What Is Everybody Getting Wrong About AI?
Writers hate AI prose. They’re right to hate it. But they’re wrong about why.
This one’s personal for me. My job is writing books in other people’s voices, and voice is the exact thing AI prose is missing. Misdiagnose why the machine’s output falls flat and you’ll use it wrong, then wonder why your own pages started sounding like it.
| The reframing | What it means | What the writer brings |
|---|---|---|
| Safe, not bad | Every sentence hedges; nothing commits or lands | Risks, even stupid ones |
| No scars, not no talent | Technically fine prose with nothing underneath it | Your divorce, grief and betrayals as voice |
| Nothing to say | Knows the word grief, not the thing itself | The strange detail nobody would invent |
| Sounds like everyone | Writes to the average of its training data | Your weird angle, a voice only you have |
| Writes scared | Softens every word to avoid offense, and impact | The raw verb and the ugly noun |
| Can’t surprise itself | Predicts the most likely next word | Discovering the story while you write it |
| Too perfect | Clean, grammatical, no glorious failures | Imperfection readers feel as human |
1. It’s not bad writing. It’s safe writing.
AI doesn’t make mistakes. And that’s exactly the problem.
Every sentence hedges. “Perhaps she felt something.” “It seemed as though.” “A sense of unease washed over her.” The prose reads like a lawyer drafted it, every word chosen to avoid liability. Nothing commits. Nothing lands.
Good writing takes risks. Great writing takes stupid risks. AI never takes any risks at all. It produces competent, defensible, utterly forgettable prose. You can’t argue with it. You also can’t remember it five minutes later. I think safe writing is the fastest way to lose a reader. A reader keeps turning pages because somebody on the page took a chance and they want to see if it paid off. AI has never taken a chance in its existence.
2. It doesn’t lack talent. It lacks scars.
AI can construct a sentence. It balances rhythm, varies length, deploys semicolons correctly. The technical execution is fine. Sometimes better than fine.
But AI has never been punched in the face. Never had its heart broken at 3am while staring at a phone that won’t ring. Never sat with a dying parent and realized there’s nothing left to say. The prose comes out smooth because there’s nothing underneath it. No grit. No detail. No truth.
Your scars are your voice. The divorce, the bankruptcy, the betrayal, the grief you still can’t talk about at dinner parties. That’s the stuff readers feel in your sentences even when you’re writing about spaceships or dragons. AI has no scars. It’s training data.
3. It’s an excellent writer with nothing to say.
Ask AI to write about grief and you get “a profound sense of loss settled over her.” Accurate, I guess. Also completely empty.
AI doesn’t know what grief feels like. The bizarre laugh that escapes at the funeral. The rage at a coffee cup your mom used to drink from. The way time stops working for weeks and then suddenly you realize six months passed. AI knows the word “grief.” It knows what other writers have said about grief. It doesn’t know the thing itself.
That gap matters most in memoir. The detail that makes a reader believe a grief scene is always the strange one nobody would invent, and a model trained on everybody else’s grief reaches for the average every time. I won’t settle for the average when somebody’s real loss is on the page.
This shows up everywhere. Love scenes that describe “passion” without any sweat or awkwardness. (See my review of Naughty Words for Nice Writers) Fight scenes with “brutal efficiency” but no fear. Dialogue where everyone says exactly what they mean. AI learned fiction from fiction. It’s copies of copies, getting blurrier every generation.
4. It doesn’t sound robotic. It sounds like everyone.
People miss this. AI prose is generic.
It sounds like every mediocre novel in the airport bookstore. Every corporate blog post about overlap and moving forward. Every writing workshop story where the prose technically works and the teacher has no specific complaints and nobody remembers it a week later. AI writes to the average. It finds the center of all the text it trained on and produces more of the same.
Your job is to write away from the average. Find your weird angle. The voice that sounds like you and nobody else. AI can help you draft. It can’t help you find that voice. You have to bring it yourself.
I tested this myself in 2024. As an experiment, I used AI to write an 85,000-word novel in six days, then spent a whole day on a manual pass. I’d call it a B-plus, and it was in no way publishable, so it never went anywhere. I never use AI to write client books, and this had nothing to do with my own novels. The process was a blast, though. You talk to the AI like a conversation and tell it what’s right and what’s wrong.
5. It writes scared.
AI softens everything. “Intimate encounter” instead of sex. “Physical altercation” instead of fight. “Passed away” instead of died. “Struggled with substance issues” instead of drunk. It’s trained to avoid offense. Which means it’s trained to avoid impact. The words that hit hardest are the ones AI won’t use. The raw verb. The ugly noun. The sentence that makes a reader flinch or laugh or catch their breath.
Fiction needs edge. AI files every edge down until the thing’s smooth, round and harmless, and I don’t have much use for harmless fiction. Your readers didn’t pick up a book to feel nothing, and they’ll put it down the moment they suspect that’s all they’re getting.
6. It can’t surprise itself.
Most people don’t understand this about writing: writers discover the story while they write it. The character does something you didn’t plan. The sentence goes somewhere unexpected. You start a scene knowing the ending and halfway through you realize you were wrong. I love those moments, and they’re where the good stuff comes from. AI can’t do this. It predicts the next word based on probability. Every output is a statistical average of what came before. It can only write what’s expected. Surprise isn’t in the math.
When you write, you’re having a conversation with your own subconscious. Weird connections fire. Old memories surface. You make leaps you can’t explain. AI has no subconscious. It’s a next-token predictor. And that predictor always chooses the most likely path. That’s by definition the least surprising one.
7. It’s a perfect writer. That’s the problem.
AI produces clean, grammatical, properly structured prose every single time. No rough edges. No weird choices. No accidental brilliance. No glorious failures.
Perfect is boring. Perfect is forgettable. Your first draft full of mistakes and strange impulses has more life in it than AI’s polished output. The sentence you wrote at 2am when you were half-asleep and forgot what you were trying to say, that sentence might have magic in it. AI never writes at 2am. AI never forgets what it was trying to say.
Imperfection is human. Readers respond to human. They can feel the writer’s hand on the page, the personality bleeding through the words. AI has no personality. It’s parameters.
So why use it at all?
Because it’s a good tool when you keep it in its place. It brainstorms twenty options in seconds, handles mechanical work like summaries and transitions, and gives you something to push against when you’re stuck. Treat it like a tool and you’ll do fine. Let it do your writing for you and you’ll have problems. You’re still the writer. Whatever AI hands you comes out safe and bloodless, and it’s your job to put the scars and the edge back in until the page surprises even you. I spent the last year building a system for exactly this: using AI as a tool without letting it flatten your voice. The result is the AI-Enhanced Writer’s Series: 46 handbooks on writing fiction with AI without sounding like a robot.
My take
Machine prose fails because it’s safe. It has no scars, and readers can feel the missing hand on the page. Keep AI for what it does well, twenty options in seconds, summaries, transitions and something to push against when you’re stuck. Then do the writing yourself. Whatever it hands you, put the edge back in until the page surprises you. You’re still the writer.
My book The Death of Thinking looks at the slow cost of outsourcing judgment, including the writer who opens a document and finds the blank page harder than it used to be.
Frequently Asked Questions
The book on this: The Birth of the Augmented Human is 212 pages on what goes right when the tools get used well, written against both of the confident positions everybody else is taking.
