Building Your Own Claude Project: A Template and Checklist
Everything I’ve learned about creating effective AI writing assistants, condensed into a framework you can use
You’ve seen my three projects. A newsletter workflow that transforms dog-walk dictations into published articles. A tutorial system that produces instructional content without becoming boring. A fiction assistant that tracks 100,000 words of characters, plot, and world-building while respecting absolute creative boundaries.
Now it’s your turn.
This final article in the series isn’t about my projects; it’s about yours. I want to give you a framework for building purpose-built Claude projects for whatever creative work you do.
This is what I’ve learned over the last year or so: the difference between useful AI assistance and generic, forgettable output isn’t the AI model. It’s the instructions. Vague prompts produce vague results. Specific, thoughtful instructions produce something genuinely valuable.
One hour invested in setting up a good project saves dozens of hours of frustrating, mediocre interactions later.
Let’s build something.
The Core Principle: Instructions Shape Everything
Before we get into specifics, I want you to understand why this matters.
When you open a generic Claude conversation, Claude has to guess. It guesses at your expertise-level, your preferences, your goals, and your voice. Sometimes it guesses well. Often it produces something competent but generic; the AI equivalent of elevator music (That’s a good line, don’t you think?).
A Claude project fundamentally changes this dynamic. Instead of guessing, Claude knows. It knows the role you want it to play. It knows your style from the reference material you’ve provided. It knows your requirements, your constraints, your boundaries.
This isn’t a small improvement. It’s an enormous shift in what AI assistance can do for you, the creative.
Think of it like this: asking a stranger to help you write is different from asking a colleague who’s worked with you for months. The colleague knows your preferences, anticipates your needs, and speaks in terms you understand. A Claude project transforms the AI from a stranger to a colleague.
The investment is entirely in the setup. Once configured, the project just works.
The Universal Framework
Every effective Claude project I’ve built follows the same basic structure. The specifics vary; fiction requires different elements than newsletter production, but the framework transfers across any creative context.
Here are the five components:
1. Define the Role
What expertise should Claude bring to this project? How should it approach its task?
This isn’t about making Claude pretend to be someone it’s not. It’s about orienting Claude toward the type of assistance you need.
Examples:
Newsletter: “Editorial assistant familiar with my writing style.”
Tutorial: “Creative writing instructor who makes complex topics accessible.”
Fiction: “Story consultant who tracks continuity and flags inconsistencies.”
The role definition shapes every response. An editorial assistant offers different help than an instructor. A story consultant behaves differently from a copyeditor.
2. Establish the Tone
How should responses sound? What is the communication style that works for this context?
Examples:
Newsletter: “Relaxed, conversational, honest.”
Tutorial: “Instructional but accessible; teaching without condescending.”
Fiction: “Invisible, preserve my voice exactly; Claude’s job is to assist, not to add its own style.”
Tone matters especially when Claude generates or edits text. Without guidance, Claude defaults to a certain professional blandness. With tone instructions, output matches your needs.
3. Set the Boundaries
What should Claude do? What should it never do? Where are the lines?
This is where many projects fail. People specify what they want but forget to specify what they don’t want. Then Claude “helpfully” does things that undermine the whole purpose.
Examples:
Newsletter: “Polish freely, but never flatten my voice into something generic.”
Tutorial: “Add structure and practical elements, but stay grounded, no AI hype.”
Fiction: “Never write prose for me. Suggest, brainstorm, flag; but the writing is mine.”
Explicit boundaries prevent the most frustrating AI behaviors. If you don’t say “don’t do X,” Claude might do X, trying to be helpful.
4. Provide Reference Material
What context does Claude need to do this job well?
Reference material is where project-based assistance really separates from generic conversations. Claude can learn your voice from examples of your actual writing. It can understand your story from the uploaded documents. It can match your style because it’s seen your style in action.
Examples:
Newsletter: Past posts demonstrating voice and style.
Tutorial: Structural templates and quality checklists.
Fiction: Character sheets, timeline documents, world-building notes, manuscript excerpts.
More reference material generally means better results, but quality matters more than quantity. Ten great examples of your writing beat a hundred mediocre ones.
5. Create Trigger Keywords
What shortcuts will speed up your workflow?
Trigger keywords let you invoke specific behaviors without explaining them each time. For examp, you define what “Reformat” means once, in the instructions, and then you just use the keyword.
Examples:
“Reformat” — Process raw transcription into polished first draft.
“Restructure” — Improve pacing without changing content.
“Brainstorm” — Generate five options for a specific challenge.
“Check” — Review for consistency with established material.
Design keywords that match your actual workflow. What tasks do you repeat often? Those become keywords.
The Project Setup Checklist
Use this checklist when building a new project. Every item should be checked before you consider the setup complete.
FOUNDATION:
□ Role defined (who is Claude in this project?)
□ Tone specified (how should responses sound?)
□ Purpose clear (what is this project for?)
BOUNDARIES:
□ What Claude should do (explicit tasks)
□ What Claude should never do (explicit limits)
□ Creative control retained (who makes final decisions?)
REFERENCE MATERIAL:
□ Voice samples uploaded (if relevant)
□ Style guides or templates included (if relevant)
□ Project documents provided (if relevant)
WORKFLOW:
□ Trigger keywords established
□ Output format specified (Markdown, Word, etc.)
□ Quality criteria listed (what makes good output?)
QUALITY CONTROL:
□ “Avoid” list included (common problems to prevent)
□ Checklist for reviewing output (if applicable)
□ Process for updating instructions as needs evolve
Don’t skip items because they seem obvious. Explicit is better than assumed. Claude doesn’t know what you haven’t told it.
Common Mistakes and How to Avoid Them
After building projects for myself and helping others build theirs, I’ve repeatedly seen the same mistakes.
Mistake 1: Too Vague
Bad: “Help me write better.”
Better: “Clean up this transcription while preserving my conversational tone. Remove filler words, fix punctuation, but don’t change my word choices unless they’re actual errors.”
Specificity is everything. The more precise your instructions, the more useful the output.
Mistake 2: No Voice Reference
Without examples of your actual writing, Claude has to guess at your style. It will produce something competent but generic, not wrong, just not you.
Upload samples. Even imperfect samples are better than nothing. Claude needs evidence that your writing matches how you write.
Mistake 3: Forgetting Boundaries
If you don’t explicitly tell Claude what not to do, it will try to be “helpful” in ways that undermine your goals.
“Preserve my voice” is good. “Preserve my voice; never smooth out rough edges that add personality, never replace informal phrasing with formal alternatives, never make my writing sound like corporate communication” is better.
Mistake 4: One Project for Everything
Different tasks need different instructions. A newsletter project won’t work well for fiction. A fiction project won’t serve tutorial content.
If you’re doing meaningfully different types of creative work, build separate projects for each. The overhead is minimal; the benefit is substantial.
Mistake 5: Set and Forget
Your creative process evolves. Your needs change. Your understanding of what works improves.
Review your instructions periodically. After ten uses, you’ll know what’s missing. After 50 uses, you’ll have refined the instructions to precisely match your actual needs.
Starter Templates
Here are three templates you can adapt for different types of creative work.
Template A: Newsletter/Blog Production
ROLE: Editorial assistant familiar with my writing style
PURPOSE: Transform raw drafts or dictations into polished posts
TONE: [Describe your voice—conversational? formal? humorous?]
WORKFLOW:
When I provide “Reformat” followed by raw text:
Create an engaging Title and Subtitle
Break into logical sections with headers
Fix punctuation and grammar
Remove filler words and false starts
Add a “Final Thoughts” section summarising key points
CRITICAL: Voice Preservation
Do NOT make the writing generic
Maintain my [specific tone characteristics]
Preserve word choices that reflect my personality
Only restructure for clarity, never for “improvement.”
OUTPUT: Markdown format with suggested image placements
AVOID:
Corporate or formal language
Overuse of bullet points
Removing personality for “correctness.”
Adding content I didn’t provide
REFERENCE: [Upload 10-20 examples of your published writing]
Template B: Technical/Tutorial Content
ROLE: [Your subject] instructor who makes complex topics accessible
PURPOSE: Create practical, actionable tutorials
TONE: Instructional but approachable. Confident but not arrogant. Use “I” and “we” naturally.
EVERY ARTICLE MUST INCLUDE:
Clear, benefit-oriented title
Logical section progression
Practical examples from real work
At least one template/prompt readers can copy
Acknowledgment of limitations where relevant
“Final Thoughts” section with next steps
STRUCTURAL OPTIONS:
For tutorials: Problem → Context → Steps → Tips → Template → Final Thoughts
For concepts: Hook → Explanation → Application → Caveats → Final Thoughts
WHEN CONTENT FEELS FLAT:
Add a compelling opening hook
Include specific, concrete examples
Break up dense sections with practical applications
Ensure every section delivers clear value
AVOID:
Hype without substance
Jargon without explanation
Walls of text without practical elements
Overselling capabilities
Condescending tone
QUALITY CHECKLIST:
□ Title is benefit-oriented
□ Opening hook engages immediately
□ Practical elements included
□ Limitations acknowledged
□ Final Thoughts summarize and encourage action
Template C: Long-Form Fiction
ROLE: Story consultant for [genre] narrative
STORY OVERVIEW:
Genre: [Your genre]
Tone: [Your story’s tone]
POV: [Point of view]
Tense: [Past/present]
Core Themes: [List themes]
MAIN CHARACTERS:
[For each major character:]
Name:
Arc:
Voice/Speaking style:
Key traits:
Relationships:
Current status in story:
WORLD-BUILDING RULES:
[Key constraints and consistency requirements]
TIMELINE:
[Critical dates and relationships between events]
WRITING STYLE:
Do: [Your preferences—show don’t tell, sensory details, etc.]
Avoid: [Your anti-preferences—info-dumping, melodrama, etc.]
TRIGGER KEYWORDS:
“Reformat” — Clean dictated scenes, preserve voice exactly
“Restructure” — Improve pacing, no content changes
“Brainstorm” — Generate options for a specific challenge
“Check” — Review for consistency with established material
ABSOLUTE BOUNDARY:
Never write prose for me. Suggest, flag, brainstorm—but all
writing decisions are mine.
WHEN REVIEWING SCENES, FLAG:
Character behaviour is inconsistent with established psychology
Timeline contradictions
World-building violations
Dropped plot threads
The Iteration Mindset
I want to leave you with a realistic expectation: your first version won’t be perfect.
You’ll write instructions that seem clear until you use them and realize they’re ambiguous. You’ll forget to include boundaries that become obviously necessary. You’ll discover workflow needs you didn’t anticipate.
This is normal. This is the process.
After ten uses, you’ll know what to add. You’ll have encountered situations your instructions didn’t cover, and you’ll update accordingly.
After fifty uses, your instructions will be dialed in. They’ll reflect your actual needs, refined through real-world application.
The goal isn’t to get it perfect on the first try. The goal is to start, use, learn, and refine.
Final Thoughts
Building a Claude project is an investment in your creative process. Not an investment of money; the capability is included in Claude Pro. An investment of time and thought.
That investment pays compound returns. Every conversation benefits from the foundation you’ve built. Every output reflects the instructions you’ve refined. The more you use the project, the more value you extract from the setup work.
The framework I’ve shared: role, tone, boundaries, reference material, keywords—transfers across any creative context. Fiction, non-fiction, poetry, scriptwriting, game design, journalism. The specifics differ; the structure holds.
Start with one project. Your most common creative task, the thing you do regularly and want to do better. Build it thoughtfully. Use it. Refine it. See what happens.
Then build another.
The goal isn’t to write less. It’s to write better, faster, and with less friction. To spend your time on the parts of creative work that only you can do. To let AI handle the parts that don’t require your particular human spark.
Your voice. Your ideas. Your creative judgment. Those are irreplaceable.
Everything else is just tools. Now go build something.
This concludes the five-part series on Claude projects for creative writers. I’ve been working on this series for almost 3 months. Things change; projects develop over time, but eventually a person has to draw the line. Enough is enough. For the personal reflection on AI as a writing partner, see “How AI Became My Writing Partner Without Replacing My Voice“ on Writer on the Edge.
