AI quietly redefines journalism’s core human tasks
The most dangerous form of obsolescence is the one you never notice because it feels like help. When a reporter in Philadelphia uses an AI tool to monitor public meetings across dozens of municipalities, the promise is efficiency: no more sitting through three-hour school board hearings to catch a single newsworthy moment. The machine watches, summarizes, and ranks developments by newsworthiness. The journalist gains time. But this is not a trade. It is a transfer. The tool does not free the reporter for deeper work — it redefines what deeper work means, moving the journalist further from the raw, unmediated experience of the story.
The Mechanism That Misses the Problem
The Philadelphia Inquirer’s Scribe tool, built with OpenAI technology, turns public meeting transcripts into concise, categorized summaries and scores developments using a newsworthiness framework created by reporters and editors. Philadelphia Inquirer This sounds like a sensible division of labor: the machine handles volume, the human handles judgment. But consider what disappears in that transaction. The journalist no longer hears the hesitation in a school board member’s voice when a controversial topic arises. They miss the body language of a parent testifying about their child’s education. They lose the context of side conversations in the hallway after the meeting adjourns. The AI captures what was said but not what was almost said, what was avoided, what was communicated through silence. Every newsroom that adopts this technology is betting that these intangible signals are replaceable. History suggests otherwise.
The Associated Press uses OpenAI technology to help journalists scan overnight news and podcasts for reportable developments, support image and video verification, and turn thousands of Supreme Court filings into searchable, structured information. AP also applies the technology to surface potential stories in government datasets and recommend member content for editors to consider sharing. Each of these applications removes a step that once required human attention, human curiosity, human serendipity. The journalist who used to scan the overnight wire with coffee in hand, catching an odd detail that would lead to a story three weeks later, now receives a pre-digested list of reportable developments. The tool finds what it is programmed to find. It misses what it cannot categorize.
The Infrastructure That Became Invisible
The American Journalism Project’s Product and AI Studio, supported by OpenAI, helps local news organizations leverage AI to strengthen their work and sustainability. American Journalism Project Centro de Periodismo Investigativo in Puerto Rico built custom GPTs that save time for its small team as they draft and translate donor communications. This innovation traveled: Enlace Latino North Carolina adapted the translation workflow to launch its first English-language newsletter, and Boyle Heights Beat used the same approach for real-time bilingual coverage during the Los Angeles fires. Enlace Latino North Carolina Boyle Heights Beat These are success stories by any measurable standard. More output, more reach, more sustainability. But notice what is never measured: the journalist who used to write that donor letter by hand, choosing each word carefully because the relationship mattered. The editor who personally translated the newsletter, catching nuances that a model would flatten. These losses are invisible because they are distributed across hundreds of small decisions, none of which seem significant in isolation.
At POLITICO, AI helps journalists analyze large volumes of public documents and data for deeper, more timely original reporting. Commercial teams use AI to tailor client sales experiences with richer data on POLITICO’s offerings. The tool serves both editorial and business functions, embedding itself in the organization’s DNA. But the journalist who once read those documents cover to cover, developing an intimate knowledge of the subject that would inform every subsequent story, now receives a summary. The analysis is faster. The depth is shallower. This is not a bug; it is the feature that makes the tool attractive. Speed always comes at a cost, but the cost is paid in a currency that quarterly reports do not track.
The Sentence Left Unspoken
Axios has built custom GPTs for tasks ranging from understanding internal policies to crafting open-records requests. The FOIA Refiner GPT helps reporters craft specific, efficient requests less likely to be denied or delayed. The Axiomizer GPT reviews story copy to suggest sharper headlines and clearer writing. These tools are designed to make journalists better at their jobs by automating the tedious parts. But there is a structural irony here that goes unmentioned: the skills these tools automate — the ability to write a precise FOIA request, the instinct for a sharp headline — are precisely the skills that distinguished excellent journalists from adequate ones. When everyone has access to the same optimization tools, the competitive advantage shifts to something else. But what? The answer is uncomfortable: the only remaining differentiator is access. Who has the sources, the relationships, the institutional memory that cannot be extracted and automated?
Le Monde launched its English-language edition in 2022 and in 2025 incorporated its translation stylebook into ChatGPT models to accelerate publication. Le Monde The editorial teams freed up time from journalists, enabling them to focus on core reporting and analysis. The logic is impeccable: translation is mechanical, analysis is creative. But this binary is false. The act of translation forces the translator to engage deeply with the original text, to understand not just what was said but how it was said, why this word was chosen over that one. The journalist who translates their own work into another language discovers things about their own writing they would never notice otherwise. That discovery is lost when the machine handles the translation, even if the output is technically correct. The efficiency gain is real. The loss is real too, but it cannot be measured, so it is not counted.
The Unspoken Sentence
PRISA Media uses OpenAI technology across a growing set of editorial tools: a trend-and-information tracker, country-specific audio news briefings, and Vera, a conversational assistant answering EL PAIS subscribers’ questions. OpenAI models also automate content vectorization, article translation, and image rights attribution. PRISA Media. Each application is practical, sensible, and individually defensible. Collectively, they represent a fundamental shift in what a newsroom does. The journalist is no longer the primary producer of content; they are the supervisor of a content production system. The machine drafts, translates, summarizes, and categorizes. The human reviews, approves, and occasionally intervenes. This is not journalism augmented by AI. This is journalism replaced by AI, with the human retained as a liability shield and quality control checkpoint.
The Daily Beast’s Data Scouts suite of OpenAI-powered agents helps newsroom and business teams move from information to action. Rather than summarizing data, Data Scouts identify opportunities, explain what is happening across the business, and recommend practical next steps. Most interactions happen in Slack, where teams already collaborate. The tool brings insights into existing conversations instead of requiring adoption of another dashboard. This is elegant design, but it obscures a deeper question: who decides what counts as an opportunity? The tool identifies patterns based on historical data. It recommends actions based on past successes. It optimizes for what has worked before. But journalism’s greatest value has always been its ability to see what has never worked before, to spot the story that does not fit any existing pattern. The tool cannot do this. It will not try. And because its recommendations are so useful for the routine decisions, the organization will gradually stop asking the non-routine questions.
What Remains
Condé Nast’s Bon Appétit launched an AI-powered Test Kitchen Assistant that combines its editorial archive with OpenAI’s models, allowing home cooks to ask questions about recipes, cooking techniques, and ingredient substitutions in real time. The Assistant lets readers experience editor-approved culinary content in more interactive ways. This is not journalism in any traditional sense. It is a service built on journalistic content, repackaged and made interactive. The original reporting — the recipe testing, the technique development, the expert interviews — still happens, but it becomes raw material for a machine rather than the finished product. The reader interacts with the AI, not with the journalist. The relationship is mediated by the model. The trust that readers placed in Bon Appétit’s editors is transferred to the AI, which presents itself as a neutral conduit but is anything but.
The US intelligence community’s embrace of generative AI, as reported by US News, mirrors journalism’s trajectory. Intelligence agencies describe their adoption as “at once wary and urgent” — the same tension that news organizations feel. They know the technology carries risks: hallucination, bias, security vulnerabilities. They also know that falling behind means losing relevance. The parallel is instructive. Intelligence analysts, like journalists, once prided themselves on their ability to read between the lines, to spot the detail that everyone else missed, to connect dots that seemed unrelated. AI tools promise to do this faster and at greater scale. But they do it differently. They connect dots that have already been identified as dots. They find patterns that have already been labeled as patterns. They miss the signal that no one has yet recognized as a signal. The agencies know this. They proceed anyway, because the alternative — not using AI while adversaries do — seems worse.
The Sentence Left Unspoken
Every newsroom that adopts these tools makes the same calculation: the risk of being replaced by AI is less immediate than the risk of being replaced by a competitor who uses AI. This is rational. It is also tragic. The technology does not make journalists obsolete by outperforming them. It makes them obsolete by redefining the job until the human contribution is marginal, then eliminable. The reporter who monitors public meetings through AI summaries is not replaced in one dramatic moment. They are replaced gradually, meeting by meeting, summary by summary, until the organization realizes it does not need a human to interpret the summaries because the summaries are designed to be self-explanatory. The journalist becomes a luxury the organization can no longer afford.
The sentence left unspoken in every newsroom strategy meeting, every partnership announcement, every case study about AI in journalism, is this: we are building the infrastructure that will make us unnecessary. Not today. Not tomorrow. But the trajectory is clear, and we are accelerating it with every efficiency gain, every automation, every tool that replaces a human judgment with a model’s output. We tell ourselves that AI handles the routine so humans can focus on the important. But we never ask who decides what is routine. We never ask whether the routine tasks we automate are actually the training ground for the important judgments we claim to preserve. The journalist who never learns to write a FOIA request will never develop the instinct for what documents might contain. The reporter who never sits through a school board meeting will never learn to read a room. The editor who never translates their own work will never feel the weight of a word choice.
The tools are here. They are useful. They are also dangerous in ways that no benchmark can measure and no case study can capture. The danger is not that AI will replace journalists tomorrow. The danger is that journalism will be redefined, piece by piece, until the thing that remains bears no resemblance to the thing that was lost. And by the time anyone notices, the infrastructure will be invisible — not because it works perfectly, but because no one remembers what existed before it.
Sources
2. American Journalism Project
3. Enlace Latino North Carolina
5. Le Monde
6. PRISA Media