I use AI every day. I’ve published 113+ books and completed 54+ ghostwriting projects, and AI is part of my process for research, brainstorming, outlining, and analysis. I have also written an entire series of handbooks on AI-assisted fiction writing. That means hundreds of hours testing what AI does well and documenting where it fails.
AI won’t replace writers and won’t make you better on its own. Used carelessly it’ll make your writing worse in ways you won’t notice until a reader does, and by then it’s printed.
The hype on both sides wears me out.
The people selling AI as a replacement for writers are wrong, and so are the people who call anyone who touches it a fraud. Treat it like a tool and you’ll do fine. Let it do your work for you and your book is going to show it.
What Does AI Writing Do Well?
AI will not replace writers and will not make you better by itself.Share on X
AI is good at pattern recognition, analysis, and generating raw material you can shape. Brainstorming. Outlining. Finding structural problems in a manuscript. Checking consistency across hundreds of pages. Analyzing voice patterns. It delivers real value in all of them.
The people selling AI as a replacement for writers are wrong, and so are the people who call anyone who touches it a fraud.
When I start a ghostwriting project, AI helps me organize research, build initial outlines, and find gaps in the structure. When I coach fiction writers, it generates examples we can pull apart and improve together.
When I’m writing my own fiction, it helps me brainstorm plot alternatives when I’m stuck, check consistency across a complex series, and catch problems I’ve gone blind to after living with a manuscript too long. For more, see the AI skills that will actually matter in five years.
AI is useful for analysis and raw material. It’s not useful for final prose, voice, or creative decisions. Those remain human work. For more, see AI writing tools.
Why is AI’s confidence a problem for writers?
AI delivers wrong information with the same confidence it delivers correct information. There’s no difference in tone, certainty, or presentation between an accurate statement and a hallucination. Your brain reads confidence as a credibility signal. Thousands of years of evolution trained you to trust confident speakers, and AI exploits that without meaning to.
I have caught AI being confidently wrong about historical dates, scientific processes, legal procedures, publishing industry standards, and the plots of novels. Each time, the delivery was indistinguishable from the times it was correct.
Ask AI for academic citations and it’ll hand you author names, journal titles, publication dates, volume numbers, and page ranges that don’t exist. The format looks perfect. The citation is invented. I tested this extensively. About a third of the time the citations led nowhere. The authors hadn’t written those papers. The journals hadn’t published those issues.
The rule is simple. Verify anything factual that matters. Treat every factual claim as a starting point for verification instead of an answer. If a fact matters to your story or your reputation, check it against sources that can be wrong in ways they’d acknowledge.
How does AI flatten a writer’s voice?
This is the problem that will damage your writing most if you don’t watch for it.
AI runs to a default voice: helpful, thorough, balanced, slightly formal. That default bleeds into everything it generates. Characters who should sound distinct start sounding alike. Narrators who should have personality get smoothed into generic competence. The prose comes out technically correct and completely dead.
AI vocabulary has tells. Delve, tapestry, testament, nuanced, landscape, journey, realm, myriad, complex, resonate, convincing, deep. They appear at rates no human writer produces naturally. If those words cluster in your manuscript, AI has flattened your voice.
Fiction runs on surprise, freshness, and unexpected word choices. AI delivers predictability dressed in formal vocabulary. AI doesn’t produce a sentence like “The Miami sky was the color of a headache.” That sentence is specific, and slightly wrong in a way that feels exactly right. Human voice lives in those small wrongnesses, and AI smooths every one of them away.
The fix. Never use AI prose as final text. Use it as raw material and rewrite it in your own voice. Then search your manuscript for the tell words, and if you’re using them at AI rates, replace them.
POV Drift
AI shifts point of view mid-scene constantly. You’re writing from one character’s perspective. You ask AI to help develop the next few paragraphs. Back comes prose that includes another character’s internal thoughts, or describes something the viewpoint character can’t see, or slides into omniscient overview for no reason.
AI doesn’t hold POV discipline, because it doesn’t experience point of view the way a reader does. It generates plausible text, and plausible text includes perspective shifts that feel natural in isolation and violate the POV you established. Without constant correction, a carefully maintained viewpoint degrades into head-hopping.
I deal with this in my own fiction. My novel Peacekeeper is a complex series with multiple POV characters shifting between first person and third limited.
AI is really bad at fiction. If you use it on fiction and don’t edit the result, it’s AI, and you didn’t write it. Light editing leaves it AI too. It takes massive editing to make it yours. I got curious about how the tools that check originality decide something came from AI, and it took a lot of research to find out. They look for patterns.
AI assistance means stating the POV constraints in every prompt and reviewing every AI-touched scene for perspective violations. For more on using AI to amplify your voice, see my interview with Diana Lee.
The fix. State the POV constraint explicitly every time you work with AI. Then review every AI-assisted scene against three questions: “Can my viewpoint character perceive this? Know this? Interpret it this way?” Flag every violation and fix it before it contaminates your own sense of the character’s perspective.
When Hallucinations Build on Each Other
Hallucinations compound when you build on them. You ask AI for information. It invents a detail. You accept the detail and ask a follow-up question. It builds on the invention, generating more content that assumes the false thing is true. By the end of the conversation you have an elaborate structure standing on a foundation that doesn’t exist.
This happens especially in research sessions where one question leads to another. The first hallucination contaminates everything that follows.
I’ve watched AI generate fictional backstories for real historical figures that fit the era perfectly. Plausible details, specific events, coherent narrative. None of it happened. The general knowledge was right and the specifics were invented. AI filled the gap between what it knew and what I asked for with probable-sounding content that wasn’t true.
That one scares me for anyone writing memoir, history or biography. A reader who catches one invented detail about a real person stops trusting every true detail around it, and the author pays for a mistake the machine made.
What Can AI Writing Not Do?
AI can’t capture someone else’s voice. Ghostwriting is listening to another person, learning how they think and speak, and translating that into prose that sounds like them. Every one of my 54+ ghostwriting projects starts with long interviews where I learn how the client says it as well as what they want to say. AI has no capacity for that kind of understanding.
Creative decisions that serve a specific story are beyond AI.
It can generate options, but choosing which option serves your narrative requires understanding what your story is about at a level AI doesn’t reach. When I coach fiction writers through plot problems, the solution usually comes from understanding the character’s psychology, the thematic intent, and the reader experience the writer is building. AI can generate twenty alternative plot directions. A human has to know which one is right for this story.
No AI can replace the emotional truth that makes readers care. Writing that moves people comes from real human experience, specific detail, and the nerve to put something real on the page. AI can imitate the structure of emotional writing without ever producing the substance.
How Do You Use AI Without Wrecking Your Writing?
Use AI for analysis, research, brainstorming, and consistency checking. Don’t use it for final prose. Verify every fact that matters. Watch for voice flattening. Maintain POV discipline manually. Treat AI output as raw material and nothing more.
Writers who skip those steps and hand the machine the final draft are putting out books that sound like everybody else’s, under their own names. I think that throws away whatever made them want to write in the first place. Use the tool hard, and keep the writing for yourself.
I’ve written three books about these issues. Using AI for Writing covers how to integrate AI into your writing process. AI Shortcomings documents every failure mode I’ve encountered and how to defend against them. Both are free. The AI-Enhanced Awful Writing Handbook covers how to identify and fix the problems AI introduces into prose, including the vocabulary tells, voice flattening, and structural patterns that mark AI-damaged writing.
My full collection of AI-enhanced writing handbooks covers every aspect of fiction craft with AI-assisted techniques, from character development to world-building to dialogue to point of view.
For ghostwriting or book coaching, start with a conversation.
My take
AI does some jobs well and will wreck your book on others. Keep it on research, brainstorming and consistency checks, and keep it away from your final prose. Verify every fact, because one invented detail about a real person costs you the reader’s trust in every true one. Hold point of view yourself. The writers handing the machine their final draft are publishing books that sound like everybody else’s under their own names. Keep the writing for yourself.
AI and Writing FAQ
AI used well amplifies thinking instead of replacing it. The Birth of the Augmented Human is my book about that path and what you gain when you keep the thinking for yourself.
One prompt I recommend for the final front-to-back review asks whether the book promised anything at the beginning that it never delivered. A book makes a lot of promises at the start, and on a long project some threads get dropped. Without something like a Kanban board, it’s virtually impossible to catch them all, and the ones you miss are fun to see come out in Amazon reviews. To me, that’s the appropriate use for AI in writing. Don’t let it write the damn thing for you.
