Ask a model for an example to illustrate a point and it’ll give you one. A mid-sized company, a plausible problem, a sensible resolution.
It reads fine. It’s also a composite of every case study ever written, and a reader who has read three business books recognizes the shape within two sentences.
The reason is simple and it doesn’t go away with better prompting. Nothing that happened to you is in the training data unless somebody wrote it down, and almost nothing that happened to you was written down.
Of everything people ask a model to do, this request does the most damage. A model can’t know what happened to you. When it hands you a story anyway, it’s handing you a fake, and anybody who prints that under their own name is cheating the reader who bought the book to hear from them.
What is missing from a generated example?
The detail nobody would have chosen. Real stories carry things that don’t serve the argument. The chair was uncomfortable. It was a microwave. The brick was propping the door for smoke breaks. None of those are the point, and their presence makes a reader believe the rest. A generated example contains only essential detail, because it was assembled to make a point. Everything in it is relevant, which in a real account never happens, and the uniform relevance is the tell.
A composite feels thin even when the reader can’t say why, because the stray details that prove somebody was there have been stripped out. It makes me wince when an author who really lived through something lets a machine describe a tidier version of it. They’ve thrown away the only part a reader would’ve believed, and most of them can’t tell what’s gone.
The book on this: Family Cybersecurity is 231 pages of things that happened to specific households, the part no model can supply.
Why does specificity matter this much in writing?
Because it’s the only thing a reader can use to judge whether you’ve been there.
Claims are cheap and every book in the category makes similar ones. What separates them is whether the author can produce the particular. A number, a name, a specific afternoon. Those can’t be reasoned to. Either you were present or you weren’t.
I can tell you that most home networks are poorly secured. So can anybody. What I can also tell you, and what went into Family Cybersecurity, is that I drove through a suburban neighborhood with a laptop and found 127 networks on default passwords and 43 with none in about twenty minutes.
The second version makes the same argument with proof of presence attached, and that proof is the reason a reader keeps going. An author with an afternoon like that in their past and a generic sentence in their manuscript is leaving their best evidence on the table.
The book on this: Behind the Wire couldn’t have been generated by anything, because it came out of a journal a man wrote after he got home.
Can you prompt your way around it?
Only by supplying the story yourself, and at that point the model is only formatting what you gave it. That’s a legitimate and useful thing to do. Give it the raw account, badly written, with the details in the wrong order, and ask it to tighten the structure. The result can be better than your draft. What you can’t do is ask it to supply the account. Any story it produces will be a synthesis, and a synthesis presented as personal experience is a fabrication with your name on it, a considerably larger problem than a flat chapter. I’d hold that line hard. Your story in the tool’s formatting is fine by me. The tool’s story under your byline lies to the people who trusted you enough to buy the book, and I won’t write one of those for anybody.
Are composite examples acceptable in business books?
They have a place, as long as the book says they’re composites.
Plenty of legitimate books use composites for confidentiality. A consultant who can’t name a client builds an illustrative case from several, says so, and the reader adjusts their expectations accordingly.
What fails is an undeclared composite presented with the detail of a real event. The reader assumes it happened, then finds out it didn’t, and every other claim in the book becomes suspect.
If you can’t use the real case, say the case is constructed. Admitting it costs you a sentence, and getting caught costs you the whole book. Authors who gamble on nobody checking are making a dumb bet, because readers do check, and the ones who catch it tell other readers.
The book on this: The Ghostwriting Advantage describes the interview process that gets these stories out through conversation.
How do you get your own stories out?
By being asked. That’s why interviews work and blank pages don’t.
Authors who gamble on nobody checking are making a dumb bet, because readers do check, and the ones who catch it tell other readers.
Memory doesn’t work by retrieval on demand. A question about a subject produces the general version. A question about a specific occasion produces the details, and a follow-up question produces the ones you’d forgotten you knew. That’s a conversational process and a model can’t start it, because it doesn’t know what to ask you about. It has no idea that something happened at a housewarming party in 2011, so it can’t ask about the ice.
The practical version for somebody writing alone: record yourself telling the story to a person. Don’t write it. Tell it. The details that arrive when somebody is listening aren’t the ones that arrive at a keyboard.
My take
Generated stories under your name are fabrications, and I won’t write one for anybody. Readers spot the composite client with the tidy problem within a paragraph. Then they doubt everything else you say. Hand the tool your own account, messy and out of order, and let it tighten the structure. Structure is a fine job for it. If you use composites for confidentiality, say so in the book. Undeclared ones cost you the reader’s trust.
What should the tool be doing instead?
Working on the material after you’ve supplied it. Tightening a rambling account. Suggesting where in a chapter a story would land best. Finding the three places you told the same anecdote in slightly different words. Producing a first structural pass so you can see the shape and rearrange it.
All of that is real work and it saves real time. None of it requires the tool to have been anywhere.
| Task | Verdict | Why |
|---|---|---|
| Tightening a rambling account you supplied | Fine | The model is formatting what you gave it |
| Suggesting where a story lands in a chapter | Fine | It works on material you supplied |
| Finding where you told the same anecdote twice | Fine | Real work that saves real time |
| A first structural pass | Fine | You can see the shape and rearrange it |
| Supplying your personal story | No | A synthesis presented as experience is a fabrication with your name on it |
| Composites declared as composites | Acceptable | Readers adjust their expectations |
| Undeclared composites | No | Getting caught makes every other claim suspect |
The neighboring pieces cover the other gaps: AI never writes in your voice and the labor split that works on a book. The AI and Writing Hub has the rest, and my ghostwriting service is mostly six interview sessions and a lot of follow-up questions, for exactly this reason.
A book built on invented stories is a counterfeit, and I think the people selling AI as a way to skip the remembering are doing real harm to the authors who believe them. Your life is the one thing in the book nobody else can supply. Spend the hours getting it out of your head and onto the page, and let the tool clean up after you.
Your real stories are the part no model can invent. The Ghostwriting Advantage explains how a ghostwriter helps you get those stories into a finished book written for you.
