When should you really use AI in your writing?

The job of a journalist requires different skill sets - not all journalists need to be writers.

Photo by Etienne Girardet on Unsplash

When I was a reporter at a daily newspaper, fellow journalists would often pull me aside with a quiet request: look over my piece before it goes to the desk. No reporter liked the editing desk - the sub-editors were necessary, even good at their jobs, but handing copy directly to them felt like surrendering. Better to have a fellow reporter catch the rough edges first. It was like cleaning the house before the cleaning lady arrives.

What those requests taught me, early on, was that being a good reporter and being a good writer are not the same thing. The colleagues who came to me most often were frequently the best reporters in the room - the ones with sources no one else had, the instinct to be in the right place, the ability to read a situation the moment they walked in. They just didn’t particularly enjoy writing, or weren’t especially fast at it, and in a daily newspaper, speed is not optional.

I’ve been thinking about this lately in the context of AI - specifically, the tendency in conversations about AI and writing to treat “writing” as a single, undifferentiated activity.

As if a breaking news report, a data investigation, a 5,000-word essay, and a novel all make the same demands on the person producing them. They don’t. And conflating them leads to both over-reliance on AI in places where it corrupts the work, and unnecessary resistance in places where it genuinely helps.

It’s worth separating the categories.

The daily reporter

A good reporter’s core skill is judgment - knowing what matters, finding the people who know things, asking the right questions, sensing when a story is more than it appears. These instincts take years to develop and have nothing to do with the elegance of a sentence.

In Indian English-language newsrooms - and honestly in newsrooms everywhere - many excellent reporters are not polished writers. That’s not a criticism. It’s a recognition that the craft of reporting and the craft of writing are related yet distinct, and that laboring over prose often conflicts with the pace daily journalism demands.

Here, AI is genuinely useful. Tightening a rushed draft, fixing grammar, restructuring a lead that isn’t landing - these are mechanical tasks, and AI handles them competently. The reporting, the judgment, the facts: those remain entirely the journalist’s. AI is the fellow reporter you can pull aside at midnight.

Long-form and analytical writing

This is where the relationship with AI needs more care. Long-form writing - essays, investigations, in-depth features - is often less about recording what happened and more about thinking on the page.

The process of writing is the process of figuring out what you actually think. A tangled first draft is not a problem to be cleaned up; it is the writer working through an argument.

AI can help at the edges: organizing a sprawl of notes, identifying where an argument loses thread, flagging structural gaps. What it cannot do - and should not be asked to do - is generate the thinking itself. The moment you outsource the actual writing of long-form work to AI, you’ve outsourced the reasoning. The output may read smoothly and still say nothing of substance.

Use AI as a thinking partner, not a ghostwriter.

Data journalism

This is where AI earns the least ambivalent endorsement. Data journalism involves acquiring datasets, cleaning messy information, running analysis, finding patterns, and translating numbers into stories a general reader can follow. Much of this work is technically demanding in ways that have nothing to do with narrative skill.

AI - and particularly code-generating AI - can dramatically accelerate this pipeline. Journalists who are not programmers can now interrogate datasets they previously couldn’t.

Those who are programmers can move faster. The editorial judgment about what the data means and why it matters still belongs to the journalist. But the tools have genuinely expanded what’s possible.

Creative writing

Here, AI’s role shrinks considerably. Creative writing is, at its core, an act of individual consciousness - a particular mind’s way of seeing, rendered into language. It is defined by the choices it makes, including the imperfect, idiosyncratic, surprising ones.

AI can be a useful brainstorming partner: helping a writer work through a structural problem, offering alternatives when stuck, generating raw material to react against. But the writing itself, if it is to mean anything, has to come from the writer.

Which brings me to why any of this matters.

I recently finished a book that was, very clearly, written by AI. I don’t know whether it was an experiment or simply an abdication. What I know is how it felt to read: smooth, competent, and utterly empty.

And yes, of course, there has always been formulaic writing. Long before AI, we had editors, market trends, genre expectations, house styles, bestseller templates, and neat little rules about what makes prose “work.” Plenty of human writing is generic. Plenty of books (and articles) are written to a formula.

But even in formulaic human writing, something usually slips through. A sentence. A word. An odd observation. A small irritation. A strange tenderness. Some little detail that tells you: a person was here. Someone noticed this. Someone lived long enough to write it.

That is the part writers are often taught to smooth out. The crooked bit. The overlong sentence. The slightly inconvenient thought. The paragraph that does not behave. And of course, not every rough edge is genius. Sometimes bad writing is just bad writing. But sometimes the broken, uneven, unpolished part is where the actual mind is.

Every time I read a book, I feel as though I am walking through the writer’s mind. I notice what they linger on, what they avoid, where they lose patience, where they become unexpectedly generous. I want the strange turn, the lived detail, the sentence only this person could have written.

In this book, there was no such presence. Everything worked, and yet the writing, even the plot twists, felt like a machine’s version of human life: competent, mechanical, dead.

It was a useful reminder that fluency is not the same as thought. And that the goal of writing — in any form that asks something of a reader — is not simply to produce text. It is to transfer something true from one mind to another.

This is why writers should be careful about surrendering too much of themselves to polish. Use the tool, yes. Use the editor. Use the structure. Clean up what needs cleaning. But keep the part that feels inconveniently yours.

That is usually where the magic is.