How Writers Can Use AI

A practical path for writers who want assistance, not authorship replacement.

There is a lot of noise around writers using AI.

• Some of it is justified.
• Some of it is fear.
• Some of it is elitism.
• Some of it is marketing.
• Some of it is bad-faith panic from people who do not understand the tools.

And some of it comes from the fact that there really is a flood of low-care AI-generated content being pushed into publishing spaces.

So I do not think writers should be careless.
But I also do not think “never touch AI” is a serious answer anymore.

Many writers already use AI-adjacent tools without thinking of them that way: grammar checkers, search engines with AI summaries, browser assistants, document tools, marketing tools, formatting helpers, transcription tools, and research tools. The writing ecosystem is already full of machine assistance.

The question is not simply:

Did AI touch the process?

The better question is:

Who carried the manuscript?

Start with the distinction: assisted vs generated

This distinction matters.

KDP currently requires authors to disclose AI-generated text, images, or translations when publishing or republishing through KDP. KDP defines AI-generated content as actual text, images, or translations created by an AI-based tool, even if the author edits it afterward. KDP says AI-assisted content does not need disclosure when the author created the content and used AI to brainstorm, edit, refine, error-check, or otherwise improve it. KDP also says authors remain responsible for ensuring both AI-generated and AI-assisted content follows content guidelines and intellectual property rules. (kdp.amazon.com)

That distinction gives writers a practical starting point.

• If you wrote the chapter and used AI to ask, “Where is the pacing weak?” that is AI-assisted.
• If AI wrote the chapter and you edited it, that is AI-generated text.
• If you wrote the poem and used AI to catch typos, that is AI-assisted.
• If AI generated the poem from your prompt, that is AI-generated text.
• If you asked AI for title ideas and wrote the book yourself, that is AI-assisted.
• If AI produced the manuscript prose, even with your outline, that is no longer the same category.

Writers need to understand this before they publish.
Not because platforms are always perfect.
Because clarity protects the work.

The manuscript stays with the writer

The cleanest rule is this:

Keep the manuscript in your hands.

For me, that means drafting in a human-led writing environment first.

I like Scrivener for this because it is built for long-form writing (and kind of old-school), not for outsourcing the writing. Literature & Latte describes Scrivener as combining notes, research, and writing in one place, allowing writers to work in sections, restructure drafts, keep research nearby, compile for self-publishing, and export to formats such as Word, PDF, ePub, and Kindle. (Literature & Latte)

That kind of tool keeps the book as a manuscript.

• Not a chat transcript.
• Not a single generated blob.
• Not a prompt result.

A manuscript.

AI can sit beside the process as a consultant.
It does not need to become the room.

Good line to keep:

Draft in your writing room. Consult AI outside the manuscript.

What AI can help with

AI can be useful for writers when the writer remains in charge.

It can help you:

• clarify a premise
• pressure-test a plot
• ask better character questions
• find continuity gaps
• summarize messy notes
• compare two outline versions
• spot repeated words
• identify pacing drag
• generate revision checklists
• draft a book description for you to rewrite
• suggest metadata options
• brainstorm newsletter topics
• organize a launch checklist
• turn scattered thoughts into a working plan

None of this requires handing over the soul of the book.

Used well, AI is not the ghostwriter.
It is the sharp reader at the table.

The assistant who asks:

• What is this scene doing?
• What changed because this happened?
• Where is the emotional consequence?
• Why does the character choose this now?
• What does the reader know here?
• What are you avoiding?
• Which thread disappeared?

That is useful.
That is not replacement.

What AI should not do for you

Here is where I draw the line.

• AI should not decide the book’s values.
• It should not imitate a living author’s voice.
• It should not generate pages you publish unread.
• It should not become your substitute for revision.
• It should not flatten your weirdness into generic “good prose.”
• It should not turn your manuscript into market sludge because the genre usually rewards certain tropes.
• It should not be allowed to decide what belongs in your story just because it says it confidently.
• It should not be fed your entire unpublished manuscript into random tools without understanding what happens to your data.
• And it should not be used to hide from the hard parts of writing.

Because the hard parts are often where the book becomes yours.

Do not dump the whole house into the tool

Writers need manuscript hygiene.

That means being careful about what you upload and where.

You do not always need to give an AI tool the full chapter, full book, full archive, full lore bible, or full private journal.

• Often, a summary is enough.
• A scene card is enough.
• A paragraph sample is enough.
• A character ledger is enough.
• A short excerpt is enough.
• A problem statement is enough.

Instead of pasting 80,000 words into a tool, ask:

• What am I trying to solve?
• Do I need the whole manuscript for this?
• Can I summarize the relevant context?
• Can I use a redacted version?
• Can I paste only the scene?
• Can I ask for a checklist instead of letting the model rewrite?
• Can I keep the manuscript in Scrivener and use AI only for analysis?

Do not feed the whole house to a tool when all you needed was help checking one window.

Keep a simple process log

Writers do not need to become paranoid.
But they should keep records.

NB: Scrivener has auto meta-data.

A simple process log is enough:

• Date
• Project
• Tool used
• Purpose
• Input shared
• Output kept or rejected
• Human decision made

Example:

Date: 2 June
Project: Novel draft
Tool: ChatGPT
Purpose: pacing check for Chapter 4
Input shared: scene summary + 700-word excerpt
Output kept: one note about unclear motivation
Human decision: rewrote transition myself; no AI prose used

That is not a confession.
That is authorship hygiene.

If you ever need to remember what happened, you will know.

Publishing paths: KDP, Lulu, direct sales

Writers also need to think about the publishing path early enough to avoid surprises.

KDP gives access to Amazon’s publishing ecosystem and has its own content guidelines, quality rules, AI disclosure categories, promotional tools, and distribution options. Its help center lists book setup, formatting, categories, metadata, Author Central, A+ Content, advertising, KDP Select, and distribution-related resources under its publishing and marketing documentation. (kdp.amazon.com)

Lulu is a different kind of path. Lulu Direct is built for selling print-on-demand books through your own site, with integrations for Shopify, Wix, and WooCommerce; Lulu says authors can manage the sale, use Lulu for print and fulfillment, retain customer data through their own checkout, and connect through Lulu’s print API or ecommerce integrations. (Lulu)

For many writers, the honest answer may be both.

KDP for reach.
Lulu for direct sales, special editions, author copies, journals, planners, workbooks, and a stronger relationship with your own readers.

But choose intentionally.
Do not publish everywhere just because you can.
And do not assume Amazon is the only legitimate shelf.

Marketing without becoming a content machine

A writer does not need to turn into a full-time influencer to sell a book.

But readers need somewhere to return.

A simple system can be enough:

• one website or landing page
• one author bio
• one clear book page
• one newsletter or supporter space
• one repeatable posting rhythm
• one soft call-to-action

That rhythm can be gentle:

• monthly writing update
• small excerpt
• behind-the-scenes note
• worldbuilding fragment
• deleted line
• cover process
• reader question
• process reflection
• launch reminder

The goal is not to shout every day.
The goal is to stay findable, trustworthy, and alive to the reader.

The Patreon / supporter layer

A supporter platform can work if it stays simple.
It does not need to become another exhausting machine.

For writers, a supporter space can offer:

• early excerpts
• monthly author notes
• deleted scenes
• behind-the-scenes process
• worldbuilding notes
• printable extras
• Q&A posts
• supporter-only reflections
• book-club style discussions

But the writer should design it carefully.

• Do not promise more than you can sustain.
• Do not turn the supporter space into a second full-time job.
• Do not make people pay for the actual book’s missing pieces.

Use it as a warm room around the work, not a pressure chamber.

AI can help with marketing, but not identity

AI can help draft:

• book descriptions
• taglines
• launch checklists
• newsletter outlines
• social captions
• reader questions
• FAQ pages
• press kit drafts
• author bios
• metadata experiments

But AI should not decide your author identity.

It should not tell you to become louder, trendier, more sexual, more controversial, more marketable, or more generic just because that is what the internet rewards.

Marketing should reveal the work.
It should not replace it.

The point is not to become a content machine.
The point is to create enough doors for the right readers to find the house.

The Atelier writer code

Here is the code I would give any writer using AI:

Write the book first.

• Use AI to think, not to disappear.
• Keep your manuscript in your own writing system.
• Use excerpts and summaries before full uploads.
• Do not imitate living authors.
• Do not publish unread AI output.
• Do not hide platform-required disclosures.
• Keep a simple process log.
• Protect your voice.
• Revise like a human.
• Market with rhythm, not panic.
• Publish only what you can stand behind.

The point

AI can help a writer.

• It can help organize the mess.
• It can ask useful questions.
• It can catch weak structure.
• It can help with revision.
• It can draft marketing copy.
• It can make the road to publishing less lonely and less chaotic.

But it should not inherit the manuscript.

• The book is not a prompt result.
• The book is a body of decisions. The writer’s decisions.

If AI enters the workshop, let it carry tools.

Not the pen.