Can AI Replace Writers and Authors?

By Dr Sonia Sharma • drsoniasharma.info • April 2026

I want to begin with the thing nobody in this debate admits: I have recently started using AI as a writing assistant. I tried it on my own fiction, a short story from my book ‘Imperfect Lives’ published in 2018, available on Amazon Kindle. On Betrayal — a story about a RAW agent, a secret marriage, and twenty-four years of sacrifice. And I am going to tell you exactly what it could do and what it could not. Because the truth of this question is not found in think-pieces written from a distance. It is found inside the work.

The statistics are already frightening, if you choose to read them that way. 

A 2025 University of Cambridge study found that 51% of published novelists believe AI is likely to entirely replace their work as fiction writers. Over a third — 39% — say their income has already taken a hit. And 85% expect their future earnings to be driven down by the technology.

51% of published novelists believe AI is likely to entirely replace their work; 85% expect future income to decline — University of Cambridge / Minderoo Centre, 2025

These are real people with real anxiety, and I will not dismiss it. But I want to ask a different question. Not 'will AI replace writers?' — because that is the wrong unit of analysis. 

The right question is: replace which writers doing what, and for whom?

 When I started working on AI, I used AI at specific moments. I asked it to suggest alternative chapter structures when I was stuck. I asked it to generate three possible versions of a scene I had written myself, so I could see what I was not doing that I could be. It’s not that I hadn’t used AI earlier, back in 2017-2018, it wasn’t called AI. It was merely Google that we all used for our research. I did too. I asked it to check factual details about the Antwerp-Breda Highway and Belgian geography, so I did not have to interrupt the writing to research.

It was useful. Genuinely. I didn’t have to travel that far to write a fiction for factual details.

But with AI, when I asked for help, the structural suggestions were sometimes genuinely illuminating — not because any of them were right, but because seeing what a very well-read pattern-matcher suggested made me more certain of what I actually wanted. The factual checking saved me hours. I started with my blog writings instead of jumping into actual book scenes. but trying it for fiction, that too about the relationship dynamics was a different ball game. The alternative scene versions occasionally it created offered a phrase I absorbed and reworked beyond recognition. It felt good because it brought me out of my writer’s block – temporarily. So I put it to test. I uploaded the script of Betrayal and asked it to elevate it or rewrite it, making it more punchier. It did. I wasn’t satisfied because the mystery was resolved before the chapter could begin. I commanded it to rewrite, maintaining the mystery and addressing my concerns. And it did. Without whinning, without fatigue, without asking twice why I was so choosy. But…. This but is where the contradiction began….?

So, here is what it could not do. It could not generate the central tension of my story. The question that is central to the plot — what does a man owe his country versus the woman he loves, and is there an answer that does not destroy him — is not a question that emerges from pattern-matching on existing fiction. It emerged from fourteen months of private discomfort. From arguments I had with myself at two in the morning. From a particular feeling I could not name that eventually found a shape in a character.

“AI gave me a mirror. What it cannot give me is the face.”

The writing world has a tendency to conflate two entirely different activities: the production of text and the making of meaning. AI is extraordinarily good at the first and structurally incapable of the second — not as a temporary limitation, but as a matter of what it is.

A language model is trained on what human beings have already written. It is the most comprehensive reader in history — it has absorbed more text than any human could read in a thousand lifetimes. When you give AI a command to analyse a text, Claude, for that matter, try checking the small text that shows its activity. I noticed it reading 2.5 billion texts. Mindboggling, right? 

But reading, however exhaustive, is not experience. And experience — specifically, the friction of being a particular person in a particular life in a particular culture with particular wounds and particular loves — is where literature comes from.

Consider: AI can write a scene of a woman watching her husband's face on a television news broadcast, identified as a criminal she never knew he was. It can do this competently. It has read thousands of scenes of grief and shock and betrayal. But it has never watched anyone it loved be reframed by a screen into a stranger. The scene it produces will be technically correct and emotionally approximate. The scene I write after sitting with that question for a year will be something else.

45% of writers already integrate AI into their workflow; authors using AI report 31% productivity gain on average — BookBub / Publishers Weekly, 2025

The Cambridge study notes that genre fiction writers are considered most vulnerable to displacement. I think this deserves honest examination. Content-at-volume writing — SEO articles, product descriptions, certain categories of genre fiction produced to formula — is genuinely at risk. Not because AI is as good as a talented writer, but because it is good enough for buyers who were never paying for talent in the first place. That market compression is real and it is happening now.

But literary fiction — the kind built on a specific human perspective, on voice that cannot be genericised, on the author's particular way of noticing the world — is not at the same risk. Not because publishers will protect it out of principle (they will not always), but because readers know the difference. When you read a novel that feels like it was generated rather than inhabited, you put it down. The experience of reading is also the experience of encountering another consciousness. Remove the consciousness, and you remove the reason to read.

I say this not as consolation. I say it as a structural argument: what AI replaces first is the work that was never really about a human voice to begin with. What it cannot replace is work where the voice IS the product.

 There is a further dimension to this that I want to name, because it is the one that makes me angriest and most uncertain at the same time. 

In 2025, Anthropic settled a major lawsuit for $1.5 billion regarding the use of copyrighted works for model training. Sixty-nine per cent of authors surveyed by the Authors Guild report feeling threatened by AI regarding their long-term careers.

These models were trained, in large part, on work that writers created without being asked, without being paid, without being credited. The stories, the novels, the essays, the poems — all of it fed into systems that now compete with the people who wrote them. That is not a technical problem. It is a moral one. And it is unresolved.

My own position is that writers should be asked. That 'opt-out' is backwards when 'opt-in' is the principle we apply to every other use of someone's creative work. That AI companies that built their capabilities on human writing owe human writers something material in return. What that looks like, legally and practically, is still being decided in courtrooms. But the principle seems clear to me.

 No. And yes. And there are two dimensions to it. 

One -  it depends entirely on which writers you mean, what you value in writing, and whether you are willing to accept approximation as a substitute for presence.

AI will replace writers who were providing a service that did not require them to be themselves. It will not replace writers who have something to say that only they can say, in a voice that is the product of their particular, irreducible life.

What it will do — is already doing — is change what writing is valued for. In a world where text is abundant and cheap, the scarce thing is not words. It is a recognisable human consciousness behind them. A perspective. A wound. A vision. The writers who will survive this are the ones who understand that their job was never to produce text. It was always to bear witness to experience in a way that makes the reader feel less alone.

That cannot be automated. Not because the technology is not clever enough. But because it requires having lived.

Two – And the larger question and a short answer to it is – Probably. Because, its not what AI writes that will make writers go out of jobs. It’s the abundance of AI clutter in visual media and shorter attention spans that there will be fewer people who’ll be interested in reading. So, it’s not the writing processes of AI that will make writers extinct, but the lack of enthusiasm to read that would discourage writers to add new works into the system. And AI will be forced to keep on recycling the old data, packaging it in new words cyclically. 

I’m attaching the link to my story, Betrayal, in this article. Both versions – Betrayal -original. And Betrayal enhanced 

Do share your thoughts on which one you like.