# How to write in your own voice with AI

Show the AI samples of your own writing, give it specific rules, and feed it the stories and opinions only you have. Then correct one chapter before it writes the next, and edit everything yourself. Examples steer it more reliably than adjectives like “warm”. The test is simple: read a page aloud, and rewrite whatever you wouldn’t say.

Last reviewed 30 October 2026. Web page: https://chapbook.ink/guides/write-in-your-own-voice-with-ai/

## Why the default voice sounds like nobody

Ask a model for a chapter with no other help, and you get its average: tidy, confident, a little too balanced. It isn’t bad writing so much as nobody’s writing.

Research points the same way. In a controlled experiment with 118 participants from India and the United States, published at CHI 2025, Agarwal, Naaman and Vashistha found that “AI suggestions led Indian participants to adopt Western writing styles, altering not just what is written but also how it is written.” The study was about writing suggestions and cultural style, not books, but the mechanism is the one to watch. If you accept the suggestion each time, your writing drifts toward the model’s.

## Show it, then tell it

Anthropic’s prompting guide calls examples “one of the most reliable ways to steer Claude’s output format, tone, and structure”. That is one vendor’s advice about its own model, but the habit costs nothing to try with any of them: lead with samples, and use rules to catch what samples miss.

| Weak instruction | Stronger instruction |
|---|---|
| “Warm and professional” | “Talks like a friend who has done this before. Says ‘I got this wrong’ at least once a chapter.” |
| “Engaging” | “Open each chapter with something that happened to a real client, never with a definition.” |
| “Concise” | “Most sentences under 15 words. No paragraph over four lines.” |
| “Avoid clichés” | “Never: journey, unlock, game-changer, in today’s fast-paced world.” |

A rule you can check is one you can hold the draft to. Rules of your own also tend to be more useful than a borrowed list, because they describe what you actually do.

## Your material is the voice

Most of what reads as “voice” is really content: the story you tell, the opinion you hold against the usual advice, the client who proved you wrong. A model has none of that unless you hand it over. Before drafting, write down the ten things you always end up saying about this topic and the three things people in your field get wrong. Those notes do more than any style setting.

## What voice matching doesn’t change

A book that sounds like you is still a book an AI drafted, if that is how it was made. On KDP, the question of AI-generated content depends on who created the text, not on how it reads. KDP’s definition says: “We define AI-generated content as text, images, or translations created by an AI-based tool. If you used an AI-based tool to create the actual content (whether text, images, or translations), it is considered “AI-generated,” even if you applied substantial edits afterwards.” Answer on those terms, and use the [AI-generated or AI-assisted helper](/tools/kdp-ai-disclosure-helper/) if you aren’t sure which applies.

## Where Chapbook fits

On your Author profile, add two or three pieces you wrote yourself, up to 5 samples of 20,000 characters each, and Chapbook reads them into a voice profile that drafting uses. The brief has its own voice field for each book, and your notes and sources hold your material.

You see chapter one before the rest is written, so you can correct it first. After that, assistant edits arrive as suggestions to accept or reject, and each chapter’s history records who wrote what, which feeds the record the KDP listing shows you.

The sample on this page is a first-person book whose brief asked for narrative chapters that each end with what the author would tell a friend. Its family and its bakery are fictional, as the brief says.

## What to do

1. **Gather three to five real samples.** Pick pieces you wrote without AI, in the register the book needs: a newsletter, a talk, a long email to a client. Five blog posts in one style beat five kinds of writing.
2. **Write rules you could check.** “Friendly” can’t be checked. “Short sentences, says ‘I’ and never ‘we’, no rhetorical questions, one joke a page at most” can. Add the words you would never use.
3. **Give it what only you know.** List the stories, client questions, mistakes and opinions the book should carry, and put them in the brief or your notes. A model can’t invent your experience, and shouldn’t try.
4. **Fix one chapter before the rest.** Draft one chapter and mark what sounds wrong. Tell it why, then draft again. Corrections are far cheaper now than after every chapter has the same habit.
5. **Edit with your own hands.** Go through every chapter. Replace generic examples with real ones, cut what you wouldn’t say, and read each page aloud. Anything you stumble over goes.

## Made with Chapbook: Slow Money

Slow Money: What a family bakery taught me about pricing, patience and staying small, by Rosa Almeida. Nonfiction book, 9,273 words, 37 pages. A sample generated from one brief and exported unedited; it is not a customer's book. Read every page: https://chapbook.ink/examples/bakery/

The brief:

> Twenty years of running a two-shop family bakery: the decisions about price, growth and hours that kept it small and solvent. Promise: A clear-eyed account of how a small business stays small on purpose, with the numbers and rules of thumb that made it work. Reader: A small-business owner weighing a second location, a price rise, or a loan. Voice: First person, narrative chapters that each end with what the author would tell a friend; a fictional family, bakery and town; no invented statistics beyond the bakery's own illustrative numbers.

The outline the planner proposed:

1. Prologue: The Second Oven (585 words)
2. The First Price Rise (1,250 words)
3. The Loan We Did Not Take (1,101 words)
4. The Second Shop, and Why Not a Third (1,186 words)
5. Hours, and the Cost of Being Open (1,085 words)
6. The Wholesale Account We Let Go (1,142 words)
7. Hiring Family and Paying Them Properly (1,216 words)
8. What the Numbers Said Each January (1,179 words)
9. Epilogue: Small on Purpose (478 words)

## Questions

### How many writing samples does AI need to match my voice?

A few good ones. Anthropic’s prompting guide says a “few well-crafted examples” improve accuracy and consistency, as checked on 30 October 2026. We suggest three to five, in the same register as the book.

### Why does AI writing sound the same?

Models lean toward the most common patterns in what they learned from. A 2025 study found AI suggestions pulled Indian writers toward Western styles, so your own samples and edits matter.

### Does matching my voice change what I must tell KDP?

No. KDP defines text created by an AI tool as AI-generated “even if you applied substantial edits afterwards”, as checked on 29 October 2026. A voice profile doesn’t change that answer.

### How does Chapbook learn my voice?

You add up to 5 writing samples of 20,000 characters each to your author profile, and Chapbook reads them into a profile that drafting uses. You can edit every word it writes.

## Sources

- [Agarwal, Naaman and Vashistha, “AI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances”, CHI 2025](https://arxiv.org/abs/2409.11360)
- [Anthropic, “Prompting best practices” (checked 30 October 2026)](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices)
- [Amazon KDP, “Content Guidelines”](https://kdp.amazon.com/en_US/help/topic/G200672390)