AI Is Already Changing How You Write — Even If You Don't Use It

Even if you never touch AI, AI is going to change your writing.

I realised this recently, and it stopped me in my tracks. Because the conversation around AI and writing has, from where I sit, been based around ethics. You either use it or you don't. You're for it or against it. Then there’s that pesky little issue of AI companies using author’s work to train their models…

But what nobody seems to be talking about is this: the mere existence of generative AI is already reshaping how authors write, edit, and submit their work. And that's regardless of which side of the fence you're on. Or if you straddle it.

Let me explain.

The new edit

If you're a writer who submits work — to publishers, literary magazines, agents, or writing competitions — your manuscript is increasingly likely to be screened by an AI detection tool.

Clarkesworld, a respected science fiction magazine, was so overwhelmed by AI-generated submissions in early 2023 that it had to shut its submissions portal down. Editor Neil Clarke reported receiving five hundred machine-written stories in under twenty days, on top of seven hundred legit ones. Other magazines confirmed the same pattern. How they identified these submissions is worth a conversation on its own, but let's put that aside for now.

Since then, many publications and competitions have been adopting AI detection measures. That adds an extra step for authors. Before you hit send, many of you will want to run your manuscript through an AI detector yourself. Not to check whether you used AI, because you know. No, you'll be checking to see whether the tool thinks you did.

And if something gets flagged? You'll change it.

Am I wrong? I don't think so.

You won't change it because you think the writing is bad. Not because you found a better way to say it. You'll change it because a piece of software told you it looked too much like something a machine might produce.

That's a brand new kind of edit. And it has absolutely nothing to do with improving your craft.

The false positive problem

Here's where it gets worse, and where I’m circling back to the earlier point.

AI detectors are not reliable. Multiple studies have shown significant false positive rates, meaning human-written text is incorrectly flagged as AI-generated.

Researchers at the University of Maryland, led by computer science professor Soheil Feizi, studied several popular detection tools and found that the majority had high false positive rates and could be easily evaded with simple paraphrasing. I can’t say that I’m as confident about the ‘simple paraphrasing’ part of those findings now, just three years later. But Feizi's conclusion was blunt and I agree: current detectors are not ready to be used in practice.

OpenAI itself shut down its own detection tool in July 2023 after it managed to correctly identify only twenty-six percent of AI-written text while incorrectly flagging nine percent of human writing as AI-generated. That’s a concern.

And it's not just researchers raising red flags. UCLA declined to adopt Turnitin's AI detection software, citing concerns about accuracy and false positives. Many other University of California campuses did the same. When a university that well known looks at these tools and says, 'We don't trust them enough to use,' I’m sure that concerns many publishers and competition organisers using those same tools.

Anyone who writes clean, structured prose, anyone whose style happens to be direct and economical, could find their work flagged.

Think about what that means for a moment.

The chilling effect

But this is the part of it all that has really shocked me. Realising that we're talking about something now that will have a chilling and changing effect on writing style itself.

If certain sentence structures trigger AI detectors, like shorter sentences, common transitions, and clean and direct prose, then writers who naturally write that way have a problem. They’ll either keep writing in their natural voice and risk being flagged, or they’ll adjust their style to avoid detection.

My guess is most would go with the second option so as not to lose opportunities. And since I began thinking about AI about a month ago (I was really in my own little world prior to that), I've been doing my own testing. I've run AI-written pieces, AI-human combined pieces, and purely human pieces through several detection programs. What I found was sometimes the tools got it right. But sometimes they said AI was human and human was AI. They highlight sentences too, and it's a truly disconcerting feeling indeed to see lines you wrote — lines that came straight from your own head — flagged as 'definitely' AI-generated.

We all know in the publishing industry that it’s important to develop your voice. Your voice is the product of everything you've read and lived, and every draft you've felt like banging your head on the desk over, yet you persevered and made it through.

What happens to that voice when you have to run your work through a program, and if it tells you something sounds like AI, you set about changing it? Diluting and distorting, not to serve the story, but to satisfy a detection algorithm. Maybe removing quirks that were worth keeping.

It gets thornier still when you consider competitions and publishing opportunities. A growing number of writing competitions now explicitly state that entries may be screened with AI detection software, and some reserve the right to disqualify flagged work. Fair enough, too. Really. I do understand it.

But … the stakes aren't hypothetical. A false positive can cost someone a prize or even a publication opportunity that they may have worked years for. My understanding at this stage is there isn't always an appeals process, and I’ve heard talk that sometimes querying authors won't even reach the hands of a human agent or publisher if their work doesn't pass the detection tool first. I’d love to know more about this from people who have more of an insiders view than I have.

The irony

There's a painful irony here, and I think it's worth giving some deep thought to.

My worry is that aside from those who never apply for competitions or publish their work, ALL writers, whether they use AI or not, will at some point soon be forced to think about it. Those who don't use AI will find themselves second-guessing whether their natural voice sounds 'too clean' or 'too predictable.' Most will do extra rounds of edits. The AI edits. We'll add them to our structural edits and our rounds and rounds of copy editing.

Only the AI edits aren't structural. They won't fix plot holes or improve quality. They exist to get your work through a flawed algorithm, to make it seem as though a human wrote it. Whether they did or not.

And there's a painful twist to this arms race. Writers who do use AI have access to tools called 'humanisers,' designed specifically to make AI-generated text sound more human. I grant you that, like the detection programs, the humanisers are choppy at best. Yet the fact remains that AI users do have tools that might help bypass the detectors while human writers are left trying to prove they're not machines.

To see just how flimsy the detection really is, consider this: when Times Higher Education tested Turnitin by simply asking ChatGPT to write like a teenager, the detection rate dropped from one hundred percent to zero. One prompt tweak. That's all it took. If detection tools can, even sometimes, be fooled that easily, it’s disturbing to think of human writers bending their voices to satisfy it.

So what do we do?

I don't have a neat answer. I wish I did.

AI is getting better at writing. Detectors have to keep up. Humaniser tools exist specifically to defeat those detectors. It's an arms race, and the writers trying hardest to stay out of it will often get caught in the crossfire.

I think the first step is recognising that this is happening, because what I mainly see in the writing community is a debate on ethics. 'We shouldn't use it at all.' Or, 'These are the elements it seems okay to use, others not so much.'

Those are valid conversations. But I want to hear what we're going to do about the fact that AI has begun to change, and will increasingly change, the work and voice of writers now and in future.

I can see some steps towards fairness, like giving authors accused of using AI the right of reply for a start, and insisting on transparent appeals processes. But I can't see a way to stop AI from influencing how authors write.

That realisation has hit me so hard. I suspect it might be the same for you? And if it isn't, I'd love to hear why. Because this feels like a conversation the writing community needs to be having loudly, and I'd rather be wrong about it than right and silent.

Are you already editing differently because of AI detectors? Have you been flagged? I want to hear from you.

Morgan xx