■ CRITICAL RISK ■ Media & Communication
For routine news copy — earnings recaps, sports summaries, rewritten press releases — yes, and it already did. What survives is reporting: finding out things that aren't in the source material. Copywriting that just reshapes existing text is exactly what language models are.
“GPT wrote this description. And your next article.”
Our AI replacement risk score — how we score jobs
News copywriters produce the connective tissue of a newsroom: rewriting wire copy and press releases into house style, drafting headlines and SEO-friendly summaries, producing quick-turn items on earnings, weather, traffic, and sports scores, and versioning the same story for web, newsletter, and social. It's deadline writing about information someone else already gathered — the reporter got the quotes; the copywriter makes them publishable in eight formats before lunch.
That job description reads like a language-model product sheet. Newsrooms were automating templated stories — earnings reports, game recaps, real-estate listings — with tools like Automated Insights years before ChatGPT existed. Modern LLMs go further: they summarize press releases, generate headline variants, adapt tone across platforms, and localize wire stories in seconds. Publishers under permanent margin pressure have every incentive to use them, and many openly do for exactly this tier of content. The economics are stark — a copy desk's daily output of routine items is now an editor reviewing machine drafts, and the same AI-driven search summaries that answer readers directly are cutting the pageview traffic that funded the desk in the first place.
The defensible territory is anything requiring new information or accountability. Original reporting — working sources, filing records requests, being in the room — produces the raw material AI can only remix. Editorial judgment about what's newsworthy, legally safe, and true still needs a human who can be fired for getting it wrong; hallucinated facts in news copy are a lawsuit, so human review persists even where drafting doesn't. But review is one job where drafting was five, and our 89 risk score is doing that arithmetic.
Automatability: our editorial assessment of current and near-term AI capability
Already in production, literally. Templated news automation predates the LLM boom, and publishers now use generative tools for summaries, headlines, and routine items as standard practice. Copy desk headcount keeps shrinking through this decade as one editor reviewing AI drafts replaces several writers producing them — while AI search erodes the traffic that paid for the desk. Original reporting roles decline more slowly, for economic rather than technical reasons.
Partially, and visibly. Automated earnings and sports recaps predate ChatGPT, and publishers now routinely use LLMs for summaries, headline options, and rewriting releases. Human copywriters increasingly review and polish machine drafts rather than write from scratch — which means fewer of them. The layoff waves across digital media reflect this shift plus collapsing traffic economics, not just belt-tightening.
Different exposure levels. Copywriting — reformatting existing information — is the most automatable layer. Original reporting, investigations, and beat expertise resist because AI cannot attend a council meeting, develop a source, or take legal responsibility for a claim. The economics are still brutal, since AI-summarized search results starve publishers of clicks, but the reporting skill itself remains the defensible core of the profession.
Reposition toward things with your name and judgment on them. Pitch original stories, however small, until reporting is on your résumé. Become the desk's best AI editor — fast, skeptical, good at catching hallucinated facts — because that role survives the transition. And build a portable audience through a newsletter or beat specialty; institutional desk jobs are shrinking faster than individual reputations.
For commodity information — scores, stock moves, weather — they already do and rarely notice. For anything involving trust, accountability, or analysis, surveys keep finding readers skeptical of undisclosed AI authorship, and publishers have taken reputational damage when AI-written content shipped with errors. That skepticism protects reported journalism somewhat. It does nothing for the anonymous rewrite desk, where provenance was invisible anyway.