, and anticipate challenges-all in the service of diagnosing GenAI impacts, defending worker power, and charting solutions through institutional arrangements. While this study has a focus-media guilds, GenAI, synthetic journalism-we see it as a case of journalists articulating their roles and values in an all-too-Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 13 common moment when they are challenged by sociotechnical forces that they did not create, but that they must nonetheless navigate and reshape.
We pause here to note our study's focus on United States-based media unions and Englishlanguage publications. We hope future research can build on this work to provide comparative insights into how journalistic communities practicing in different jurisdictions, and in different languages, experience, discuss, and respond to GenAI (van Dalen 2024; Cools and Diakopoulos 2024).
We created a corpus of three types of materials related to organized labor and GenAI in journalism, drawing upon trade press publications, public union statements, and collective bargaining proposals and agreements, spanning April 2022 through July 2024.
First, we collected twelve texts published either in journalistic trade publications (Columbia Journalism Review, Digiday, Nieman Lab, and Poynter) or trade press of other news organizations (In These Times, and TechCrunch, and WIRED). We searched the websites of prominent trade press publications, tracked newsletters, and selected articles on organized journalistic labor actions or reactions to GenAI.
Second, we collected twenty-three public statements on GenAI by representatives of organized journalistic labor, ranging from national umbrella organizations (Communications Workers of America (CWA), The NewsGuild-CWA, National Association of Broadcast Employees and Technicians (NABET)-CWA, Writers Guild of America, East (WGAE), Writers Guild of America, West (WGAW), National Writers Union (NWU)), to union locals at individual media organizations (Atlantic Media, CNET, Dow Jones, G/O Media, Insider, IGN, Los Angeles Times, Pittsburgh Post-Gazette, Sports Illustrated, Vox Media, Ziff Davis). We formed this subcorpus by searching union social media and websites, press releases, remarks before legislative bodies, platform or principles documents, petitions, reports, surveys, and contract proposals.
Finally, we collected fourteen collective bargaining agreements or agreement excerpts containing new contractual provisions explicitly addressing AI. These agreements include those recently bargained for by publishers and unions at the Associated Press, Dow Jones, Financial Times, Philadelphia Inquirer, Politico, Wilmington News Journal, Delaware Online News, Future PLC, The Onion, Desert Sun, Arizona Republic, Wirecutter, Omaha World Herald, and MinnPost.
We formed this subcorpus by searching union social media and websites for announcements and press releases as well as monitoring journalistic trade press for negotiation updates.
After compiling this combined corpus of forty-nine sources, we iteratively identified and coded discourse patterns. In the first stage, authors independently reviewed all sources and created two separate lists of codes deemed relevant to GenAI problems, worker roles, and proposed solutions. We subsequently synthesized our lists to create a joint list of fifty-eight codes across five categories, shown in Table 1: fieldForces, workerRoles, techProblems, laborProblems, solutions. In the second stage, we independently re-reviewed each source, applying codes to individual text passages resulting in two parallel sets of coded data. We then compared our analyses and resolved discrepancies.
In the third stage, we identified corpus patterns along two dimensions: breadth and depth.
On one hand, we sought to determine how broadly or often a code was discussed across the entire corpus. For each code, we counted the number of sources in which that code appeared. On the other hand, we sought to determine how deeply or in how much detail that same code appeared within sources across the corpus. Accordingly, we counted the total number of times we applied each code. We completed these two calculations separately for each of our sets of coded data, averaged the results, and identified the twenty codes with the highest average counts along each dimension.
Although we make no quantitative claims about our coded data, this method revealed remarkable overlap between these two ways of evaluating the data. As shown in Table 2, nineteen of the top twenty codes were the same whether calculated by breadth or depth, suggesting a correlation between the number of sources within the corpus that mentioned a particular GenAIrelated problem, role, or solution, and the relative attention sources paid to it. The patterns in the coding data are evidence supporting our qualitative analysis, detailed below.
Across our corpus's trade press, union statements, and collective bargaining agreements, we found six dominant frames of GenAI problems, worker roles, and proposed solutions .
should collaborate with workers to establish guidelines.
Publishers are seen as the primary instigators of GenAI developments impacting news organizations. They "determine how to adopt AI into their businesses" (19). They "push to implement AI, particularly generative AI in newsrooms" (13). They "explor[e] options to negotiate with tech companies" (21), prompting "a wave of new licensing deals" (44). And publishers, rather than workers, believe and promote GenAI hype. A WGAE petition notes that publishers are "eager to deploy generative artificial intelligence (AI) tools throughout our industry" (1). A representative from the Insider Union questioned the "breathless excitement…over the 'potential opportunities' our management believes artificial intelligence may present when it comes to creating content" (7).
Publishers are fueling and capitalizing on hype to rapidly impose GenAI on workers.
Publishers are in a "rush"-"rushing out AI-generated content" (1) and "rush[ing] to use AI to churn out low-cost content" (23). As Hamilton Nolan, an elected council member of the WGAE, told the Columbia Journalism Review, "there are plenty of companies in this industry (hello, G/O Media)
where the owners and CEOs will just rush the implementation of AI in whatever way they think will make them the most money the fastest" (25). Workers are rarely involved in decision-making about whether to start using GenAI. Digiday reported that "[m]ost of the newsrooms already using AI did not have preliminary discussions about its use with the unions and the employees they represent" (19). Instead, are left reacting to "shocking news" about a new announcement by management or information gleaned from outside reporting (47). As publishers forge ahead, one G/O Media editor said: "We're all scrambling to try to figure out how to handle it" (19).
But as workers organize and negotiate for greater power, this dynamic is not static. Unions demand publishers "[w]ork collaboratively with us, the workers, before implementing AI tools in our workplaces" (1). They clamor for "a say in how AI is implemented" (20) and "to be in the room and at the table for these decisions" (13). They seek "guardrails" (3)-"[e]mployers must sit at the table and negotiate with workers over ethics codes and work rules" (13). The Writers Guild has "asked employers across the board to engage with workers on AI rules" (24). In response, some publishers offer "only excuses and delays" (Vox Media, 43), or "reserve the right to unilaterally make decisions regarding AI" (Gannett, Dow Jones, 20). Yet others have agreed in new contracts to "discuss in advance the introduction of any new technology" (FT Specialist US, 31) or have "at least one union member involved in conversations about using new tech like AI" (Insider, 19). While the Philadelphia Inquirer "retains the right to experiment and evaluate the ways in which AI could impact its journalism and workflows," it has agreed to "meet and discuss the Company's use of AI" (30). Months after Ziff Davis unilaterally established a strategic partnership with GenAI company Xyla, the editorial union for its publications Mashable, Lifehacker, and PC Mag struck an agreement whose "goal is to create a space, legally binding in the contract, for our union members to be part of discussions about AI": "if Ziff Davis moves ahead with implementing generative AI tools, the contract would require management to form an 'AI subcommittee' to discuss those plans, and union members would be required to have a seat at that table" (48). The discourse reflects widespread unease about publishers' prerogatives leading to union demands for seats at the table.
Publishers are generally not transparent about their GenAI plans. "Many workers across sectors were concerned about the lack of transparency in management attempts to implement AI and enter agreements with AI companies, as well as the lack of transparency surrounding content created by AI appearing on their websites." (13). Some publishers implement GenAI without notice.
"CNET ran chatbot-generated articles without telling anyone they were doing it" (8), and workers there "face a lack of transparency and accountability from management around…plans for artificial intelligence" (16). Other publishers "provide little transparency" around the licensing deals they make with GenAI companies (3). Workers are "informed without warning" (40), learning about these deals for "the first time" in "sudden" announcements (43), without "any inkling that their work would be handed over to OpenAI" (47). There is "little transparency" (44), a "complete lack of transparency" (42), a "lack of information and transparency" that is "extremely concerning" (43).
The Atlantic "refused to answer questions" from its union, directing workers to "outside reporting" (42). "The NewsGuild regularly deals with employers that balk at…requests for information" (3).
GenAI. Unions already "do not trust employers to ethically implement artificial intelligence" (13, 3).
Members have "little to no faith in their employer to use AI responsibly and ethically" (9, 13). G/O Media has given its workers "no reason over the years to believe" the company's assurances about GenAI (6). A "foundation for meaningful collaboration," however, is "advanced notice from the company regarding the development of new workplace technologies and information sharing that provides the union a full understanding of the proposed systems" (15). Workers "deserve to know" about publishers' GenAI plans (42). Unions are "demanding more transparency" (44,10,43) and seeking contractual assurances. Unions at the AP and some Gannett-owned publications have proposed contract language that would "require their respective companies to provide 90 days advance notice of new technology affecting newsroom functions" (20). The Onion Union now "must receive 20-day notice of any AI policy changes" (34), and the Wirecutter Union bargained for "[n]ew requirements granting transparency around the procurement of AI" (38). The discourse points to the possibility of resolving the GenAI trust deficit, at least partially, through greater transparency.
3) Workers ensure the creativity and quality of news work, and they should be trusted to make judgments about whether and how to use GenAI.
Since journalists are responsible for the quality of news work, they must both oversee GenAI's use and have discretion not to use it. Workers are "the first, last and best line of defense against the vast wasteland of [GenAI] garbage that's already coming right at us" (9).
The creativity and humanity of workers are fundamental to journalistic quality. "[H]uman ingenuity and creativity" are "essential to a free press" (12). This "creativity" is "not readily replicable with AI" (2) because "the 'creativity' of AI itself is really just an illusion based on stealing and remixing creative work that humans have already done" (25). No matter how GenAI is integrated into newsrooms, "it takes a team of working journalists to collect, organize, verify and publish facts as news stories" (3), and to "bring contextual knowledge and background information to stories that AI systems cannot replicate" (2). "Humans," therefore, must be "centered in AI processes to add necessary oversight and transparency into machine decision-making." (15). The AP will use Generative AI to "perform the work of news production only with the direct involvement and oversight of employees in compliance with AP standards" (37). At The Onion, The A.V. Club, Deadspin, and The Takeout, "stories, articles or graphics created by [GenAI], in part or in whole, shall first be reviewed by an editorial employee" (34). MinnPost "will treat AI-generated content as source material that requires editing and fact-checking by a human prior to being published." (45) Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 20 More than this, though, unions demand the right to opt out of using GenAI altogether when doing so would conflict with journalistic judgment. Workers are "the experts on their jobs and workplaces" (15). They are "uniquely positioned to help identify beneficial uses of generative AI in our industry and flag potential problems" (1). "It should be workers, not employers, who determine what is and is not assistive and augmentative" (4). The CNET Media Workers Union is "seeking the editorial discretion to not use AI if it fails to meet publishing standards, [and] the right to opt out of using AI without reprisal" (20). The Atlantic Union proposes "that writers can use AI at their discretion, in accordance with journalistic principles and ethics, but they can't be made to use it" (47). Union demands to center workers derive from a shared belief that journalism is a fundamentally creative human practice.
Automation resulting from GenAI was both the most broadly and the most deeply discussed problem. Workers fear that the economic incentives driving publishers to "spend less and extract more, regardless of how it affects quality" (6) will cause further automation across the industry and, ultimately, job replacement. There is also a professed conviction that, despite the hype, some journalistic functions can never be "replaced by bots" because of GenAI's lack of reliability and accountability (3).
While publishers may see GenAI as "a 'cost effective' silver bullet" ( 5), workers see it as a "dangerous" (13) and "ever-growing existential threat" (1) because "[t]asks and even whole job categories have the potential to be automated" (2). Workers worry "about how AI use could play a role in cost-and job-cutting" (7), are "concerned that news companies are attempting to implement AI to juice profit and cut costs" (3), and are "concerned about the blurring of editorial and monetization strategies" (16). "The chief concern among many is the potential loss of jobs" (20). Adam Rogers with the Insider Union captured the general sentiment: "I don't want to be replaced by a machine, of course" (8).
Workers "want to fight to make sure we do not lose jobs to automation" (13). Responding to "[t]he Post-Gazette's attempt to replace our labor with artificial intelligence," Zack Tanner, president of the Newspaper Guild of Pittsburgh, stated: "As newsroom jobs continue to disappear due to corporate greed and mismanagement, we stand firmly against any use of AI that takes work out of union members' hands" ( 14). Such defiance is not only based on a desire to protect workers' livelihoods; it also stems from a repeatedly expressed belief that "AI can never replace actual workers" (7). This belief derives in part from the "destructive, careless, and borderline fraudulent ways that generative AI has already been deployed in our own industry," ways that are inconsistent with journalism as a "fact-based profession" (9). The discourse focuses on publications-CNET, Gannett, Men's Journal, MSN.com among them-that ran "AI-generated stories" that were "riddled with errors" (8,9,17,47). It highlights others-Sports Illustrated, Reviewed-that created "AIgenerated authors" and violated "basic journalistic standards" (10, 9, 47). The Gizmodo Media Group Union declared: "The hard work of journalists cannot be replaced by unreliable AI programs notorious for creating falsehoods and plagiarizing the work of real writers" (6).
It also derives from a conviction that GenAI simply cannot generate journalism because legitimate journalism "requires accountability" (23), and GenAI cannot be "accountable" to "editors" and "readers" (9). No "technological leap" will erase the "qualitative difference between human beings and AI that will never change" (24). The "front-end, creative, thoughtful side of journalism…is something that we as journalists should always be able to explain and be accountable for" (25). New collective bargaining agreements with unions at the AP and MinnPost Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 22 incorporate the AP's "standards around generative AI," which affirm that "[i]t is the responsibility of…journalists to be accountable for the accuracy and fairness of the information we share" (37,45). Agreements with unions at the Arizona Republic and Desert Sun incorporate the "Ethical Guidelines and Policy for Gannett Journalists Regarding AI-Generated or Assisted Content," which state that "journalists must be accountable to their audience and take corrective action if errors are found" (35,36). While publishers may use GenAI to automate the "logistical, busywork, backend side of reporting" (25), there is "no such thing as journalism without journalists" (9). The fight against job loss is thus framed in the discourse not only as a fight to preserve journalists' livelihoods but also as a fight to preserve accuracy as a professional value and, ultimately, accountability in journalism.
Unions want journalists to have more control over the products of their labor. There is widespread alarm over the lack of worker consent both in technology companies' ingestions of news content to train GenAI systems and in publishers' use of those systems to modify or create news content. The lack of control is characterized as a problem of automation and replacement, since "employees can have years of their work product fed into and used to train generative AI, and that technology can then be used to supplement or replace the employee's work" (3). It is also considered a problem of autonomy and reputation. Technology companies have not given workers "a way to opt out of these training corpuses if we don't want to participate" (4). Publishers have disseminated "AI-generated content under the bylines of former staff writers without permission" (48), and workers worry that publishers might use "their image or likeness for content that they do not agree with or that compromises their professional integrity" (3).
In response, the CNET Media Workers Union wants "assurance that AI won't be used to modify content after employees leave CNET" (24). The Future Union ratified a new contract that grants "protection from byline misuse on articles using generative AI" (18,32). Unionized journalists at Dow Jones secured a contractual provision that "[p]rohibit[s] the use of voice recordings in connection with AI-generated speech without the consent of the employee" (46), while those at Ziff Davis stopped publishers' use of GenAI "to impersonate a specific employee or team without their consent" (48).
Organized journalistic labor is generally aligned with publishers when it comes to attempts to wield copyright law against technology companies to take back control of their intellectual property. "The smart move is for journalists-and authors and artists and media companies more broadly-to try to enforce their copyright claims to the absolute max right now" (25), and "media organizations shouldn't allow AI companies to train large language models (LLMs) on journalistcreated content for free" (13). But unions want some control over the licensing deals that publishers strike with OpenAI and others, including "input on the handling and use of [worker] data, especially over its use to train AI systems." (15) "The people who continue to maintain and serve
The Atlantic deserve to know what precisely management has licensed to an outside firm and how, specifically, they plan to use the archive of our creative output and our work product." (42) Workers seek stability and security in their professional lives, and they see contractual limitations on the use of their names, likenesses, and content, along with the ability to opt out of GenAI training, as crucial means to those ends.
Collective bargaining for AI protections and contractual guardrails are perceived as "vital" (47) and as playing a "central role…in negotiating over adoption of AI and other new technologies" (2).
Workers are wary of publishers ' "empty 'trust us' responses" (41). "Though many news outlets have pledged not to replace workers with AI, at least one has already gone back on its word. As a result, unions are pushing companies to commit to protections against AI by enshrining them in contracts, which usually remain in effect for several years." (20) Justin Murphy, a reporter for the Democrat & Chronicle and member of the Newspaper Guild of Rochester, told Digiday: "Whatever the language is, we're asking for acknowledgement [in our contract] from the company that [it] might use AI in any kind of way as it develops, but it is not going to be to replace either the work of journalists or the journalists themselves. We want that in writing." (18) But organized journalistic labor views the "power to fight for protections at the negotiating table" as "tied to organizing campaigns to engage members to put pressure on management." (13).
Unions "look at every potential avenue as a way to push back against AI implementation" (47), most prominently strike authorizations and social media campaigns. Local unions-including at Dow Jones, Insider, and Ziff Davis-go on strike or threaten to do so to support negotiations around "advances in artificial intelligence" and "a fair contract valuing our humanity" (7,48,49). Workers attempt to marshal external pressure by directly engaging the public. "Journalists at Politico and E&E News, called PEN GUILD, recently launched a campaign called 'Journalists, Not Robots' to share out the individual copy editors, newsletter writers and reporters who make the actual journalism, pressuring the company to respect human ability over bots." (13) In its negotiations with G/O Media, the Gizmodo Media Group Union "call[ed] on the public to help, taking to Twitter to urge readers not to click on stories with a 'bot ' byline" (19).
Organized labor also recognizes the limitations of "fight[ing] these battles alone in America's newsrooms" (3). They call on policymakers-state and federal, legislative and executive-to help, including by strengthening workers' bargaining power, promoting investment in labor augmenting technology, and protecting workers' copyrighted material and likenesses. The National Labor Relations Board "should act upon its authority to issue guidance" (3,15). Congress should "ensure that journalists and media workers are a core priority in any proposed legislation…to address Artificial Intelligence" (12). The discourse reveals that unions view direct worker action as limited, with lasting gains arising out of parallel efforts to work outside of and around their publications.
Across and within this diverse discourse stretching more than two years, we see patterns of problems, roles, and actions that, taken together, illustrate how organized journalistic labor both understands and is working to reshape GenAI's power.
Scholarship critical of AI's purported power and wisdom abounds. Campolo and Crawford warn that "enchanted determinism" drives "myth-making about deep learning's transcendent, superhuman capacities" (Campolo and Crawford 2020, 1). Slota and colleagues debunk fantastical promises that AI brings entirely "new modes of living, working, and community engagement" (Slota et al. 2020, 1). And Bourne (2024) This paper offers a contribution to this scholarship by investigating how one community of practice-unionized journalists-talks about GenAI as a problem, values human labor over automation, and proposes solutions to prioritize workers' control over GenAI and its power to structure their labor. Analyzing this discourse across a variety of sources and types of communication, we find patterns: publishers often start and pace GenAI experiments; workers want management trust and transparency but find both often lacking; GenAI cannot match journalists' craft and creativity; GenAI is too unreliable to replace journalistic news work; workers fear for control over their data and creative identities; and collective bargaining is a necessary but insufficient solution to the risks and harms of GenAI. Across trade press, guild statements, organizer testimonies, and contracts we find union organizers pushing hard-through public communication, solidarity appeals, collective bargaining, worker action, and management negotiation-to slow the pace of GenAI's deployment, control its use in newsrooms, curb its power over working conditions, and rebalance a labor landscape that has largely been defined by publishers' rush to reduce production costs and fuel GenAI hype.
We note, though, that some concerns appeared little in the discourse-topics that we were expecting to see more discussion of but did not. Three stand out. First, we saw very little discussion in the discourse about GenAI's ecological impacts, about the dramatic water and energy needed to sustain datasets, train models, and power interfaces. At least in the discourse that we studied, extremely few union organizers are concerned about GenAI's environmental harms (Crawford 2024). They could take a broader view of GenAI to question how publishers' embrace of these tools and infrastructures hastens climate change and harms workers, but we did not see mention of this. Ananny, M., & Karr, J. (In Press). How Media Unions Stabilize Technological Hype: Tracing organized journalism's discursive constructions of Generative Artificial Intelligence. Digital Journalism. Retrieved from osf.io/preprints/socarxiv/xdfe9 27 Second, while we saw some concern about the provenance and construction of datasetsas biased, extractive, or copyright busting-we saw little concern for the human labor building, training, and sustaining large datasets, the often ignored "ghost workers" (Gray and Suri 2019) who make AI seem automated and intelligent . While other scholars and organizers have focused on such workers (Irani and Silberman 2013), contemporary journalism union organizers do not seem to define media labor in ways that include the people who build the GenAI infrastructures used for reporting, editing, and publishing. GenAI's labor problem is largely framed as a media labor problem, not a technology labor issue.
Finally, while we saw some concern over GenAI's power to speed up the pace of journalists' work, we saw little talk about how working with GenAI might impact journalists' sense of job satisfaction, professional reward, or workplace stress. Even though some scholars are increasingly documenting the emotional tolls of journalism (Bélair-Gagnon et al. 2024;Holton et al. 2023)being publicly visible, enduring harassment, navigating precarious job markets-few media unions seem to be foregrounding broader senses of worker wellbeing or even professional joy when discussing the power of GenAI.
These three absences suggest that while media guilds have quickly developed sophisticated ways to talk about and correct GenAI's power over their news work, they have not yet prioritized GenAI's impact on the climate crisis or dataset workers, and they leave the emotional aspects of work largely unaddressed. Future research might examine whether these topics fall outside of media guilds' expertise or interests; if they were strategically de-prioritized in favor of concerns seen as urgent and tractable; or how such exclusions connect to longer histories of labor organizing that must contend with the boundaries of technological infrastructures and media work.
Most broadly, we saw a field struggling with hype. Returning to the three-pronged panics dominating contemporary journalism-economics, labor, technology-we see people (media guilds and unionized journalists) using language (stories, statements, testimonies, contracts) to interpret and control an invasive, popular technology (GenAI). Our hope is that the discursive patterns we chronicle here help scholars and practitioners alike see how journalists, owners, technologies, infrastructures, and audiences may work together to create better media systems.