will AI replace journalists?

Twenty-six of the thirty tasks in your job have zero measurable AI penetration right now. That's not a rounding error. Tasks like developing sources, arranging interviews, and investigating breaking news require physical presence, trust, and the kind of judgment that takes years to build. An AI can't cold-call a city council member who owes you a favour. It can't sit in a community meeting and read the room well enough to know which grievance is the real story.

Source relationships are the clearest example. According to O*NET task data, establishing and maintaining relationships with credible sources is one of the most time-intensive parts of the job, and it's irreducible. A source who talks to you won't talk to a chatbot. That relationship is personal. It's built on coffee, on follow-through, on being the person who called back when you said you would. No tool replicates that.

Investigative work is the same story. When you're digging into a disaster, a crime, or a company's financials, you're doing something no language model does reliably: you're verifying. You're cross-referencing documents, spotting inconsistencies, asking follow-up questions that only make sense given what you just heard in person. The Anthropic Economic Index places information-gathering and first-hand observation tasks among the lowest AI-exposure categories in the profession, and that tracks with what journalists actually report. The work that matters most is still yours.

The biggest problem with AI in journalism is that it sounds confident when it's wrong. Language models generate plausible text. They don't generate verified facts. If you ask GPT-4 to summarise a news story, it might get the names right, the timeline wrong, and present the whole thing in a tone that reads as authoritative. That's dangerous in a profession where a single factual error can end your credibility with a source, a publication, or a court.

There's also the accountability gap. When a journalist publishes something wrong, there's a correction, a byline, a paper trail, and sometimes a lawsuit. When an AI gets something wrong, there's no one to hold responsible. News organisations that have tried to run AI-generated articles without disclosure, like Sports Illustrated in 2023, found themselves facing reader backlash and staff revolts. The trust problem is real, and it goes both ways: readers are right to be sceptical.

AI also can't access anything that isn't already in text form. It can't attend the off-the-record briefing. It can't notice that the mayor's press secretary looked nervous when you asked a specific question. It can't follow a tip down a hallway. The parts of journalism that produce the most important stories are precisely the parts that require you to be physically present and paying attention.

The tasks AI actually handles today are the ones that were never the interesting part of the job anyway. Tools like Otter.ai and Sonix can transcribe a 45-minute interview in under three minutes, with speaker labels and searchable timestamps. That used to take an hour of typing. It's genuinely useful, and it works.

On the research and organisation side, tools like Perplexity AI can pull together background information on a topic quickly, surfacing public records, prior coverage, and data points that would have taken a database search and twenty browser tabs. It's faster than doing it manually. But you still need to verify everything it gives you before it touches your draft. Treat it as a starting point, not a source. Similarly, tools like Jasper and even standard GPT-4 prompting can help you restructure a draft for different formats, such as taking a long feature and pulling the key facts into a short web summary. That's the kind of reformatting work that can eat an afternoon and now takes ten minutes.

For broadcast journalists specifically, tools like Descript let you edit audio and video by editing a text transcript, which cuts post-production time significantly. None of this changes what you report. It changes how fast you can process and package what you've already found. The Anthropic Economic Index puts AI's actual penetration into journalistic tasks at around 21%, and most of that sits in these back-end production tasks, not in the reporting itself.

The clearest shift in the daily rhythm is on the back end of a story. Transcription used to be the tax you paid after a good interview. Now it's done before you've finished your coffee. You get to the meaning of what someone said faster, which means you spend more time on the next interview and less time replaying a recording with your headphones on.

What hasn't changed: the front end of the job is identical. Making calls, getting turned down, finding a different angle in, attending the briefing, waiting outside the courthouse. All of that is the same as it was ten years ago. The AI tools covered above don't touch any of it. Your day still starts with who you need to talk to, not what software you need to open.

What you spend more time on now is judgment. Because the mechanical tasks compress, you're left with more decisions about what the story actually is, which angle serves the reader, and whether the sourcing is solid enough to publish. That's the part of the job that's harder than people realise. It doesn't show up in a word count, but it's where the professional value sits.

The 3.9% decline BLS projects through 2034 is real, but it's not primarily an AI story. It's a business model story. Print advertising revenue collapsed over the past two decades, and local newsrooms have been the casualty. The American Press Institute estimates that more than 2,500 local newspapers have closed since 2005. The journalists who lost those jobs didn't lose them to a language model. They lost them to a funding crisis.

AI adds pressure to a market that was already under pressure. The tasks most at risk — fact aggregation, brief rewrites, data summaries — are exactly the entry-level tasks that used to be how younger journalists built their careers. If those tasks get absorbed by automated tools at wire services and aggregators, the traditional path into the profession gets narrower. That's a pipeline problem more than a replacement problem, but the effect on job numbers is the same.

The 4,100 annual openings BLS projects are real too. Retirements, new beats, specialist roles in data journalism and video, and the growth of independent and nonprofit journalism outlets keep seats opening up. The roles that survive will skew toward journalists who can do what AI can't: report from the ground, cultivate sources, and make editorial judgments that hold up under scrutiny. The market is shrinking, but it isn't disappearing, and the journalists who stay will be doing the parts of the job that actually required a journalist in the first place.

The 21% AI exposure score for journalism is likely to stay roughly flat over the next five years. The high-penetration tasks are already automated where automation makes sense. For the exposure score to climb significantly, AI would need to solve source cultivation, physical presence, and real-time verification. None of those are close. Source relationships run on human trust, and no model trained on past text can build a new one.

The scenario where this score jumps is a breakthrough in AI agents that can autonomously browse public records, cross-reference databases, and flag inconsistencies in government documents. Tools like Harvey, built for legal document analysis, hint at what that might look like in an adjacent field. If something similar arrives for investigative document work, the research side of journalism gets faster again. But the reporting side, the interviews, the observation, the judgment calls about newsworthiness, stays human for the foreseeable future. Ten years out, the job looks different in its tools. It doesn't look different in its core purpose.

The tasks to double down on are the ones with zero AI penetration: source development, on-the-ground investigation, and the kind of contextual judgment that only comes from knowing a beat deeply. If you cover local government, know every name in that building. If you cover tech, understand what engineers actually build versus what press releases claim. That depth is the thing that makes your byline worth something, and no tool replaces it.

Data journalism is one of the clearest growth areas. Knowing how to file FOIA requests, clean a dataset in Excel or Python, and turn public records into a story is a skill set that very few journalists have and that AI tools can assist with but can't lead. The Washington Post and ProPublica have both built significant audience around data-driven investigations, and smaller outlets are looking for journalists who can do that work without a dedicated data team behind them.

On the career structure side, consider where the business model is sustainable. Nonprofit outlets, subscription-based publications, and specialist trade press have held headcount better than general-interest print. Video and audio are also worth building into your workflow now if you haven't: multimedia journalists who can report, shoot, and edit get hired over single-format reporters when budget is tight. The documentation tools covered earlier can help you get the mechanical work done faster, which frees time to build these longer-term skills. The journalists who'll be fine in ten years are the ones who used the time they got back from transcription to make one more call.