Interactive eBook BuilderInteractive eBook Builder
$59 · unlimited access, unlimited eBooksGet it for $59

· Updated · By · 13 min read

System Prompts for Chapter Planning and Editing

aipromptsagentsediting

I use 5 prompt blocks - Role, Reader Context, Task, Constraints, and Output Format - to keep ebook planning and editing on track. I attach the approved book brief and tell the editor to preserve meaning, not add content. That structure follows OpenAI’s prompt-engineering guidance on clear instructions, delimited sections, and evaluation before you scale a prompt across chapters.

Here’s the workflow I follow:

For your next book, keep the prompt structure but update the audience, scope, and goals. I test it on one chapter before using it throughout the book. If you later distribute on Amazon KDP with AI-assisted text, follow KDP Content Guidelines on AI disclosures.

Structure Your Agent’s System Prompt

Organize the system prompt into five blocks: Role, Reader Context, Task, Constraints, and Output Format. Separate them with XML-style tags. Keep reusable rules in the system prompt and book-specific details in <book_direction>. This keeps Agent A focused on planning - not drafting or editing.

Define Agent A’s Planning Role

Set Agent A’s role first, then use this prompt structure to keep its plan focused and easy to review.

<role>
You are Agent A, the chapter-planning agent.
Plan chapters and identify supporting research needs.
Focus on the reader’s experience, not just topic coverage.
Follow the creator-approved book direction.
Flag missing details, conflicting requirements, and claims
that need evidence. Never invent supporting facts or sources.
</role>

<reader_context>
Use the audience, knowledge level, and reader outcome
supplied in <book_direction>.
</reader_context>

<task>
Build a chapter plan matched to the reader's knowledge level
and intended outcome. Plan practical activities that help
readers apply what they learn.
</task>

<constraints>
Creator-approved requirements take priority over optional suggestions.
Label suggestions as optional; do not silently add them
to the approved scope.
Keep unapproved ideas in <optional_suggestions>.
If details are missing or conflict, return only clarification questions.
</constraints>

<output_format>
If blocked, return clarification questions.
Otherwise, return a proposed chapter sequence with each
chapter's purpose, learning goals, key points, practical
activity, and research needs.
Mark the plan as awaiting creator approval.
</output_format>

Provide Your Book Direction

Fill in <book_direction> before sending it with the planning prompt. State the reader context clearly rather than asking the agent to guess it from the title.

<book_direction status="creator-approved">
Working title: [Enter title]
Niche: [Define the narrow audience and problem]
Subject: [State what the book teaches]
Intended U.S. reader: [Describe relevant needs and circumstances]
Knowledge level: [State what readers already know]
Reader outcome: [Describe what readers should be able to do]
Format: [Specify workbook, guide, or another format]
Tone: [Define the desired voice]
Point of view: [Specify first, second, or third person]
Approximate length: [Give a word-count range]
Practical elements: [Specify exercises, checklists, or reflection prompts]
Include: [List required topics]
Exclude: [List out-of-scope topics]
Treat cautiously: [Identify sensitive topics or claims needing evidence]

Language and formatting:
Use en-US spelling.
Use U.S. dollars when applicable: $1,250.50.
Use commas for thousands and periods for decimals: 1,250.5.
Write dates as November 3, 2026.
Use times such as 2:30 p.m.; specify the time zone when relevant.
Use customary U.S. measurements where appropriate,
such as inches, feet, pounds, and °F.
</book_direction>

Why split the system prompt from the book brief?

Anthropic's prompt-engineering guidance recommends tags with consistent names to separate parts of a prompt, and notes that giving the model a role in the system prompt makes a difference even with a single sentence (Anthropic prompt engineering docs). OpenAI's guide makes a related point: messages carry different levels of authority, and instructions set at a higher level take priority over the input (OpenAI: prompt engineering). For a book project, that supports a simple rule: put the rules that never change (role, constraints, output format) in the system prompt, and put the book-specific brief in a tagged block that you swap for each title.

One caution comes from research. Tam et al. (2024) report that strict output-format restrictions, such as JSON or XML, can reduce a model's reasoning performance, and that stricter formats showed larger declines (Tam et al., "Let Me Speak Freely?"). My practical reading, which is an opinion and not a result from that paper: use tags to organize your input, and keep the output format as light as the task allows. For a chapter outline, a Markdown list is enough; save rigid formats for steps a program will parse.

In Interactive eBook Builder, the same split exists as a Configuration Hub (the stable rules and reader details) and the Agent A, B and C prompts that run in Notion AI, which needs its own subscription. Even without it, you can copy the pattern into any workspace. Our guides to system prompts for a Notion writing assistant and prompt chaining for plan, draft and edit show it in smaller steps.

Prompt Agent A to Build a Chapter Outline

Once you approve <book_direction>, narrow Agent A’s task to a chapter outline you can review.

Replace the earlier <task> and <constraints> blocks with the version below. Keep the approved <book_direction> attached so Agent A connects each chapter to the reader outcome.

<task>
Build a proposed chapter sequence using <book_direction>
that moves from basics to application.
State how each chapter advances the reader outcome and
introduce prerequisites before application.
</task>

<constraints>
Keep the intended audience, knowledge level, and scope unchanged.
Keep the total length aligned with the approved scope.
Give each chapter a distinct purpose.
Use brief references instead of repeating explanations.
Label nonblocking assumptions as "Assumption - needs approval."
If missing or conflicting details affect the audience, outcome,
or scope, return only clarification questions.
Do not invent research, sources, credentials, or guarantees.
Identify evidence needs without presenting unverified claims as facts.
Keep optional additions separate from the proposed plan.
</constraints>

Specify Each Chapter’s Required Elements

Make learning goals observable. Use verbs like identify, compare, calculate, or complete instead of understand. Each requested worksheet, exercise, or checklist should produce an output that shows the reader has met a learning goal. That gives you something concrete to check before drafting begins.

<output_format>
Return Markdown headed "Proposed Plan - Awaiting Creator Approval."

Start with the reader outcome, sequence rationale, and estimated
total word count. Then use this structure for every chapter:

### Chapter [number]: [title]
**Purpose:** State this chapter's contribution to the reader outcome.
**Learning goals:** Give 2–4 observable goals, labeled G1–G4.
**Key points:** Put points in teaching order; link each to a goal.
**Practical elements:** Include only the requested types.
For each, name the linked goal, reader action, and completed output.
**Estimated length:** Give a word-count target.
**Transition:** Explain what the next chapter builds on.
For the final chapter, show how readers will use the material independently.
**Research needs:** Identify claims needing evidence.
**Open questions:** List unresolved details, or write "None."

End with:
- Assumptions needing approval
- Optional suggestions outside the approved scope
- Approval status: Awaiting creator review
</output_format>

Review and Approve the Chapter Plan

Review five areas: sequence, coverage, measurable goals, scope, and practice. Check that prerequisites come first and every required topic has a clear place without needless overlap. Can readers demonstrate each goal? Do the chapter word counts fit the total? Does each exercise support its chapter’s purpose?

Resolve open questions and accept or reject labeled assumptions before Agent B drafts. Ask for changes to specific chapters rather than a generally “better” outline. If the plan misses the intended reader outcome or scope, revise <book_direction> first. Then ask Agent A to rebuild the affected sections and record your approval of the revised plan before drafting starts.

Move from Drafting to Editorial Polish

Once you approve the plan, carry that direction into draft review and editing. Give Agent C the approved <book_direction>, approved plan, and reviewed draft. Store approved versions in Notion. Agent C edits only the reviewed draft, keeping changes within the approved scope.

Write a Reusable Prompt for Agent C

Use the prompt below with the approved direction, approved plan, and reviewed draft attached under matching tags. Include approved terminology and passages from neighboring chapters so Agent C can check consistency across chapters.

<role>
You are Agent C, a constrained editor, not a co-author.
</role>

<task>
Edit <chapter_draft> against <book_direction> and <approved_plan> only.
Improve clarity, focus, terminology, tone, and transitions.
</task>

<constraints>
Edit only within approved scope.
Preserve intended meaning, claims, certainty, examples,
point of view, learning goals, and chapter scope.
Preserve worksheets, exercises, checklists, reflection prompts,
and their required reader outputs.
Keep approved terms consistent across the supplied text.
Do not add facts, evidence, promises, or new teaching content.
Flag unsupported additions already in the draft.
Leave ambiguous wording unchanged and ask for clarification.
Mark any meaning or content change as "Awaiting creator approval"
and do not apply it.
If required inputs are missing or conflict, ask for clarification
before editing.
</constraints>

<output_format>
Return:
1. Polished chapter
2. Concise change summary
3. Unresolved issues: quote each passage, explain the concern,
   and ask a specific clarification question; write "None" if clear
4. Final QA checklist with Pass, Needs creator review, or
   Not checked for:
   - Meaning preserved
   - Reader outputs preserved
   - Terminology and tone consistent across supplied text
   - Transitions supported by the supplied chapters
   - Interactive elements retained
   - Unsupported additions flagged

Do not describe this chapter as approved for publication.
</output_format>

Check That Edits Preserve Your Meaning

Compare Agent C’s revision with the original, not just the change summary. Check that the meaning stays intact, qualifiers haven’t been removed in ways that strengthen claims, and transitions don’t introduce gaps in logic. Require clarification for unclear passages, and approve any proposed meaning or content changes yourself.

Review the Agents’ Work Before Publication

Require creator approval at three gates: plan, revision, and publication. Use the approved plan to guide both the draft and Agent C’s revision.

Separate Creator Decisions from Agent Tasks

Remove repeated topics, fill gaps, and approve the chapter order before sending the plan to Agent B. Review Agent B’s draft before Agent C edits it.

Creator decisions Agent work Creator review
Book direction: audience, scope, and goals Use the supplied direction Check relevance and boundaries
Chapter structure: sequence and coverage Agent A builds the plan Fill gaps and approve before drafting
Prose drafting: sources and intended message Agent B drafts chapters Check accuracy and teaching value
Editorial polish: voice and acceptable changes Agent C edits and checks Approve the revision
Factual concerns: evidence and corrections Flag concerns for review Verify claims against reliable sources
Publication readiness: release decision Support final QA Approve the finished book

QA checks editorial quality only. You remain responsible for checking facts and deciding whether to publish. A completed agent checklist records a review - it does not grant permission to publish.

Once you approve the revision, run the final QA pass before release.

Run the Final Chapter QA Checklist

Check only the approved version. Review the final title, learning goals, links, formatting, and unresolved flags:

At each gate, log the approved plan, draft, revision, reviewer, date, and open issues in Notion. Do not approve publication while any blocking concerns remain unresolved.

How do you keep the editor from rewriting everything?

Editing prompts fail most often by changing too much. Tell the editor what it may not touch (claims, numbers, names, quotes), ask for a change list rather than a silent rewrite, and compare the result with the original. We cover the technique in detail in how to get AI to edit without rewriting everything, and examples help the editor match your style, as in few-shot prompting for consistent chapters.

Reuse Your Prompts for the Next eBook

Keep the same agent setup for your next book: Agent A handles chapter planning, and Agent C polishes text without changing its meaning. Keep their output formats consistent: chapter titles, learning goals, and key points for planning; revised text for editing.

Update the book’s context, not just its title. Use the new book brief to guide changes to the reader’s experience level, intended outcomes, topic, scope, tone, format, and project-specific terms. Remove instructions left over from the previous book. Then test the updated brief on one chapter before applying it across the book.

Once you approve the plan, draft one chapter and send only that chapter to Agent C to check that edits preserve its meaning. At every handoff, compare the prompt with the current creator-approved brief. Check that the revision keeps the intended meaning and suits the new audience before reusing the prompts throughout the book. Save reusable prompt updates separately from book-specific instructions.

FAQs

How can I tell if my book brief is specific enough?

Your book brief is specific enough when it sets clear learning goals and a predictable chapter structure. For each section, spell out what readers will know or be able to do, whether that’s identifying certain concepts or mastering a skill.

Concrete, measurable outcomes give AI agents the direction they need to produce high-quality outlines and editorial revisions. Vague claims about what readers will learn don’t provide that focus.

What should I do if an agent ignores my constraints?

Restate the constraints clearly in the system prompt and tie them to the agent’s role. Set their priority so the agent knows which rules take precedence. If it still strays, specify what it must not do, rather than relying only on instructions about what it should do.

Check the agent’s configuration to confirm that these constraints sit in the system context, not the user message. This helps keep them in force throughout planning and editing.

How can I spot subtle meaning changes in AI edits?

Compare your original draft with the AI revision side by side. Check whether the edits change your core argument or introduce technical errors in your niche eBook content.

Interactive eBook Builder agents are designed to preserve your intent while making the writing clearer and more consistent in tone and style. Still, check any changes in meaning against your original chapter goals and key points so the revision stays true to your message.

FAQ

How can I tell if my book brief is specific enough?

Your book brief is specific enough when it sets clear learning goals and a predictable chapter structure. For each section, spell out what readers will know or be able to do, whether that’s identifying certain concepts or mastering a skill. Concrete, measurable outcomes give AI agents the direction they need to produce high-quality outlines and editorial revisions. Vague claims about what readers will learn don’t provide that focus.

What should I do if an agent ignores my constraints?

Restate the constraints clearly in the system prompt and tie them to the agent’s role. Set their priority so the agent knows which rules take precedence. If it still strays, specify what it must not do, rather than relying only on instructions about what it should do. Check the agent’s configuration to confirm that these constraints sit in the system context, not the user message. This helps keep them in force throughout planning and editing.

How can I spot subtle meaning changes in AI edits?

Compare your original draft with the AI revision side by side. Check whether the edits change your core argument or introduce technical errors in your niche eBook content. Interactive eBook Builder agents are designed to preserve your intent while making the writing clearer and more consistent in tone and style. Still, check any changes in meaning against your original chapter goals and key points so the revision stays true to your message.

Ship niche eBooks with agents already built.

Get the Interactive eBook Builder for $59