Can I just have AI write my non-fiction book?
You can, and you’ll be able to tell, and so will your readers. AI is genuinely useful for the work around the writing: organizing research, building a first-pass outline, summarizing source material, suggesting questions you hadn’t thought to ask. It is not good at producing a finished non-fiction book a reader will trust and enjoy. It invents facts, it writes in a flat sameness that gives itself away, and it has no judgment about what actually matters to your reader. Used as a tool under a writer who knows its failure modes, AI earns its place. Used as the writer, it produces something competent, generic, and quietly wrong in places you won’t catch unless you already know better. For the publishing and copyright side, whether an AI-assisted book can be copyrighted and what the platforms require, see my
AI and Your Book FAQ.
Why does AI hallucinate facts, and why does that matter for non-fiction?
An AI model doesn’t know things. It predicts the next likely word based on patterns, which means it will produce a confident, fluent, completely false statement as readily as a true one. It invents statistics. It attributes quotes to people who never said them. It generates citations to studies and books that don’t exist, complete with plausible authors and dates. For fiction this is harmless. For non-fiction it’s fatal, because your entire credibility rests on the facts being right. One invented statistic that a reader catches makes them doubt everything else in the book. This is exactly why AI-drafted non-fiction has to be fact-checked claim by claim against primary sources, not skimmed for plausibility. Plausible is what AI is best at, and plausible is not the same as true.
What is AI drift, and how does it hurt a book?
Drift is what happens across a long manuscript when the model slowly loses the thread. Chapter one establishes a clear argument and a consistent set of terms. By chapter seven, the definitions have shifted slightly, the throughline has gone soft, and the book is arguing something subtly different from where it started, without ever announcing the change. AI does this because it has no real model of the book as a whole. It handles each stretch locally and can’t hold the full structure of an argument in view the way a human author does. Catching drift means reading the whole manuscript as one piece and checking that the spine holds from first page to last. That’s editorial work, and it’s one of the first things I check when I clean up an AI-assisted draft.
Why do AI chapters all feel the same?
Because the model falls into one internal template and repeats it. Open with a broad framing statement. State the point. Give three supporting subpoints, often in a tidy list. Add an example. Close with a tidy restatement. Every chapter built on the same skeleton, so the book develops a mechanical rhythm a reader feels even if they can’t name it. Real non-fiction varies its structure to fit the material: some chapters are a single sustained argument, some are a story, some are a walk through a process, some are short and sharp. The sameness is one of the clearest tells that a machine drafted the chapters, and fixing it means rebuilding chapters so their shape follows what they’re actually doing, not a default mold.
What are the specific tells that AI wrote something?
There’s a recognizable fingerprint once you know it. Sentences that open with gerunds used as filler: “Understanding your audience is key,” “Building trust takes time.” The “not just X but Y” and “not only X but also Y” constructions, over and over. Throat-clearing hedges: “it’s important to note that,” “it’s worth mentioning,” “at the end of the day,” “in today’s world.” Triads everywhere, three items in every list, three adjectives per noun, because the model loves the rhythm of three. Business jargon standing in for plain words: leverage, streamline, utilize, robust, holistic, seamless, delve, tapestry, journey. And a steady drift into passive voice that drains the action out of sentences. Individually any one of these is just a writing habit. Stacked together at machine density, page after page, they’re a signature.
What’s wrong with the way AI builds sentences?
AI writes to a rhythm rather than to a meaning. You get the relentless “this, then this, then this” cadence where every sentence is the same length and shape, so the prose flatlines. You get triple constructions stacked on triple constructions, “clear, concise, and compelling,” because three sounds complete to the model. You get passive voice that hides who did what: “mistakes were made,” “results can be achieved.” And you get hedging that softens every claim into mush so nothing lands with conviction. Good non-fiction prose varies its sentence length, commits to active voice, says things plainly, and lets a strong claim stand without three qualifiers around it. The fix isn’t subtle once you can hear the difference, but you have to be able to hear it, and most people drafting with AI can’t yet.
What about AI detectors like Originality.ai and GPTZero?
Treat them as a weak signal, never a verdict. The irony is built in: these tools use AI to detect AI, and they are wildly inaccurate in both directions. They flag human writing as machine-written and pass machine writing as human, routinely. A positive result does not prove a person used AI, and it should never be used to condemn a piece of work on its own. Plenty of people write in ways that happen to trip these detectors: clean, plain, well-organized prose, or writers who learned in a structured corporate or academic setting, or simply anyone whose natural style is tidy. Non-native English speakers get falsely flagged at especially high rates. There is some use in running a detector as one input among many, a prompt to look closer at a passage, but the moment you treat the score as proof you are trusting a tool that is wrong often enough to ruin someone unfairly. The reliable read on whether something was machine-drafted comes from a human who knows the actual tells and checks the facts, not from a number a detector spits out.
Can you tell when a manuscript was written by AI, and can you fix it?
Yes to both, and it’s become a real part of the work. I run non-fiction manuscripts through a forensic pass that catches the patterns: the jargon, the gerund openers, the “not just X but Y” tic, the hedges and throat-clearing, the passive voice, the triads, the structural sameness across chapters, and the factual inconsistencies that drift produces. Then I rewrite to put a human voice back in: plain words for jargon, active voice, varied sentence and chapter structure, claims stated with conviction, and every fact checked rather than trusted. The goal isn’t to disguise that AI touched the draft. It’s to produce a book that reads like a person wrote it because, in every way that matters to the reader, a person did. If you have a draft that came out of AI and feels off but you can’t say why, that off feeling is the fingerprint, and it’s fixable.