■ CRITICAL RISK ■ Media & Communication
For grammar, consistency, and house style, AI is already a faster, cheaper copy editor that never gets tired at page 300. The human role shrinks to a final-judgment layer — and there will be far fewer chairs at that layer.
“AI catches errors you'd miss on your third read-through.”
Our AI replacement risk score — how we score jobs
Copy editing is a stack of tasks with very different automation profiles. At the bottom: catching typos, subject-verb disagreements, comma splices, and inconsistent hyphenation — work that grammar checkers did passably a decade ago and language models now do genuinely well, at speed, across an entire manuscript, without the attention fatigue that makes human error rates climb after hour three. Style-guide enforcement — is it 'health care' or 'healthcare' per house rules — is exactly the kind of consistent rule-application machines were born for.
The middle of the stack is where it gets contested. Fact-checking a suspicious date, flagging that a paragraph implies something legally actionable, noticing that the writer used 'flaunt' where they meant 'flout' in a way that changes meaning — LLMs catch some of this and confidently miss the rest, occasionally 'correcting' text into being wrong. Query-writing, the diplomatic art of telling an author their favorite sentence doesn't work, requires tact and institutional knowledge of what this publication and this writer will tolerate.
At the top sits judgment that resists automation: preserving a writer's voice while fixing their sentences, knowing when a rule should be broken, catching the error of sense rather than syntax — the sentence that is grammatically perfect and factually catastrophic. The problem for employment is structural: publishers, already operating on famine budgets, are exactly the industry that will accept 'good enough' automation. Newsrooms cut copy desks to the bone before AI arrived; the technology now finishes an amputation that cost-cutting started. Our 78 reflects fewer jobs, not zero — one senior editor supervising AI output where five once sat.
Automatability: our editorial assessment of current and near-term AI capability
Already biting. AI editing tools are in daily use across publishing, marketing, and journalism, and organizations that once employed copy desks now employ a person who supervises a tool. Through the late 2020s expect steady contraction in staff roles and rate pressure on freelancers, with surviving positions concentrated in book publishing, legal-sensitive media, and anywhere errors are expensive enough to justify a human's final pass.
Dying is strong; consolidating brutally is accurate. AI now handles mechanical correction and style enforcement well enough that publishers need far fewer copy editors per page. Surviving roles skew senior: people who supervise AI output, handle sensitive material, and do substantive editing. Entry-level proofreading-adjacent work is largely gone or going, which means the traditional way into the profession is closing too.
On mechanics — grammar, spelling, consistency — it's arguably better, because it never fatigues and never skims. Where it falls short: errors of sense, legally risky implications, voice preservation, and knowing when a rule should be broken. It also occasionally introduces confident errors while 'fixing' things. That gap keeps humans employed, but as reviewers of machine output rather than first-pass editors.
Reposition above the machine. Learn the AI tools thoroughly so you can offer supervised, guaranteed-quality editing at competitive speed. Push your services toward substantive and developmental editing, sensitivity reads, and high-stakes content where clients fear errors. And cultivate direct relationships — the freelance editor a nervous author trusts personally is harder to replace than a line item in a publisher's budget.
Longer than most, yes. Book publishing has quality expectations, long production cycles, and authors who negotiate for human editing, so human copy edits will persist there while newsrooms and content mills automate first. But even in books, expect AI to do the initial pass with humans reviewing — meaning fewer total hours of paid human editing per title, and rates that reflect it.