On Writing Too Strange, Too Much Like Myself

This seems to be a real worry for non-native speakers and writers, and for neurodivergent people too. What if you're both? You end up having to explain that English isn't your first language and that you're neurodivergent (hello, fellow ADHDers!) so people believe you wrote your own work. Someone said it to me yesterday, and I told them I was sad that I have to write so I don't sound like AI.

I'm still sad about it, honestly.

Before all this (pre-AI), people would tell me my writing had rhythms. Habits. Certain words I leaned on, certain shapes my sentences liked, the odd metaphor people said they'd never have thought of. That was a good thing back then! It was called having a voice. Years ago one of my MA tutors told me never to lose my voice. I still think about that.

Now the same habits get read as "tells".

Elise Gomes, a Brazilian designer who's neurodivergent and writes in her second language, wrote a piece called The AI That Isn't about being told that she writes like a robot. Detectors have flagged things she wrote twenty years ago, long before these tools existed. People reply to her work emails asking to speak to a real person. Her point is that the tools are built around what "normal" writing is supposed to look like, and some of us just don't write that way.

I've found research on this too. A 2023 Stanford study ran essays by non-native English speakers through popular AI detectors, and more than half got flagged as machine-written. Essays by American schoolkids mostly passed. Why? When you write in a second language, you reach for the words you're sure of, the structures you know work. Which, funnily enough, is exactly what makes text look like AI. 

Jane Quiani’s article Whose Writing Gets Recognised as Human? explains why AI detectors miss the mark. They run statistical checks on sentence lengths and word choices. Anyone writing in structured English (especially non-native or neurodivergent writers) ends up scoring high on those exact metrics. The system ends up treating straightforward human writing as if it came from a prompt.

What gets me is the irony of it. You've got people furious about AI taking over human writing, and to catch those who supposedly use it, they use... AI. They feed a piece of writing to a machine to see a percentage pop up; they let a bot do their thinking for them.

And if your first language is Arabic, there's another layer. Arabic writing loves parallelism. We're taught to balance one clause against another, to repeat a structure for emphasis, to pair two words that mean nearly the same thing because saying it twice makes it stronger. Good Arabic prose has a rhythm you can hear, and some of it actually rhymes. Sentences run long, stitched together with wa, which means "and," over and over. At school, that's the kind of writing that got praised. 

A linguist called Robert Kaplan noticed this back in 1966, when he found that Arabic-speaking students writing in English built their paragraphs out of parallel pairs. Nobody called it suspicious then; it was just their first language shaping their second.

It goes further than that. Arabic rhetoric treats antithesis, the "not this, but that" pattern I've been told to cut, as a mark of really good writing. Students learn it on purpose. Arabic also lets a sentence stand without a verb, so what English calls a fragment can feel perfectly whole to me. Then there's the cognate accusative, where a verb gets paired with a noun from the same root for emphasis. In Arabic you'd literally say "he smiled a smile." Bring that into English and you get "she laughed a bitter laugh," exactly the sort of line people now squint at. 

And metaphors. Arabic is full of them. We start learning simile and metaphor in Arabic class at school, and classical poetry compares things you'd never think to put side by side. Zuhayr, one of the oldest and most famous Arab poets, said death stumbles around like a night-blind camel: whoever it hits dies, whoever it misses gets to grow old.

People talk like that too. In Egyptian Arabic, if someone's funny, their "blood is light". If someone cooks for you, you tell them "may your hands stay safe" (which is just how you say thank you). If someone's "heart is white," they're kind and hold no grudges. And if something's never going to happen, "it'll happen in apricot season". Put any of that into English and it sounds a bit odd. That's how I grew up talking.

Now look at the usual list of AI tells: balanced pairs, repeated structures, sentences that start with And, metaphors and similes that come out of nowhere. It's pretty much a description of how I learnt to write.

So if you're non-native and neurodivergent, you write in patterns because that's how your brain works, and you write differently because it's not your first language. Both apparently read as machine.

The list

Here's what I should bear in mind:

• Fewer negations, or none at all.
• Vary your sentence lengths. (But not too much, apparently, because that's a tell as well.)
• As few metaphors and similes as possible.
• No em dashes. Ever. 

Every one of these is also just writing. AI learned these moves from us; it hoovered up millions of human books; and now we're being told not to write like the books it learned from.

Today I'm second-guessing my sentences before I've even finished them. Every "not this, but that" makes me flinch, and half my brain is busy checking whether a sentence sounds like a machine. It's a strange place to write from. Horror's meant to be about the thing in the dark, not a comment section.

It's got me doubting bigger things too. I've got a novel I've been working on for years, and I'm not sure anymore whether to send it to traditional publishers, or whether I'll just get accused of using AI. Maybe self-publishing is the way to go? I don't know yet.

"I think it's AI" isn't the same as "I didn't like the story". You can dislike a story. You can find the style irritating or the ending flat. Say that. A writer can take it, and some of it might even help. But once someone throws "AI" at you, it follows you into the next draft and the one after... It makes you doubt sentences you were proud of. It can do real damage to your confidence and to your mental health. 

For now, first drafts are mine. I'm writing them the way I always have, negations and fragments and em dashes and all. Then I'll edit with a checklist. Count the fragments. Hunt down the dashes. Swap the words. Change a sentence. Cut the negations.

If you're a non-native writer, or neurodivergent, or both, and you've been told your writing "has AI tells," what do you do? Ignore it and carry on? Edit around it? Stop reading the comments? Give up?

Because following a list of rules makes me wonder:

Am I now turning into a bot?

All stacked negations, em dashes and extra fragments have been removed from this post. I've left the commas... I think.I'm also slowly going back through my older posts and editing out the obvious "tells." I'm not happy about it, but I suppose that's just how things are now.

Update - 8 October:

I'm not going back to rewrite my old posts. I'm not stripping out my own words to fit someone else's idea of how I'm supposed to sound.

Trying to change my sentences so I don't "sound like a bot" is the very thing that would turn me into one. Writing is the one place I don't have to pretend, where my mind runs unfiltered; and that's the point of it. If my natural rhythm sets off someone's algorithm, that's their problem. 

I'm keeping my voice.