In this guide
The question “what should my novel be about?” is a bad question, and idea generators answer it badly. They produce premises that are plausible, familiar and weightless: a detective with a secret, a kingdom on the edge of war, a woman who inherits a house with a history.
Each of those could be a book. None of them is a book yet.
The writers who get somewhere start from the other end. They notice something — a person sitting in a fountain they should not be in, a warning on a toothpaste tube, a tiny bone on the kitchen floor — and ask what would have to be true for that to happen. The idea comes from a question they cannot stop answering.
A generator can help with the next step. It cannot supply the question.
There is a second kind of book that needs its own version of this advice. If you are writing from experience rather than imagination — a coaching manual, a business guide, a memoir of a trade — your idea is not a what-if. It is an angle into material you already own. The same rule applies: the generator expands; it does not originate.
What generators are actually good at
Once you have a what-if, a generator is useful for expansion under constraint:
- World rules. What has to be true for this situation to exist, and what does that cost the people in it?
- Cast. Who benefits from the rule, who is crushed by it, who is pretending not to notice?
- Conflict. What does the protagonist want, and what would they have to give up to get it?
- Variations. Ten directions, so you can pick one instead of accepting the first.
For a practical book, the expansion looks different but works the same way:
- Scope. Which part of your experience is the book about, and which parts are a different book?
- Reader. Who is this for — the beginner, the peer, or the person who has to teach it?
- Obstacle. What do they try today, and why does it fail them?
- Promises. What will they be able to do by the end, and can you prove each one?
What the generator is bad at is deciding which of the ten is worth six months of your life. That decision needs a stake the model does not have.
That is also why “AI wrote a novel in an hour” lands so badly with readers. The complaint is not about the tool. It is that a book made without investment reads like it: if you did not care enough to spend the time, why should I care enough to read it?
The three-constraint test
Before you generate anything, answer three questions in one sentence each.
- Who wants what? Not a trait. A want with a deadline.
- What blocks them? An obstacle that would also block a reasonable person.
- What does failure cost? Something the protagonist cannot shrug off.
A premise that passes all three has a spine. A premise that fails one is a mood.
| Book | Who wants what | Obstacle | Cost of failure |
|---|---|---|---|
| Fiction: a woman inherits a haunted house | Unclear | Ghosts | Unclear |
| Fiction: Mara has nine days to get her brother’s name off the lease before an audit | Clear | The landlord already knows, and wants something | Her brother loses the only home he has |
| Practical: a coach writes a shooting manual for badly taught teenagers | A parent wants their kid to stop getting benched for a fixable flaw | Online advice is generic and contradictory | Another season on the bench, and a kid who quits |
The second and third premises are not more original. They are more playable. Every chapter can be about the deadline, the landlord’s price or the brother’s ignorance — or, in the manual, about one specific flaw, one drill and one case that proves it.
The thirty-chapter test
A novel is not one idea. It is thirty to sixty changes that escalate.
Take the premise and try to write thirty chapter goals, one line each, as a change:
- Topic: “Mara talks to the landlord.”
- Change: “Mara trades the deposit for a week and learns the landlord already knows about the audit.”
If you can reach thirty and the pressure keeps rising, the idea can carry a book at a workable chapter length. At 1,200 words a chapter, that is roughly 36,000 words; a 70,000-word novel is about 58 chapters. The plan itself can run from 3 to 60 units, so you are not committing to the whole book at once — but you should be able to see the whole arc.
For a practical book, the test becomes: can you list thirty things the reader will be able to do, each one harder than the last? If you run out at eight, you have an article or a short guide, not a book.
If you stall, you have three options:
- Shrink the unit. It is a short story, a pamphlet or a long article.
- Add a second pressure. A second audience, a second constraint, a second situation the advice must survive.
- Change the reader. The person you are writing for may be the wrong person for this material.
Do not solve a thin premise by generating more prose. Prose cannot add a spine.
Lock the premise before you draft
Once you choose the idea, write it down as things the book must honor:
- The premise. One paragraph: who the book is for, what changes for them, what it costs them to stay where they are.
- The cast or cases. Five facts per principal person, client or example.
- The rules. Three things that are true in this book and cannot be broken for convenience.
- The timeline or scope. The zero point, the deadline, the boundary of what the book covers.
- The ending’s promise. What the reader is waiting for, stated plainly enough that you can tell whether you paid it.
In our editor, those go into Global knowledge as facts, and every chapter is written against them. The rough notes behind each case can sit in a knowledge base mounted to the book, so the chapter can quote the real numbers instead of inventing plausible ones.
That is what stops the book from drifting into a different, more average idea by chapter twenty. It also keeps the book honest: if a case is not in your notes, the right move is to leave a gap or go find the file, not to let the model fill it with something that sounds true.
Quality checks that do not require taste
You do not need to know whether the idea is brilliant. You need to know whether it is workable. Six checks:
- Agency: Can the protagonist change the outcome, or are they only reacting? For a practical book: can the reader act on the advice tomorrow?
- Pressure: Does each chapter goal make the next one harder?
- Specificity: Could this book happen to anyone, anywhere? If yes, it is a shelf, not a premise.
- Cost: Is there something the reader loses by not acting?
- Scenes or cases: Can you name three moments you are excited to write? If not, the idea may be an essay.
- Reader promise: Can you say in one sentence what the reader gets? If not, neither can the cover.
A short workflow
- Notice a what-if — or a piece of your experience that keeps coming up in conversation.
- Write it in one line.
- Answer the three constraints.
- Ask a generator for ten expansions: scope, reader, obstacle, promises, and three possible endings or structures.
- Pick one. Discard the rest without regret.
- Run the thirty-chapter test. Adjust the premise until the goals escalate.
- Lock the premise, cases, rules and scope as facts; mount the notes the book may use.
- Draft the first three chapters and read them together before continuing.
The bottom line
A novel idea generator is a useful amplifier and a poor origin. Bring the what-if, the person, the specific small wrongness you noticed — or the experience you cannot stop explaining. Then use the generator to expand it under constraints and test it with thirty chapters. The idea that survives is not the cleverest one. It is the one you can keep writing when the first ten chapters are done.
Key takeaways
- 1Bring the what-if or the angle; use the generator to expand, not to originate.
- 2Three constraints — want, obstacle, cost — separate a premise from a mood.
- 3The thirty-chapter test proves an idea can carry a book, in fiction or in a guide.
- 4Lock premise, cases, rules and scope as facts before drafting.
- 5If a detail is not in your notes, leave a gap; do not let the model invent it.
Frequently asked questions
Can AI generate a good novel idea?
It can generate many plausible ones. The useful move is to bring your own what-if and let the generator expand it under constraints — who wants what, what blocks them, what failure costs. Without constraints you get the average of the genre.
How do I know if an idea can carry a whole novel?
Try to write thirty chapter goals as changes. If you can list thirty moments where the pressure rises, the idea has a spine. If you run out at eight, it is a short story or a scene.
What if I am writing a practical book from experience?
The idea is an angle, not a story. The same three constraints apply: who the book is for, what blocks them today, and what staying stuck costs. Then run the thirty-chapter test as thirty things the reader should be able to do by the end.
Do I have to disclose that I used AI for ideas?
Follow the platform’s rules and your own conscience. Research suggests disclosure lowers evaluations even when quality is unchanged, and a vague “AI-generated” label is less useful than a specific description of how AI was used.
Sources
- Google Search Central — Creating helpful, reliable, people-first content — read on 2026-11-16
About the author
One useful essay, every other week
Notes on writing, publishing, and the unglamorous parts of shipping software. No roundups, no growth hacks.





