How to Use AI Ethically When Writing a Novel

The argument has been going for about 3 years now. It's got louder, not quieter. Authors are getting angry at each other on forums, readers are discovering their favourite novelists used ChatGPT for scene outlines, and Amazon's KDP team is somewhere in a meeting room trying to figure out what counts as "AI-generated content."

Everybody's got an opinion. Very few people have drawn a line they can actually defend.

Here's my attempt at one.

---

The Spectrum Nobody Talks About Directly

When people say "using AI to write," they mean about 6 different things simultaneously, and the conversation falls apart because everyone's arguing about a different point on the scale.

At one end: you ask an AI to suggest synonyms for "walked" because your brain has gone blank at 11pm. At the other: you dump a plot outline into a chatbot and publish whatever comes back with your name on the cover.

Between those two poles there's a gradient.

Research tool. You're writing a thriller set in 1987 Beirut. You ask an AI to summarise the political situation, name key factions, give you a feel for daily life. You then verify everything against primary sources. Same as googling, faster.

Structural sounding board. You've written yourself into a corner at chapter 14. You describe the problem to an AI. It suggests 4 ways out. You pick one, modify it, execute it yourself. Same as talking to a writing group, quieter.

Line editor. You paste a paragraph. The AI flags passive constructions, repeated words, a sentence that runs three clauses too long. You decide what to keep. Same as Grammarly, smarter.

Prose generator with heavy editing. You generate 500 words, keep about 80, rewrite the rest. You've pulled the shape of an idea out of the noise. More contested.

Co-writer. You generate text and publish it with light edits. The AI wrote significant chunks of what the reader is reading. Very contested.

Ghostwriter. You publish AI-generated work as your own with minimal human input. The reader assumes you wrote it. You didn't. Most contested.

The ethics live in those last 3 categories. The first 3 are, I'd argue, fine, because the tools are different but the process of writing (your thinking, your judgment, your creative decisions) is still yours.

The reason this matters: when people argue about AI in writing, they're almost always talking past each other because one person means "research assistant" and another means "ghostwriter." Get specific about which category you're discussing before you form an opinion.

---

Where Authors Draw Lines, and Why It Moves

Ask 100 authors where the line is and you'll get 100 different answers. That's not wishy-washy, it's structurally honest, because the line genuinely depends on context.

Genre matters. A literary novelist whose whole deal is voice and interiority will draw the line somewhere around "I asked AI to suggest synonyms." A thriller writer churning through 6 books a year for a genre audience that prizes plot over prose might draw it much further along. Neither is obviously wrong, because different genres carry different implicit contracts with readers.

Community norms matter too. Romance readers are, on average, more willing to accept AI assistance than literary fiction readers. Some romance authors have been open about using AI for first drafts and received minimal community backlash. If a Booker Prize longlister did the same and it came out, the fallout would be considerable.

Sci-fi sits somewhere in the middle. A genre that's historically fascinated by AI as subject matter has, predictably, a complicated relationship with it as a tool.

Your own authorial claim matters. If you've positioned yourself publicly as someone who laboured over every sentence for 3 years, and you actually spent 2 weeks editing AI output, you've deceived readers about something they bought into. If you've made no such claims, it's murkier.

The craft tradition also factors in, though people rarely admit this. An author who came up through MFA workshops, whose identity is built around close attention to language, will feel AI prose generation as a betrayal of something specific. An author who started in fan fiction, always collaborative and iterative, might not feel that weight at all. Both positions are internally consistent.

The practical test: would you be embarrassed if readers knew exactly how much AI was involved? If yes, you've probably crossed your own line.

---

The Disclosure Question

This is where things get genuinely complicated, and I'll try not to flatten it.

Should authors tell readers they used AI? The honest answer: it depends on what you used it for.

If you used AI the way you'd use spell-check, a thesaurus, or a structural editor, probably not a disclosure issue. Nobody expects a copyright page that says "typeset in InDesign" or "spell-checked with Microsoft Word." Tools that assist writing have always existed.

If AI wrote substantial portions of the prose the reader is consuming, I think you probably do owe the reader a disclosure. The reasonable expectation when buying a novel is that the named author wrote it. If they didn't, in any meaningful sense, that's worth knowing.

Some readers won't care. But the decision about whether they care should be theirs to make, not made for them by omission.

The tricky middle ground is where most authors actually live: they used AI for research, for brainstorming, maybe for drafting a scene or two they then heavily rewrote. Is that material? I don't think so. But I'd understand a reader who disagreed, and I think the honest response is to be direct if anyone asks rather than reach for vague language about "writing tools."

There's also the question of what disclosure even means in practice. "AI-assisted" as a label doesn't tell a reader much. It doesn't distinguish between "I used it to check my grammar" and "I used it to draft 40,000 words." If disclosure becomes standard, the field's going to have to get more specific about what it's disclosing.

One thing I'd push back on: the impulse to hide it suggests the author already knows they've crossed a line. That's a useful signal.

For the KDP-specific picture (disclosure thresholds, exact wording Amazon requires, what counts as "generated" vs "assisted"), we've got a full breakdown at [/blog/kdp-ai-disclosure-guide/].

---

Platform Policies: What the Rules Actually Say

KDP requires authors to disclose AI-generated content. The threshold is "AI-generated," which Amazon interprets as text, images, or translations produced by AI tools. AI-assisted content (human-written, AI-refined) sits in murkier territory, and Amazon hasn't fully resolved where that line is.

Traditional publishers vary wildly. Some major houses have issued blanket bans on AI-assisted manuscripts. Others have said nothing officially, which means their slush readers are probably guessing. A handful of literary agents have started asking about AI use in their standard submission queries, though with no standardised definition of what they're asking about.

Writing contests are where it gets sharp-edged. Most major contests (from the Booker to genre award bodies) have banned or heavily restricted AI-generated content. Some require a signed declaration. Submitting AI-assisted work to a contest that prohibits it isn't just an ethics question; it's grounds for disqualification.

The practical point: know the rules of whatever you're publishing to. They're different everywhere, they're changing fast, and "I didn't know" isn't a defence that tends to work.

---

The Economic Argument

There's a harder conversation underneath the ethics one, and most people skip past it because it's uncomfortable.

When you use AI for research or editing, you're generally not displacing compensated human labour. You would have googled it yourself, or run it through Word's editor. No developmental editor lost work because you talked through your plot with an AI.

When AI writes substantial portions of your novel, the picture shifts. You might have hired a co-writer. You might have sent the manuscript to a structural editor. Human creative workers (editors, writing coaches, developmental readers) exist in an ecosystem that's absorbing real pressure from AI-generated content flooding the market at scale.

The self-publishing side of this is already visible. KDP's romance and thriller categories have been hit by waves of AI-generated content that's difficult to distinguish from human-written work at first glance. The effect on legitimate authors trying to be found in those categories is real.

Readers who've been burned by low-quality AI content have started buying less speculatively, which means new authors—the ones who can least afford it—take the hit.

This isn't a clean argument and I won't pretend it is. Authors also struggle economically. AI tools let some writers produce more work, which is a genuine financial argument for using them. The counter is that flooding the market with AI-generated content may push down returns for everyone, including those same authors.

My position: tools that help you write better are worth supporting. Tools that write instead of you, then get passed off as human creative work, are doing something categorically different, and the economic consequences are landing on real people.

---

VSProse's Position

We build editing tools. That framing is deliberate.

VSProse's AI assists with prose editing: catching inconsistencies, flagging pacing problems, identifying overused constructions, suggesting alternatives you can accept or reject. The manuscript is yours. The decisions are yours.

Editing has always been a collaborative process. You've had beta readers, copyeditors, developmental editors. Every published book has been shaped by people other than the author. What we've built sits in that lineage: tools that work in service of the author's vision, not tools that substitute for it.

The line we've drawn internally is this: does the author write the book? If yes, we're a tool in that process. If we're writing the book, we've built the wrong thing.

There's a practical guide on how to use editing AI without having it sand down your voice into something generic at [/blog/ai-editing-without-losing-your-voice/].

---

A Practical Framework

Since everyone seems to want one, here's mine. 6 questions. Answer them honestly.

1. Did you write the prose the reader is reading?

If AI generated significant chunks that you published with light edits, that's worth examining. If you wrote it and AI helped you refine it, you're fine. The question is who made the creative decisions, not whether you used a tool.

2. Would you be embarrassed if readers knew exactly what AI did?

The embarrassment test is underrated. It's not about shame; it's your own honest assessment of where you landed on the spectrum. Most people know, instinctively, whether they've done something they'd rather not explain.

3. Does the platform require disclosure?

If KDP, a contest, or a publisher requires a declaration of AI content, declare it. Don't gamble on the definition of "generated" shifting in your favour.

4. What did you claim about your process?

If you've positioned yourself publicly as a craftsperson who writes every word by hand, you've made a claim. AI use that contradicts that claim is a problem even if the ethics were otherwise murky. Your platform, your reader relationship, your responsibility.

5. Are you supporting tools that support authors?

Editing tools, research assistants, structure aids: these broadly help authors do their work better. Wholesale prose generation is a different category with different knock-on effects on the people around you in the industry.

6. Can you defend your choices to a reader who asks directly?

If yes, you're probably fine. If you'd have to reach for vague language, that's a sign worth paying attention to.

---

There's no clean answer here, because there isn't one. The ethics of AI in writing are being worked out in real time: in forums, in publisher policy documents, in court cases nobody's brought yet.

But the authors who navigate this well seem to have one thing in common: they figured out where their line was before they crossed it. They made a deliberate choice and they can explain it.

Doing that now, before you're three books into a process you'd rather not have to justify, strikes me as worth the hour.