Will AI Replace Journalists?
AI will not replace journalism as a whole, but it is already replacing specific tasks inside it: earnings summaries, sports recaps, and first-draft transcription. The jobs most at risk are the ones built entirely around those repeatable tasks. Original reporting, verification, and the human relationships that produce a scoop are much harder to automate, and nobody has done it at scale yet.
The U.S. Bureau of Labor Statistics projects a 4 percent decline in jobs for news analysts, reporters, and journalists between 2024 and 2034. That decline started well before generative AI existed. Newsroom employment has been shrinking since the 1990s because of collapsing ad revenue, not because software started writing the news.
So the honest answer sits in the middle: fewer jobs overall, driven mostly by economics, with AI now speeding up a few of the tasks that used to take up a reporter's day.
What the data actually says
The BLS Occupational Outlook Handbook puts total 2024 employment for this occupation at 49,300 jobs, projected to decline 4 percent by 2034. Despite the shrinking headcount, BLS still expects about 4,100 openings a year on average over the decade, almost all from people leaving the field or retiring rather than new demand (BLS Occupational Outlook Handbook).
BLS is explicit about the cause: declining advertising revenue across radio, newspapers, and television, combined with stations moving content online where traditional ad formats sell for less. That is a business-model problem that predates ChatGPT by two decades. AI adoption inside newsrooms is a newer factor layered on top of an existing decline, not the thing that started it.
Which tasks are exposed
Some newsroom tasks are close to fully automatable today, and several outlets already run them that way:
- Earnings-report summaries and other data-driven financial writing
- Sports recaps built from box scores and play-by-play data
- Weather and traffic blurbs pulled from structured feeds
- Transcribing recorded interviews before a reporter writes from them
- First-draft SEO listicles built around a known template
- Rewriting wire copy into a slightly different version for a local audience
These share one trait: the underlying facts already exist in structured form somewhere, and the writing step is mostly formatting. That is exactly the kind of work large language models handle well.
Which tasks are protected, and why
Original reporting depends on relationships a model cannot build. A source talks to a specific reporter because of a track record of accuracy and discretion, not because a byline exists. That trust takes years and can end instantly if a reporter burns a source, which is a risk no AI system currently bears any responsibility for.
Verification and legal accountability sit with a named human under libel law. If a story is wrong and someone is defamed, the publication and the reporter carry that liability. Editors need someone accountable who can be deposed, corrected, and held to a standard, which is a different kind of responsibility than checking a model's output for plausibility.
Investigative work often requires physical presence: sitting in a courtroom, knocking on doors, meeting a source somewhere they'll actually talk, requesting documents through a slow bureaucratic process. None of that has a text-generation equivalent.
Editorial judgment about what counts as news, which angle matters, and what to leave out is a value judgment shaped by audience knowledge and ethics, not a pattern to predict from training data.
Protecting a source's identity, sometimes under legal or physical pressure, is a human obligation a model cannot hold.
What is already happening
The Associated Press has run automated earnings reporting since 2014, using Automated Insights' Wordsmith software to generate quarterly earnings stories. The result was a roughly tenfold increase in earnings coverage, from around 300 stories a quarter to 3,000, extending coverage to many small companies AP never had the staff to cover before. AP has said no jobs were cut as a result, and reporters freed from routine earnings writing shifted toward user-generated content, multimedia, and investigative work (Poynter).
CNET took the opposite lesson in January 2023. The outlet quietly published more than 70 personal-finance articles under an AI byline, and when the errors were reviewed, 41 of the 77 AI-generated articles needed corrections, some for basic math mistakes and some for language close enough to existing text to raise plagiarism concerns. One widely cited example: an article on compound interest claimed a $10,000 deposit at 3 percent would earn $10,300 in a year, when the real answer is $300 (Washington Post). CNET paused the program and rewrote its AI policy months later.
The Reuters Institute's 2025 research on generative AI and news found journalists increasingly using AI for research, transcription, and drafting suggested headlines and summaries, with UK newsrooms leaning on it most for language tasks where small errors are lower-stakes. The same research found the public is far less comfortable with AI writing news than journalists are with using it as a tool: on average across six countries, only 12 percent of people say they're comfortable with news made entirely by AI, compared with 62 percent comfortable with news made entirely by a human (Reuters Institute, Generative AI and News Report 2025).
What to do about it
Specialize in work that depends on being physically present or personally known to sources: courthouse reporting, city government, a specific industry beat where sources return your calls because they know you. That kind of access does not transfer to a general-purpose model.
Build direct source relationships deliberately rather than relying on press releases and wire copy. A reporter with three sources who call first, before a press office sends a statement, is doing something a summarization tool cannot replicate.
Consider becoming the person who fact-checks and edits AI-assisted drafts rather than someone competing against the drafts. CNET's own error rate shows why that role matters: a newsroom that uses AI for first drafts still needs someone who catches a $10,000 miscalculation before it publishes. That skill, checking AI output against reality, is becoming its own specialization inside newsrooms rather than a side task.
Five days to take back your core tasks
The tasks you still do by hand, without checking a tool first, are the ones that keep you valuable. The free 5-Day AI Reset is a five-email course built around exactly that: Day 2 has you take one task back and do it unassisted. One small change per day.
FAQ
Is AI already writing news articles?
Yes, for narrow categories like earnings reports, sports recaps, and some SEO-driven finance content. AP has automated earnings stories since 2014 without job cuts. CNET's 2023 attempt at broader AI writing produced errors in more than half its published articles and was paused.
Will AI replace investigative journalists?
Not with current technology. Investigative reporting depends on building trust with sources, obtaining documents through legal processes, and being physically present, none of which a language model can do on its own.
Why are journalism jobs declining if AI isn't the main cause?
BLS attributes the projected 4 percent decline through 2034 mainly to falling advertising revenue as audiences move online, a trend that started in the 1990s, long before generative AI existed.
How many journalism jobs will open up despite the decline?
BLS projects about 4,100 openings a year on average through 2034, almost all from people leaving the occupation rather than new positions being created.
Do audiences trust AI-written news?
Reuters Institute's 2025 research found only 12 percent of people across six countries are comfortable with news made entirely by AI, versus 62 percent comfortable with news made entirely by a human journalist.
What is the safest specialization for a journalist right now?
Beat reporting that depends on direct source relationships and physical access, plus editorial and fact-checking roles that catch errors in AI-assisted drafts before publication.
Want a personal read on your own risk?
BLS data and industry cases describe the occupation. Your own daily tasks are more specific than that. Take the AI-proof-your-job assessment to see which parts of your actual workload are exposed and which are protected.