In Journalism in the Age of AI, Rodrigo Zamith, Tomás Dodds, and I make the case that artificial intelligence is reshaping how we create, how we learn, and how we understand the world around us — and that journalism, the institution society depends on most to make sense of it all, is at the center of that transformation. But we also think the way journalism and AI are developing together right now isn’t working. It’s largely accelerating a broken system that has been grinding journalists and public trust down for decades.
Our book, published by Polity, argues it doesn’t have to be that way. AI can also offer journalism a rare chance to break that cycle and build something better. We draw on questions of creativity, democratic accountability, platform power, and human agency to offer readers a toolkit for reimagining journalism in the age of AI — not just as an industry, but as a profession dedicated to serving the public good.
The book engages with AI’s impacts on multiple levels, covering everything from the hands-on realities of AI-assisted news production and the shifting architecture of media power to AI’s implications for democratic life, journalism education, and the profession’s future. Below is a portion of Chapter 2, which examines how AI is already reshaping every stage of news production and asks whether these tools will reinvigorate journalism or merely accelerate its hamster wheel. This excerpt focuses on emerging applications of AI across five stages of the newsmaking process — story ideation, sourcing, verification, storytelling, and distribution — highlighting both their transformative potential and their limitations.
You can download a digital copy of Journalism in the Age of AI from our website for free starting August 6, and purchase paperback copies starting November 6 in the U.K. or January 19, 2027 in the U.S.
AI is already helping reshape journalistic practices across five stages of news production: coming up with story ideas, sourcing information, verifying content, telling stories, and distributing news.
AI and story ideation
Journalists are in the business of hunting for stories. Traditionally, this hunt involves attending public meetings and press conferences, mining records, reviewing competitors’ coverage for ideas, and cultivating sources who provide tips. During these activities, journalists assess potential stories against familiar newsworthiness criteria: Is it timely? Does it involve recognizable people? Will audiences care enough to share it? And, crucially, they must manage risk — newsrooms need reliable output, so some stories must be “sure things.”
This practical reality produces templates that make news production efficient and predictable. Journalists learn to recognize certain patterns — the political scandal, the human-interest feature, the controversial quote — and actively seek ideas that can be fitted into the proven molds. Templates offer efficiency, but they also create tension. Journalism, after all, is supposed to be a creative enterprise. Editors and reporters constantly seek to add something extra to their stories: a fresh angle on a familiar event, an unexpected voice, a compelling visual treatment. This drive stems from both intrinsic motivation (breaking the monotony) and market pressure (standing out in a crowded information landscape).
Fitting templates and sparking thoughts: Some newsrooms already deploy AI to identify stories that fit established templates. These systems scan competitors’ articles, press releases, social media feeds, and search trends to bring story ideas to light, essentially automating the hunt for low-risk, newsworthy topics. Sisi Wei, chief impact officer at CalMatters, says that “AI can be an incredible tool for good” and that her organization uses it to keep up with government officials. The technology, Wei writes, “tracks pretty much everything we can track about California’s state legislators, parses that fire hose of data, and then generates story ideas for our reporters.” Some AI-generated tips, she notes, “would have taken a data-savvy political reporter weeks, if not months, to find on their own.” The resulting stories have already influenced legislative action.
But AI doesn’t have to be restricted to the familiar; it can spur creativity. Journalists increasingly use large language models (LLMs) as sounding boards: tools for developing nascent ideas, working through story angles, and overcoming creative blocks. Meghan Murphy, director of the Online News Association’s AI in Journalism Initiative, describes AI as “an interesting thought starter.” While it often “spits back what I already know or what I wrote anyway,” Murphy finds it sometimes highlights angles worth exploring further. The key, as data journalist Paul Bradshaw suggests, lies in well-constructed prompts that describe the situation, specify constraints, and provide examples.
A serendipity engine: As AI becomes further embedded in browsers, operating systems, and information gateways, it is poised to reshape how journalists encounter ideas without requiring direct prompting. AI agents could be configured to interrupt journalists’ information consumption routines, functioning as a “second brain” that highlights unexpected connections or reveals valuable information outside a journalist’s typical information-seeking patterns.
This vision aligns with cognitive models showing that creativity often emerges from collisions between seemingly unrelated information. Human creativity requires both access to diverse inputs and the capacity to recognize non-obvious connections. AI systems designed as serendipity engines could deliberately introduce productive friction into routine behaviors, helping journalists overcome cognitive fixations and confirmation biases. Such systems would blend structured support with strategic surprise — providing scaffolding for creative work while systematically introducing the unexpected.
Any AI system that generates story ideas must be trained on what counts as newsworthy, though, which requires translating editorial values into code. To maximize creative potential, these systems would need to point journalists toward novel, worthwhile paths rather than merely replicate their existing judgment. This is no easy task given the pattern-focused design of most current AI systems.
Journalistic creativity involves ethical and empathetic judgments about which stories matter and how to tell them in ways that respect and resonate with human experience. The future, in our view, likely lies neither in pure human ideation nor in fully automated story generation, but in a hybrid approach where human intuition works alongside both visible and invisible digital assistants to collectively decide what stories need telling and how to best tell them.
AI and sourcing
Journalists depend on sources to provide context, interpretation, access, and insider knowledge for stories they cannot directly observe. These sources — ranging from academic experts and government officials to advocacy organizations and ordinary citizens — help journalists understand complex issues, gain access to restricted spaces, and illustrate broader social concerns through personal narratives. But sourcing practices have never been neutral: They reflect deep structural inequalities that AI both promises to address and threatens to reinforce.
Two key factors shape which voices appear in news coverage: authority and availability. Individuals and organizations within established power structures — government representatives, corporate leaders, institutional experts — are treated as especially newsworthy because their decisions shape public life and their institutional standing signals legitimacy. But even diligent and well-intentioned journalists face practical constraints. Time pressures push them toward sources who are easily reached and responsive, such as the contacts already in their address books and the spokespeople who return calls promptly.
These structural patterns are reinforced by interpersonal dynamics. Journalists and sources are subject to homophily: the human tendency to associate with people similar to ourselves. For example, journalists are more likely to contact, and receive cooperation from, sources who share their gender and race. As a result, female experts and ethnic minority community members have historically been underrepresented in coverage, and their perspectives have consequently been marginalized in important public debates.
A sourcing auditor: AI systems have been deployed to identify systematic sourcing biases and suggest alternatives beyond journalists’ existing networks. Several organizations have experimented with automated tools that review draft stories and flag imbalanced sourcing patterns. UNHEARD, a project led by investigative journalist Bette Dam, “aims to help news organizations reveal potentially overlooked narratives by using AI to audit who is quoted in their articles.” The ultimate goal is to allow “news organizations to analyze reporting, identify potential imbalances, and suggest alternative sets of sources to improve accuracy as a story unfolds.”
This isn’t a pipe dream. Research demonstrates that automated systems can identify sources with high accuracy, enabling them to predict when articles would benefit from additional perspectives and even recommend specific individuals to expand representation.
However, AI-powered auditing doesn’t necessarily eliminate sourcing hierarchies. Journalists’ disproportionate reliance on government sources doesn’t stem from limited imagination but from the perceived authority those voices confer. And scientific sources tend to be underutilized not because journalists forget about them, but because academic timelines are often incompatible with journalistic deadlines — not to mention scientists’ tendency toward cautious, jargon-laden responses. While AI systems can help journalists more quickly identify more diverse voices — an improvement, no doubt — they cannot resolve the deeper professional imperatives.
An interview assistant or interviewee: AI is also helping journalists conduct preliminary research, learn about unfamiliar topics, and generate interview questions tailored to specific sources and story angles. While these questions typically need refinement, they can help journalists break away from formulaic templates, provided that they use them with such intent.
In extraordinary cases, AI has even been used to automate elements of the interview process itself. For example, the Swedish tech company United Robots offers tools that, with limited human involvement, analyze sports recaps, generate questions for post-game interviews with coaches, send those questions via text messages, and insert the responses into match reports.
But interviews are rarely mechanical transactions. As Ruby Voge, opinion co-editor at Boston University’s student newspaper The Daily Free Press, contends: “ChatGPT can’t interpret the nuance of a fleeting facial expression or a subtle vocal inflection. It can’t wait outside the mayor’s office for an early-morning interview or embed itself into the life of a profile subject. It can’t interview disaster survivors with empathy and understanding or hold world leaders accountable with skepticism and perseverance.”
Even more complicated is when an AI agent becomes the source itself. In April 2023, the German magazine Die Aktuelle published what it claimed to be the “first interview” with Michael Schumacher, the Formula 1 legend who suffered a near-fatal brain injury in 2013 and hadn’t spoken publicly since. But the interview wasn’t real; it was an AI-generated confabulation of what the magazine imagined he would say a decade after his accident. There was immediate public backlash, leading the magazine’s publisher to fire its editor, apologize for a “tasteless and misleading article,” and reach a compensatory settlement with the Schumacher family.
Other incidents of AI stand-ins have followed, raising concerns about authenticity, consent, and the blurring of lines — and whether such practices constitute legitimate journalism or exploitation of emotions like grief. While AI can facilitate novel forms of connection, and is being used increasingly by the “digital afterlife industry” to simulate deceased loved ones through so-called “deadbots,” it also risks commodifying human tragedy, desensitizing people, and trivializing death.
A rapid transcription tool: Audio transcription represents an area where AI has already become indispensable. Journalists routinely use AI to automatically transcribe recorded interviews, identify noteworthy segments, make audio searchable, and cross-reference sources’ current statements with past remarks. This seemingly basic automation saves hours of work weekly — though it can be subject to mistakes, such as when popular agents like Whisper occasionally “hallucinate” and transcribe something very different than what was said. Similarly, AI can analyze large volumes of digitized documents, extracting key information and summarizing findings. The detection doesn’t need to be perfect — if an AI assistant directs a journalist toward a smaller set of likely relevant documents, it has already proven valuable.
Michigan Public Radio in the United States, for example, created an AI-powered tool called “Minutes” that automatically scrapes and transcribes videos of city council meetings, allowing reporters to monitor public proceedings they cannot attend in person. Reporter Dustin Dwyer finds the tool to be both a time-saver and a way to diversify his reporting: “I use it to skim transcripts to get a general idea of the meeting, without having to spend hours sitting through the whole thing. And the new keyword alert system allows me to track issues and find sources in communities that I normally don’t cover.” The tool enables the station to continue performing monitorial journalism even as its resources shrink.
These examples show that AI-powered tools can improve sourcing practices in multiple ways, but journalists need to exercise care. Current tools remain best suited to identifying readily accessible, digitally present sources rather than marginalized voices who may lack online profiles or institutional affiliations, which can reinforce rather than challenge established hierarchies. Furthermore, AI-generated interview questions can inadvertently perpetuate problematic assumptions about whose perspectives matter and what questions are worth asking because of how their models were trained. Perhaps most significantly, AI systems struggle to replicate the human capacity for developing rapport with sources, interpreting nonverbal cues, or approaching vulnerable subjects with the empathy their stories demand — the very kind of high-touch interactions that need to be fostered for journalists to rebuild trust in communities. A growing number of people, though, report establishing relationships with chatbots that they perceive to be meaningful, suggesting this may yet be possible. Nevertheless, the path of hybridity again strikes us not only as the most fruitful but also the most prudent given the state of the technology.
AI, verification, and fact-checking
Accuracy and truth-telling have become more than professional values in journalism. As audiences drown in an ocean of information sources, verification and fact-checking practices serve as both credibility markers and a way for outlets to stand out. News organizations are not just competing for clicks; they are competing for audience trust in the hope that it translates into repeat visits and revenue.
A fact-checking boom began around 2016, when Donald Trump’s election and the Brexit referendum ushered in what some called a “post-truth” era. For example, fact-checkers worldwide created ClaimReview, a standardized tagging system launched in 2015 that helps search engines and social platforms highlight verified information in places like their search results and news feeds. Google initially embraced the initiative, which now includes hundreds of thousands of fact-checks from organizations like Agence France-Presse, The Washington Post, and Rappler. Fact-check snippets on Google products were viewed over 120 million times in the first half of 2024 in the European Union alone.
This matters because fact-checking works. Research consistently shows that exposure to fact-checks reduces belief in misinformation, with corrective effects persisting over time. But just as fact-checking gained prominence, it became exponentially harder.
Traditional verification and fact-checking: Three forces have converged to challenge traditional fact-checking. First, the democratization of publishing tools empowered anyone to create professional-looking content and distribute claims widely online, producing a surge of misleading and fabricated information.
Second, budget-strapped newsrooms began facing even harder choices between fact-checking erroneous claims and pursuing new stories, and as dedicated fact-checkers and copy editors were increasingly laid off, those who remained were further overextended by being forced to take on the additional verification and reviewing tasks.
Third, fact-checking itself became politicized, prompting major platforms, including Google and Meta, to retreat from verification partnerships. Furthermore, studies have found that initial falsehoods circulate far more widely than the fact-checks that follow, making fact-checks only partly successful most of the time.
When generative AI arrived, it turbo-charged the challenge. The speed and scale at which AI can produce and spread false information have rendered real-time, human-scale verification nearly impossible. As investigative researcher Henk van Ess observes: “Traditional fact-checking takes hours or days. AI misinformation generation takes minutes.”
Professional fact-checking typically follows several stages. Journalists first select claims worth verifying within their story or by scanning social media, news reports, and online comments to identify statements being made elsewhere that warrant verification. Selection criteria include the claim’s verifiability, potential impact, and whether multiple sources raise the same concern.
Next comes assessment and contextualization, which recognizes that technically accurate statements can mislead without proper context. Journalists trace claims to their origins, search for primary documents, review previous reporting, identify authoritative data, and assemble evidence.
Finally, they decide whether to include, refute, omit, or contextualize the claim. Dedicated fact-checking organizations typically conduct internal deliberations before publishing separate fact-check stories, followed by editorial review for accuracy, clarity, and neutrality.
Automating and augmenting verification: Recognizing they can’t outpace AI, researchers and journalists have developed tools to automate some parts of that process while supporting others. These systems detect claims in speeches, social media posts, and articles; identify synthetic media like deepfakes; and spot disinformation campaigns through pattern analysis.
Real-world implementations reveal both promise and pragmatism. Full Fact, Maldita.es, and the European Fact-Checking Standards Network built a multilingual system monitoring short-form video from YouTube Shorts, TikTok, and Instagram Reels. It extracts claims using speech-to-text and text recognition and then clusters them into narratives, helping journalists “prebunk” false information before it spreads further.
“AI tools help us find, filter, and follow claims at scale,” reflects Full Fact senior product manager Kate Wilkinson. “Human fact checkers provide the expertise, judgment and accountability. Together, they give us a fighting chance to make public debate more accurate and fair.”
Full Fact’s hybrid approach acknowledges another challenge: truth rarely fits binary categories. Journalists routinely confront ambiguous, unverifiable claims or those dependent on future events. More critically, many information sources needed for verification don’t exist digitally and thus don’t appear in AI training data.
A comparative study of fact-checking in Brazil and Germany found that journalists regularly encounter cases where reputable sources of information cannot be located or inquiries go unanswered, leaving claims unresolved despite the journalists’ rigorous efforts. AI excels at verifying simple, measurable claims involving publicly documented information or previously fact-checked claims. It cannot navigate the messy reality of missing documents and reluctant sources.
Moreover, the propensity of LLMs to “hallucinate,” or to unintentionally produce incorrect or fabricated information and present it as fact, compounds the problem. Even when AI provides technically accurate information, it may lack contextual grounding for proper interpretation. Journalists must therefore verify what their verification tools tell them — a frustrating irony. Consequently, some journalists remain skeptical about AI’s efficiency gains. A study of 22 fact-checkers found existing tools often failed to fit their workflows and produced erroneous or irrelevant results that increased rather than reduced their workload.
Risks to credibility: News organizations also face a new threat that goes beyond their own use of AI: having their credibility damaged by AI systems they don’t control. In December 2024, the BBC publicly criticized Apple Intelligence after it created a misleading notification that falsely implied BBC News had reported that Luigi Mangione — the man accused of murdering healthcare executive Brian Thompson — had shot himself. The BBC never published this claim, yet its credibility suffered among those who saw the error. Worse, the BBC had no way to know who was exposed to the erroneous notification or on what scale. While Apple paused that specific feature, it merely added disclaimers about potential inaccuracies elsewhere in its operating system.
This represents a troubling inversion: rather than using AI to combat misinformation, news organizations find their own brands weaponized or placed at risk of being tarnished by AI-generated falsehoods.
The lesson emerging from newsrooms is that while AI can help journalists rapidly process certain types of information and spotlight contentious claims, the broad practice of verification still requires a degree of judgment, contextual awareness, and accountability that algorithms currently appear to be incapable of exercising. The implication is clear: Human fact-checkers will need to learn to work effectively with AI technologies to defend truth in an environment where information, misinformation, and disinformation all operate at machine speed.
AI and storytelling
Journalists use storytelling as both craft and strategy, carefully selecting, sequencing, and framing information to engage with audiences while maintaining professional standards. Compared to other forms of storytelling, journalism tends to be more formulaic, especially in genres like breaking news.
For example, the inverted pyramid, which front-loads the most important information, has dominated breaking news in Western journalism for over a century. Of course, other structures also exist — the “martini glass” blends summary leads with chronological narration and the “kebab” bookends analysis with humanizing anecdotes — and they serve specific purposes, from accommodating time-pressed readers to adding emotional resonance to feature stories.
Yet even though these conventions provide consistency and efficiency, critics have long derided news writing as overly rigid. In recent years, facing fierce competition from digital upstarts and social media influencers who favor formats like bullet points, Q&As, and visual illustrations, traditional news organizations have become more open-minded about storytelling structures and styles. Indeed, research suggests that audiences increasingly gravitate toward emotionally resonant content that breaks from formulaic norms. Now, AI is accelerating this evolution — in ways that are sometimes promising and other times disastrous.
Early automation of newswriting: One of the more visible early applications of AI in newsrooms — even though some observers objected to calling it that — was the automation of newswriting in certain genres, including finance and sports. In 2014, the Associated Press started automating the production of thousands of stories about corporate earnings reports and sports recaps by extracting information from databases or standardized reports, applying simple logic to determine newsworthiness, and inserting the information into pre-written templates. Proponents celebrated this as liberation from drudgery — few journalists relish writing yet another earnings brief, after all.
But critics charged that those automated stories were remarkably dull and risked alienating audiences, who would scroll past or, worse, develop negative associations with the news organization. This wasn’t merely content with little value; it potentially carried negative value by crowding out higher-quality content and reinforcing perceptions of journalism as increasingly catering to the lowest common denominator.
New possibilities with generative AI: The emergence of generative AI as begun to address some limitations of early automation while introducing fresh concerns. It is now routinely used to generate multiple headline variants for editors to choose from, create useful article summaries that are appended to stories, and personalize content in increasingly sophisticated ways, such as by localizing data-driven stories more naturally to make them more relevant and interesting to specific communities. These applications streamline workflows and free editorial staff to focus on higher-value tasks.
GenAI also enables some fundamentally novel storytelling opportunities, such as the use of chatbots to create individualized dialogic news experiences. For example, Time magazine launched the Time AI Agent in November 2025, which integrates language understanding, voice synthesis, translation, and search capabilities. Readers can ask questions, request summaries, generate audio versions of stories, and translate reports — all through a single interface that can, for instance, produce an audio summary of recent interviews with world leaders. Jason Droege, the CEO of Scale AI, which powers the tool, described it as “a blueprint for how publishers can use AI agents to create a more meaningful relationship between their audience and their content.”
The emergence of Time’s AI Agent — along with similar tools like The Washington Post’s Ask the Post AI and The Financial Times’s Ask FT, both launched with fanfare in 2024 — represents the materialization of a long-imagined future: one in which people get their news from conversations with AI-powered agents rather than discrete articles. This marks a fundamental reimagining of the journalist–audience relationship, simultaneously bringing audiences closer to a synthetic intermediary while distancing them from the human journalists whose work feeds these systems. Some tech companies are pursuing similar visions, potentially sidelining news organizations from developing direct relationships with their audiences altogether.
Prototyping new ideas: Another key development is the emergence of vibe coding: an informal approach to software programming where a person, coder or not, uses plain language to describe the “vibe” of what they want to create, and the AI tool mostly does the rest in generating the code. The human focuses on the goal, and the AI handles the line-by-line writing. This approach, further discussed in Chapter 6, has empowered journalists to prototype nascent ideas, experiment with specialized back-end tools, and create public-facing products without having to rely on technical specialists.
Joe Amditis, associate director of operations at the Center for Cooperative Media, exemplifies this shift. Despite describing himself as not “a real software developer,” he has used LLMs to build dashboards, create interactive features, process data dumps, and develop web tools for local news organizations across New Jersey. According to Amditis, “the most valuable skills for vibe coding are the skills of a good project manager,” such as identifying requirements, clearly articulating what you want, providing direct and clear feedback, and organizing ideas and documentation. He adds: “These skills are now more immediately valuable than the skills of a traditional software developer. You’re a manager now, delegating the coding to an LLM.”
This isn’t just making some work easier. It’s a form of upskilling that allows journalists without previous technical training to do new things they otherwise would not have been able to, such as developing interactive graphics, data visualization tools, and small applications that enhance their storytelling capabilities.
When automation goes wrong: The promise of automated storytelling comes with serious risks, particularly when organizations move too quickly or fail to maintain adequate oversight. AI’s tendency toward hallucination becomes even more vexing when false content reaches audiences.
In late 2023, Sports Illustrated faced a firestorm when other journalists discovered the magazine had published articles under AI-generated author names with fabricated biographies. Around the same time, then-CNET Editor Connie Guglielmo acknowledged that 77 machine-generated stories appeared on its website, several requiring corrections. The scandal damaged the publications’ credibility and cost several executives their jobs.
Smaller organizations have also stumbled. In November 2025, reporters at Suncoast Searchlight, a Florida-based nonprofit news organization, sent a letter to their board of directors after discovering their editor-in-chief’s undisclosed use of AI editing tools — which included instances where the AI inserted hallucinated quotes into reporters’ drafts. Some staff members were unaware that AI was being deployed on their work at all, raising important questions about transparency, editorial control, and ethical boundaries.These uses of AI illuminate the emerging complications around authorship, accountability, and transparency — questions that grow more vexing as AI becomes so integrated into workflows that some people no longer consider it worth mentioning. While the need for disclosure is often underscored in debates about AI-generated news, the thresholds become murkier as AI permeates routine support activities and sometimes even directly modifies journalists’ work without their knowledge or consent. The line between tool and author therefore continues to blur, challenging foundational assumptions about who is responsible when storytelling goes wrong.
AI and distribution
Media distribution fundamentally shapes who can access journalism and under what conditions. For decades, news organizations had considerable direct control over their distribution pathways. However, the rise of digital intermediaries — search engines, social media platforms, news aggregators — has dramatically altered this landscape. Today, algorithmic systems increasingly determine which stories reach which audiences and provide the marketplace for monetizing them, forcing news organizations to adapt their editorial practices to accommodate the demands of algorithmic distribution. At the same time, AI empowers news organizations to rapidly remix their own content for multiple platforms and to expand accessibility.
Optimized distribution: Media publishers have traditionally owned large segments of their distribution channels. They often owned the printing presses and the delivery trucks. They owned the cameras and microphones to record the news, as well as the antennas to beam their stories across the country. With the advent of the internet, newsrooms also had significant control over their own websites. As social media platforms started to gain popularity in the 2000s, though, media publishers gradually became more dependent on platform distribution by tech companies like Google and Facebook, including their social media feeds, search engine results, and news aggregators.
Thus, news organizations have had to manage their own proprietary distribution channels while optimizing their content for algorithmic visibility. Some newsrooms even required reporters to build personal social media followings, effectively transforming individual journalists into distribution channels. Meanwhile, audiences increasingly encountered news through algorithmic feeds, where platform logic, not editorial judgment, dictated the final presentation.
Several news organizations have adapted by integrating AI into their distribution through two main strategies. Algorithmic optimization uses AI to decode platforms’ shifting preferences, advising journalists on ideal formats, optimal publishing times, and strategic keywords to maximize visibility and reach across platforms. Targeted delivery harnesses AI for precision audience segmentation, enabling organizations to serve personalized content to specific demographic or interest groups.
BBC News exemplifies this approach. In March 2025, the broadcaster announced a new department dedicated to AI-powered personalization. As then-CEO Deborah Turness wrote to staff: “We must become ruthlessly focused on understanding our audience needs, on delivering the kind of journalism and content they want, in the places they want it, designed and produced in the shape that they enjoy it.” The path forward, she argued, requires deploying “AI to support, enable and accelerate our innovation and growth.”
It isn’t just major organizations doing this, either. The Rural News Network, a collective of more than 500 local news nonprofits across the United States, created Text RURAL to solve a distribution challenge. With member organizations producing nearly 5,000 stories monthly, the AI-powered tool automatically selects, aggregates, and summarizes the most relevant content for individual users based on their location and interests. It then delivers personalized weekly news roundups via text message. It’s a practical solution that makes local journalism more accessible to rural communities underserved by traditional distribution channels. However, it is an automated solution that bears risk, as even major publishers have discovered in their experiments with AI summaries.
Automating content adaptation: News organizations are also using generative AI for automated adaptation, or the rapid generation of derivative products tailored to different platforms and accessibility needs. AI systems can automatically produce social posts, video summaries, audio versions, and translations from content originally created by human journalists. While these derivatives are typically reviewed before publication, they nevertheless reduce the time, cost, and labor required to meet audiences across multiple platforms while expanding accessibility.
Dow Jones Newswires launched a custom AI language service in 2024 targeting this exact opportunity. The system produces what it describes as “fluent” translations of hundreds of English-language stories daily into Japanese, Korean, and French. “It gives us an ability to very quickly, almost instantaneously, translate our very rich existing English-language content into a language that will attract other audiences,” observes Chip Cummins, The Wall Street Journal’s chief Newswires editor. The service reaches professionals “who prefer to read their business news in their native language or who are suddenly tapping into the US market as a place where they want to invest, or their clients are investing.”Similarly, the U.S. television station WCPO 9 uses AI to automatically reformat content for different platforms. “AI helps us take horizontal video storytelling you might see on our broadcast or YouTube and convert it to vertical on Instagram or TikTok,” notes senior manager PJ O’Keefe, “so that, again, we can make as big of an impact as possible, on as many platforms as possible.”
Dow Jones’ approach includes transparency measures: each translated article carries a warning about AI translation and links to the original story. However, as WBEZ reporter Araceli Gómez-Aldana observes, current models still struggle with the “accuracy of the translation for news stories, cultural understanding, translating quotes for tone and understanding — not only a literal word-by-word translation — and adapting to various dialects and literacy levels.” In other words, there are still risks associated with this kind of functionality.
Despite these powerful capabilities, what AI cannot do is liberate news organizations from platform dependencies. No matter how sophisticated their AI tools become, newsrooms cannot alter the ethical, economic, or political decisions made by platform owners regarding content moderation, algorithmic demotion of news, or sudden rule changes. Platforms can unilaterally reshape their algorithms — as Meta has done repeatedly in prioritizing and then de-prioritizing news content on Facebook — leaving news organizations scrambling to adapt their carefully crafted AI-optimized strategies to new, often opaque, rules. This structural power imbalance constrains AI’s potential to transform news distribution in ways that truly benefit journalism.


