AI Writes. I Bleed

When I came across a writer’s online rant about his original work being flagged as 60% AI-generated, I thought he must be a bad writer. Not because the detectors are infallible, but because I’ve come to believe AI itself is a bad writer. Clean, articulate, even clever, but insipid. It writes with zero soul. So, if a human-written piece gets flagged as AI-generated, it likely feels lifeless. But let’s not conclude yet.

I dug into my archives and ran some of my best pre-AI pieces through detectors. They were the kind of essays that carried story, tension, the pulse of real experience. Even ZeroGPT, one the most paranoid of the tools, returned them as 100% human. To stretch the theory, I fed the abstract and introduction of a research paper, arguably the most robotic parts of any academic work, into ZeroGPT. It, too, came back as 100% human. So did the most rigid, clunky sections of my old writings, back when I was still learning to write well. This wasn’t the result I expected, but it made me pause.

More than confirm my hypothesis, these results, paired with a fascinating piece by Michael Todasco, titled “AI Detection Arms Race: Judging the AI Writing Detectors,” left me quietly awed by how reliable these tools have become. Todasco ran his own tests using a mix of AI and human-written stories. While the detectors were good at flagging older AI outputs, they stumbled with newer, more sophisticated ones. Tools like GPTZero and Copyleaks sometimes let highlypolished machine texts slip through. But interestingly, when Todasco fine-tuned his human-written stories to sound “more professional,” the detectors still recognised them as human.

What this means is simple, and also a little poetic: a machine-generated piece, if infused with soul by a human, through careful editing, storytelling, emotional intelligence, can pass detection. Meanwhile, a lifeless, formulaic piece written by a human might get flagged. Authorship, it turns out, is less important to the detectors than aliveness. They don’t care whether it was written by a person or a prompt; they care whether it feels like someone was there.

The paradox, though, is that while AI may seem to be redefining creativity, it might actually be returning us to the original concept of creation. In many religions, creation is layered and delegated. God created man, but often through the hands of another. In Yoruba mythology, it was Obatala who was tasked with moulding the human form. He shaped the head, the limbs, the body. But it was Olodumare who breathed life into that form. Without His breath, the sculpture was just clay. It was the breath that made it human. That’s where we are with AI and writing. Machines can mould sentences, structure thoughts, even imitate style. But they cannot breathe. That part – the life, the presence, the creation of feeling – still belongs to us.

AI is eloquent because it has access to more words than any native speaker will ever hear in a lifetime. It doesn’t struggle to recall the exact word for an emotion, the precise verb for a gesture, or the rare synonym that ties a sentence together. A language model like ChatGPT isn’t one mind but an aggregate of countless minds, ideas, styles and tones. That’s why ittalks too much, like a student trying to impress the teacher with words. Paragraph after paragraph of polished emptiness. But that’s also why we aren’t winning the fight against it.

When it comes to surface-level brilliance, AI has the advantage. Its grammar is flawless. Its sentences land. Its metaphors, even when hollow, are impressively structured. So, if you ignore  it, as some purists and preservationists have suggested, you’re already out of the competition. Because while you’re busy romanticising the struggle, others are learning how to blend their voice with the machine’s precision. There’s nothing noble in pretending the tide doesn’t exist.

The solution is collaboration – not prompting. There’s a difference. Prompting is passive. Prompting is asking the machine to create so you don’t have to. But collaboration is active. It means letting AI do everything it can while you do everything it can’t. AI can give scaffolding, vocabulary, rhythm and tone. It can build, arrange and tighten. But only I can feel. Only I know what haunts me at night, what breaks my voice mid-sentence, what memory I haven’t written down because I still don’t have the strength. AI can tell, but it cannot show. Because showing – true showing – is the product of experience. You cannot describe what you have not wrestled with. You can’t summon imagery from a place you’ve never emotionally been. What you can’t think about, you can’t show. And what you can’t feel, you can’t write with depth. That’s the work it can’t do, the work I must do. That’s collaboration.

One of the biggest mistakes people make with AI writing is trying to ‘humanise’ it. As if adding a little slang, a dash of personality or a dramatic opener suddenly gives it soul. But the truth is, trying to make AI text sound human often reveals just how not human it is. It’s like putting blush on a mannequin. It might look closer to life, but it will never live.

Humanising AI text is not the same as giving it life. What we call “human” in writing isn’t tone or relatability or even voice. It’s presence. It’s risk. It’s  memory. It’s contradiction. It’s the tremble in the sentence where the writer clearly doesn’t know how to go on but does anyway. AI can mimic coherence, but it cannot mimic struggle. This is why AI-generated articles often fail when they try to feel real. They’re clean, yes. They’re informative. But they never bleed. When people try to humanise AI writing, they’re usually just trying to fix a lifeless piece without facing the harder truth: it has no pulse because no one felt anything while making it.

So, don’t humanise AI texts. Don’t dress them up and parade them as your own. Instead, start with what only you can feel and then let AI help you say it better. Let it serve your truth, not disguise its absence.

When AI detectors flag part of your writing as AI-generated, it’s probably because it’s lifeless. But the answer isn’t pressing the ‘humanise’ button. The answer is to rethink. To ask: what am I really trying to say here? What part of me is absent from this paragraph? Because rewriting, not rephrasing, is the only real solution.

Damilola Ayeni is an award-winning creative writer, journalist and editor. He is a Black Orpheus Fellow exploring dissent and racial diversity through the eye of one of Africa’s earliest literary journals. He is the former editor of the Foundation for Investigative Journalism, and his work focuses on storytelling at the intersection of memory, justice and technology.