Why using AI could be hurting your writing

AI is now part of many writers' workflows – and it's not surprising. Large language models like Claude and ChatGPT are no longer standalone platforms. They’re baked into software, productivity tools, email platforms and website builders. And the message is clear: adapt or get left behind.

Which is probably why I'm seeing more AI-assisted books hitting my inbox. To be honest, most are from authors who are self-publishing, but I’ve seen a few from publishers too. At best, the author has used AI tools to outline or edit sections they were struggling with. At worst, they’ve cut and pasted large chunks of text, changing only a few details.

And I get it. Authors want to speed up the process and polish their writing. But no matter how many prompts you use, or how much you try to hide the ‘evidence’, AI could be hurting your writing rather than strengthening it. Here are some of the pitfalls.

Your skills deteriorate

The beauty of writing is that it helps you work through ideas: organising, discarding, clarifying, refining. As author and teacher William Zinsser put it, ‘Writing is a tool that enables people in every discipline to wrestle with facts and ideas … It compels us by the repeated effort of language to go after those thoughts and to organize them and present them clearly’.

When you outsource this process, your thinking and writing become lazy. In fact, you fall prey to a phenomenon known as cognitive surrender, which means that you accept AI’s reasoning with minimal oversight or verification, believing it to be your own output. As a result, your thinking and writing skills decay.

And there are studies to back it up. A 2026 cross-domain review of research on AI-overdependence found that it was 'a significant threat to higher-order cognitive skills, especially critical thinking, complex decision-making, and analytical reasoning, while also undermining creativity, memory, and executive functions'.

It went on to explain that '[t]hese effects stem from well-documented mechanisms such as cognitive offloading, metacognitive laziness, and automation bias, compounded by the perceived authority and convenience of AI systems.' In short, people are intrinsically 'lazy' and look for shortcuts, and they trust machines more than themselves, even when the machines are flawed, biased or wrong. The result? Skill decay.

(For a broader look at this topic, see The Silent Erosion: How AI’s Helping Hand Weakens Our Mental Grip.)

You lose your voice

Generative AI uses predictive patterns, not logic or thought, which means its output is generic and … boring. Sure, it may look polished and error-free, but it lacks the specificity and personality found in human writing. Even if you feed it reams of your work and prompt it to clone your voice, it’s still an approximation (or even a parody) of the real thing.

Think about your favourite writers. What makes their work memorable? Is it a distinctive voice and point of view, unusual sentence patterns, lived experiences? AI can’t replicate this. Without those vital ‘human’ components, you’ll end up sounding like everyone else using the same prompts.

It gives you the illusion of productivity

AI platforms promise speed and convenience. You can write an essay, a report or a novel in a fraction of the time. But more words don’t necessarily mean more progress. In fact, many studies argue that the productivity ‘promise’ is often a myth.

Sure, you can generate content within seconds, but you need the right prompts (often multiple) before you get something passable. Then you need to check the output. Is it too wordy? Are the sources correct? Has it made anything up?

In the end, after tweaking and cross-referencing and editing – and trying to make the final product not sound like AI – you’ve sunk hours into the task, possibly more than if you’d written the thing yourself.

It masks weak arguments

There’s no doubt that AI output looks polished. But polished writing can mask weak arguments. This is particularly true if you’re using it to write about an unfamiliar topic. When you read the output, you may be unaware that its arguments are biased, clichéd, lacking in depth or even outdated.

Large language models are trained on publicly available datasets, so they won’t be making unexpected connections or introducing new ideas. They're pulling from what's already out there (good or bad – it can't differentiate). And while its output looks confident and persuasive on the surface, it lacks nuance and falls apart under scrutiny.

Let's look at an example taken from a short story that was recently flagged as AI-assisted. Here's one of the lines: 'The girl smiled like sunrise over a sink'. When you break it down, the simile doesn't make logical sense. How can a smile be like a sunrise, and why over a sink? It's just an algorithmic arrangement of words that look pleasing or even poetic at first glance but fails to impart any meaning.

Of course, this doesn't just happen with fiction. To see how this plays out in college classrooms, check out this article by Professor Wendy Belcher: 10 ways  AI is ruining your students’ writing.

It uses predictable language patterns

You can't go on social media these days without seeing someone talk about AI giveaways and how to avoid them. That's because AI outputs are saturating the market and easily recognisable.

Large language models use publicly available datasets, usually from the internet (but also from pirated books), that predict language patterns, so they repeat the same words or structures, even in the same paragraph. This means that AI-generated content can be predictable or repetitive. And while these ‘tells’ change on an almost weekly basis, here are some common ones:

• overuse of em (or en) dashes

• repetition of the rule of threes

• overuse of contrast framing (e.g. ‘It’s not X; it’s Y’)

• overuse of words or phrases like ‘delve’, ‘leverage’, ’landscape’, ‘elevate’, ‘quietly’, ‘here’s the thing’, ‘in today’s fast-paced world’

• repetition of transitional words like ‘Furthermore’, ‘In addition’

• use of consecutive sentence fragments

• use of consecutive one-sentence paragraphs

• non-sensical metaphors or similes

• generic or abstract statements.

This list is by no means complete, or a sure indication that a piece of writing was assisted by generative AI. These devices or patterns existed long before large language models were created – and they're not incorrect – but they become suspect when overused.

It makes things up

Most people know that AI has a tendency to make things up, also known as 'hallucinating'. This happens when the system doesn’t understand information or a question, or when it can’t find an answer in its dataset. Because it’s designed to comply with every request, it fills the gaps with incorrect information or sources.

Of course, if you’re an expert in your field, you’ll know if the facts presented are right or wrong. But even experts have been fooled by seemingly polished and persuasive arguments. So unless you fact-check the output, run it by a professional or write the content yourself (this is probably your best bet), it has the potential to damage your credibility and reputation.

For a list of high-profile hallucinations that had legal, financial or reputational consequences, see 8 AI hallucinations examples.

Readers hate it

If recent writing controversies are anything to go by, readers hate stories written by AI. Now that the tech has been around for a while, they can spot the signs a mile off – and they feel cheated. Not convinced? Here are some recent examples.

In March 2026, Mia Ballard’s book Shy Girl attracted publicity when it was accused of being AI-generated. Originally self-published, it was picked up by Hachette UK and published commercially, only to be dismissed by readers as AI slop. The backlash was so strong that it was pulled from sale and release of the US edition was cancelled.

A couple of months later, the Commonwealth Short Story Prize was criticised when a regional winner's story was also accused of using AI. While it wasn’t disqualified or stripped of its prize, debate was heated, and it cast doubt on the author's credentials (he had extolled the virtues of AI in a recent social media post) and the prize's reputation.

There are more examples of this kind, but you get my drift. These responses suggest that an AI backlash is gaining momentum, and the message is clear: discerning readers want stories and content shaped by personal insight, lived experience and human imagination – not an AI algorithm.

Your work may not be protected by copyright law

Laws around copyright and AI-generated copy are hazy and constantly evolving, and they can vary depending on your country or region. In Australia, there is no copyright legislation for AI-generated content – yet. So if the output is mostly AI, it's not protected by copyright law (AI doesn't have legal status, so it can't assert any moral rights). This means it's not an infringement of copyright if you publish AI-generated content, or if that work is used (stolen?) by someone else.

Does that mean you're in the clear if you use unedited outputs? Not really. Since large language models have been trained on public datasets, they may borrow heavily from an existing source (possibly one that was pirated), which may result in copyright infringement. In fact, one study showed that ChatGPT had a 45 per cent plagiarism rate, which is alarmingly high. So will you know which part was plagiarised? Sadly, no – at least not until the lawyers come knocking.

On the other hand, if there was substantial human input, such as using multiple prompts, rewriting, editing and reformatting, then you're more likely to be protected by Australian copyright law. But if queried, you may need to provide proof of human intervention, which can be tricky, particularly if you used a free version of the platform that doesn't save outputs or prompts.

Note: These days, tools like Grammarly and Microsoft Word Editor are powered by generative AI, which means that writers are unwittingly using it during grammar or spell checks. At this stage, these tools are deemed acceptable because they don't do the bulk of the work and writers can either accept or reject flagged issues. That said, some publishers are strict about how much AI can be used in a published work, so always check.

Takeaways: Why using AI could be hurting your writing

While it may be tempting to use AI to take the grunt work out of writing, the costs often outweigh the benefits. Dependence can weaken the critical thinking skills that make your writing effective and uniquely human in the first place.

If your aim is to churn out as many books as possible for a fast buck, then AI will help you do it. But if you're drawn to the craft of writing, in all its flawed and messy glory, ditch the prompts and let your own voice shine again. If recent controversies have taught us anything, it’s that readers still value the insights and experience only a human author can provide.

Got an AI-free manuscript ready for editing? Get in touch through my contact form or request a sample edit.