AI Can Write for You. But Can It Teach You to Write?

By Jim White

Years ago, I met two guys who ran a successful retail design agency.

They had written a book about their field and were using it, very effectively, to market their business. At the time, I wanted to write a book for RealityCheck, so I asked them the obvious question:

“How did you find the time to write a book?”

Their answer surprised me.

They didn’t.

Instead, they hired a professional writer, a former Chicago newspaper reporter. Once a week, the three of them met for a couple of hours. The agency owners would talk through the next section of the book: what they believed, the argument they wanted to make, the examples they wanted to include. The writer would listen, ask questions, go away and draft the section. The owners would review it, provide feedback, and they would repeat the process the following week.

I remember thinking it was a brilliant solution.

Years later, I realized I now work in almost exactly the same way.

Except my ghostwriter is AI.

In fact, this article was written that way. I worked out the argument, talked it through with AI, reviewed the drafts, challenged what didn’t work, rewrote passages and kept sending my ghostwriter back for another try.

While the technology is revolutionary, the working relationship is familiar.

My AI Ghostwriter Is Really Good

I write with AI almost every day now.

Lately, much of that work has involved developing a series of briefing books based on proprietary RealityCheck research. These are substantial pieces of writing that bring together academic research, our own consumer data, strategic frameworks and business implications.

I don’t ask AI to “write me a book.”

The process is much closer to working with a talented professional ghostwriter.

I work out the central argument. I decide what the story needs to be. I develop an outline and explain what each section needs to accomplish. Then I have AI draft sections one at a time.

From there, I edit, write and rewrite.

I challenge weak reasoning. I cut clutter. I tell it when a paragraph is overwritten, when the lead is buried, when the evidence doesn’t support the claim or when a sentence simply doesn’t sound like me. Sometimes I rewrite the passage myself. Sometimes I explain the problem and ask the AI to try again.

Then we keep going.

Before we began the briefing-book series, I worked with AI to develop an editorial framework based on my own writing habits and on principles from writing teachers and editors I have long admired, most notably William Zinsser.

The process can produce remarkably good writing and, at least for me, be quite enjoyable. But that is also what has started to worry me: not because AI writes badly, but because it can write so well.

The Catch

The partnership works because, over decades of writing, I’ve learned how to supervise the writer.

I have spent most of my career writing and being edited. Before market research, I was a newspaper reporter. I have written research reports, articles, proposals, presentations and more pages of business writing than I care to count.

Most of these pages did not begin as good writing.

That is how I learned.

I learned by staring at paragraphs that did not work and figuring out why. By having editors cross out sentences I loved. By discovering halfway through a story that I had started in the wrong place. By realizing that an argument I thought was brilliant collapsed as soon as I tried to explain it clearly.

Those experiences built judgment.

Today, AI can generate the first draft. But I still have to know whether the draft is any good.

That requires a different skill from simply producing sentences.

It requires editorial judgment.

The question I’ve been thinking about lately is how young professionals are going to develop that judgment when the ghostwriter is sitting beside them from Day One.

Writing Is Thinking

Years ago, I came across one of my favorite observations about writing from historian David McCullough:

“Writing is thinking. To write well is to think clearly. That’s why it’s so hard.”

Anyone who writes regularly knows the experience. You sit down believing you understand something. Then you try to explain it.

Suddenly, two ideas that seemed connected are not connected at all. The evidence turns out to be weaker than you remembered. A paragraph contains three arguments fighting for space. The insight you thought was central becomes less interesting, while something you had treated as secondary starts to look like the real story.

Writing exposes fuzzy thinking.

At RealityCheck, we have built this belief into our reporting process. We call it Insights Journalism®. We do not treat writing as something that happens after analysis. Writing is part of our analytical process. It is one of the ways we test ideas, discover relationships and find the story in the data.

A lot of people in the market research industry like to talk about “storytelling.” But most have no clue what that actually means or, more importantly, how to do it. If you want to be a storyteller, learn to write.

AI can change the process of learning to write because it can remove much of the struggle. The question is how much of that struggle we should want to remove.

Formative Friction

There is a useful concept in education and learning called formative friction: some forms of difficulty are not obstacles to learning. The effort required to work through them is part of how capability develops.

Not every difficult task deserves to be preserved simply because it is difficult. I have no desire to return to typewriters, correction fluid or library card catalogs. Technology has eliminated enormous amounts of pointless friction from knowledge work, and I am grateful for it.

But some forms of friction build capability.

Learning to play guitar requires fumbling through scales. Becoming a good interviewer requires conducting some mediocre interviews. Learning mathematics requires working through problems you initially do not know how to solve.

I believe learning to write requires the same kind of formative friction.

The blank page is frustrating partly because it requires the writer to make decisions. What is my point? Where do I begin? What follows from what? What evidence do I need? What can I leave out?

AI can make many of those decisions for us. That is often useful. But can a writer delegate those decisions before they’ve developed the ability to make them?

The question is not whether AI should eliminate friction from writing. It is which friction is waste and which friction is forming the writer.

What the Research Is Beginning to Show

The research is still developing, and I do not think we know enough yet to draw sweeping conclusions about what AI will do to writing ability. But two recent studies raise questions worth paying attention to.

A 2025 study from Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about how they used generative AI in real-world tasks. The researchers found that greater confidence in AI was associated with less reported critical-thinking effort, while greater confidence in one’s own expertise was associated with more critical thinking.

That sounds familiar to me. Over many years, I’ve developed confidence in my ability as a writer. I’d like to think that allows me to work with AI critically rather than simply handing the thinking over to the machine.

A more recent study of AI-assisted writing makes another useful distinction. Students who had relatively open access to AI produced strong work while the AI was available. But on a later writing task they had to complete without AI, they did less well than students whose use of AI had been more limited and required greater reflection.

The authors are careful about the limits of the study, and the findings should not be treated as definitive. Still, the study raises an important possibility: students who lean heavily on AI may produce better work in the moment without necessarily becoming better writers themselves.

That gets very close to the question I am asking.

A person can produce an excellent document with AI without necessarily developing the ability to produce, diagnose or improve that document independently.

From Doing the Work to Supervising the Work

This shift from producing to supervising is already visible in professional work.

When I review an AI draft, I am constantly asking questions such as: Is this interesting? Is this the lead? Does the argument follow? Is that claim supported? Is this sentence saying anything? Is that phrase a cliché? Does this sound like me? What can I cut? What is missing?

Those questions are not prompting skills. They are judgment skills that come from experience.

The more people use AI for writing, the more valuable that judgment may become. But the developmental problem remains: how does someone learn to answer those questions if they have never spent much time struggling to produce the underlying work?

Can someone become a strong editor without first spending much time being a writer?

Maybe. But I would not assume it.

What Happens When the Ghostwriter Arrives First?

The two retail design executives I met years ago had something important before they hired their ghostwriter. They had expertise.

They knew their business, had a point of view and understood the argument they wanted to make. The writer helped transform their knowledge into a book.

AI has democratized that arrangement.

Today, an 18-year-old college student can have access to a ghostwriter more capable than the one most executives could have hired 20 years ago. So can a 22-year-old analyst on her first day at work.

I think that is mostly a wonderful thing.

Younger professionals should use these tools. The idea that we can somehow keep AI out of education or professional life is neither realistic nor desirable.

But their apprenticeship will be different from mine.

My generation learned to write first and acquired the ghostwriter later.

The next generation will have to learn the craft while the ghostwriter is sitting next to them, constantly offering to finish the sentence. And that offer will be tempting because good writing is hard.

Journalist and author Gene Fowler is widely credited with one of my favorite descriptions of the process:

“Writing is easy. All you do is stare at a blank sheet of paper until drops of blood form on your forehead.”

That’s funny because there’s truth in it. When something is that difficult and a machine offers to do it for you, resisting the offer takes some discipline. That changes the job of educators and managers.

Teaching Writers in the Age of AI

The answer cannot simply be, “Make them write without AI.”

I think there will always be value in some unaided writing, particularly when people are learning to form arguments, organize ideas and confront the blank page. But the larger goal should be teaching people to become good enough writers to work intelligently with AI. That means teaching more than prompting.

Young writers should learn to critique AI-generated drafts. Give them a mediocre lead and ask them to improve it. Ask what can be cut. Have them identify unsupported claims, restructure weak arguments and explain why one paragraph works better than another.

Expertise in writing develops when people learn not only to recognize that something feels wrong but also to understand and articulate why.

I would also teach AI as a junior writer rather than an oracle. The human should originate the argument, brief the AI, review what comes back, reject weak work and direct revisions, all while assuming it won’t get things right the first time. That keeps responsibility where it belongs.

And we should continue teaching great writing.

Zinsser is no less relevant because ChatGPT exists. Clarity still matters. Economy still matters. Structure, voice, logic and humanity still matter. In fact, when machines can generate competent prose instantly, those qualities may become more important because competent prose itself is no longer scarce.

The Writer Still Matters

AI has already made me a more productive writer.

In some ways, I think it may be helping me become a better one. I can test ideas faster, explore alternative structures and spend more time editing instead of grinding through every first draft.

But I bring something important to the partnership as well: decades spent learning what good writing looks like.

The next generation of professionals will have to acquire that judgment differently than I did. That is not a criticism of them. It is simply the environment in which they will learn.

The goal should not be to keep AI away from young writers. It should be to make sure they become good enough writers to tell the machine when it is wrong, when it is boring, when it is unclear, when the argument does not hold together and when it needs to try again.

Having a great ghostwriter is useful. Knowing what you want the ghostwriter to say is still a writer’s job. And if AI becomes everyone’s ghostwriter, we should spend some time figuring out how everyone still learns to write.