| name | voice-matching-wizard |
| description | Transform writing samples into a codified voice style that can be replicated consistently. This wizard guides you through analyzing samples, extracting patterns, and generating a custom voice skill. |
| tier | light-setup |
| setup_time | 15-20 minutes |
| requires | 2-5 writing samples (500+ words each) |
| outputs | voice-[your-name].skill.md |
Voice Matching Wizard
Create a voice skill that captures the patterns, rhythms, and sensibilities of any writing style.
What This Does
Voice matching is the art of identifying what makes writing recognizable. Not just the obvious markers—vocabulary, sentence length—but the deeper patterns: how ideas unfold, where the writer pauses, what they leave unsaid.
This wizard walks you through the process of:
- Gathering representative samples
- Extracting patterns across multiple dimensions
- Synthesizing a usable voice profile
- Generating a working skill file
The output is a voice-[name].skill.md you can use alongside other skills (like anti-ai-writing) for consistent, authentic content.
Before You Begin
What makes a good sample:
- At least 500 words
- Writing you're proud of (or writing you want to emulate)
- Representative of the voice you want to capture
- From the same medium (all newsletters, all blog posts, etc.)
What to avoid:
- Heavily edited committee writing
- Content written under constraints
- Mixed formats (tweets + longform together)
- Very old work that doesn't reflect current voice
How many samples:
- 2-3 samples: Basic patterns
- 4-5 samples: Good coverage
- 10+: Comprehensive capture
Three Paths Through This Wizard
Path A: "I have samples of my own writing"
You want to codify your existing voice so AI can match it consistently.
→ Skip to Phase 1: Gather Your Samples
Path B: "I want to write like [someone I admire]"
You want to emulate an author, publication, or brand voice.
→ If the writer is well-known (major author, publication), you may be able to use their name directly in prompts. For custom or lesser-known voices, you'll still need samples.
→ Proceed to Phase 1: Gather Your Samples
Path C: "I'm not sure what my voice is yet"
You want to discover and develop your voice through this process.
→ Start with Phase 0: Voice Discovery (below)
Phase 0: Voice Discovery (Optional)
Skip this if you already have samples or know what voice you want.
To discover your voice, we need to understand how you naturally communicate when you're not performing.
Answer these questions:
-
How do you explain things to friends?
- Do you use stories and examples?
- Do you build logical arguments?
- Do you make jokes?
- Do you ask questions to draw them in?
-
What writers or publications do you gravitate toward?
- Not who you think you should read—who do you actually read?
- What about their style appeals to you?
-
What's your natural sentence length?
- Short and punchy?
- Long and flowing?
- Variable depending on the point?
-
How do you feel about jargon?
- Love it (signals expertise)?
- Hate it (pretentious)?
- Use it sparingly when precise?
-
What's your relationship with your reader?
- Peer/friend?
- Mentor/teacher?
- Curious explorer?
Based on your answers, find 2-3 pieces you've written that feel authentic. Use those for Phase 1.
Phase 1: Gather Samples
Collect 2-5 writing samples. Paste each one into a separate document, or note where they can be found.
For each sample, note:
- Source/context (blog post, newsletter, etc.)
- When it was written
- Why you chose it
Phase 2: Extract Patterns
Analyze each sample across these dimensions:
2.1 Sentence Architecture
| Element | What to Look For |
|---|
| Length | Average sentence length (short/medium/long) |
| Variation | Does length vary deliberately for rhythm? |
| Complexity | Simple declarative? Compound? Complex? |
| Punctuation | Em dashes? Semicolons? Minimal? |
| Fragments | Used for emphasis? Never? |
Signature structures to identify:
- Opening patterns ("Here's the thing:")
- Rhythm breaks ("But.")
- List patterns ("First... Second... Third...")
- Closing moves (questions, callbacks, calls to action)
2.2 Word Choice
| Element | What to Look For |
|---|
| Vocabulary level | Simple/moderate/sophisticated |
| Formality | 1-5 scale (casual to formal) |
| Contractions | Always/sometimes/never |
| Jargon | Industry terms? Avoided? |
| Strong language | Profanity level/style if any |
Identify:
- Favorite words (appears 3+ times across samples)
- Avoided words (never appears despite opportunity)
- Signature phrases
2.3 Tone & Attitude
| Element | What to Look For |
|---|
| Emotional register | Warm? Cool? Intense? Measured? |
| Reader relationship | Peer? Mentor? Friend? Expert? |
| Humor style | Sarcasm? Wordplay? Self-deprecation? None? |
| Certainty level | Definitive or exploratory? |
| Attitude toward subject | Passionate? Skeptical? Curious? |
2.4 Structural Moves
| Element | What to Look For |
|---|
| Openings | Story? Question? Bold claim? Scene-setting? |
| Transitions | Seamless? Signposted? Abrupt? |
| Closings | Call to action? Question? Summary? Callback? |
| Paragraph length | Short punchy? Long flowing? Mixed? |
2.5 Distinctive Techniques
What makes this voice recognizable?
- Rhetorical devices (questions, analogies, callbacks)
- Signature moves unique to this writer
- How evidence is presented (data? anecdotes? logic?)
- Pattern interrupts (how does the writer surprise?)
2.6 What They DON'T Do
Just as important:
- Phrases never used
- Structures avoided
- Topics skipped
- Formality levels never hit
Phase 3: Synthesize
The Voice in One Paragraph
Write a single paragraph capturing the essence:
"[Name]'s voice is [primary characteristics]. They write like [relationship to reader], using [key techniques]. Their tone is [tone], with [humor style if applicable]. Sentences tend to be [structure]. They favor [word choice] and avoid [avoidances]. The overall effect is [feeling/impact]."
Voice Spectrum
Rate the voice on these scales (1-5):
Formal ←――――――――――→ Casual [ ]
Expert ←――――――――――→ Peer [ ]
Serious ←―――――――――→ Playful [ ]
Reserved ←――――――――→ Opinionated [ ]
Abstract ←――――――――→ Concrete [ ]
Phase 4: Generate the Voice Skill
Using your analysis, create voice-[name].skill.md with this structure:
---
name: voice-[name]
description: Write in [Name]'s distinctive voice. [One sentence characterization]. Use with anti-ai-writing for authentic content.
---
# Voice: [Name]
[Your one-paragraph voice description from Phase 3]
## Voice Spectrum
- Formal/Casual: [1-5]
- Expert/Peer: [1-5]
- Serious/Playful: [1-5]
- Reserved/Opinionated: [1-5]
- Abstract/Concrete: [1-5]
## Core Characteristics
### Sentence Architecture
[Your findings from 2.1]
### Word Choice
[Your findings from 2.2]
### Tone & Attitude
[Your findings from 2.3]
### Structural Moves
[Your findings from 2.4]
### Distinctive Techniques
[Your findings from 2.5]
### Avoidances
[Your findings from 2.6]
## Example Patterns
### Signature Openings
[3-5 examples from samples with pattern explanation]
### Signature Transitions
[3-5 examples]
### Signature Closings
[3-5 examples]
### Signature Sentences
[5-10 exemplary sentences that capture the voice]
## The [Name] Test
Before publishing, ask:
- [ ] Would [Name] actually write this sentence?
- [ ] Does it match the voice spectrum above?
- [ ] Are signature patterns present?
- [ ] Are avoidances absent?
- [ ] Does it *feel* right?
## Anti-Patterns
If you see these, you've drifted from the voice:
- [List 5-10 patterns that would violate this voice]
Phase 5: Validate
Test your voice skill:
- Write a short piece using only the voice skill
- Compare to original samples
- Identify gaps or mischaracterizations
- Refine based on testing
Validation checklist:
Human Writing Fundamentals
Every voice skill should build on these principles from anti-ai-writing.
The Energy Transfer Principle
The best writing is a transfer of energy from writer to reader. When analyzing samples, notice how the writer transfers energy:
- Do they write conversations or speeches?
- Do they speak WITH their audience or AT them?
- Where do they use specific, concrete language vs. abstract ideas?
The SUCKS Framework
Apply this when generating voice output:
S - Specific: Who is the ONE reader?
U - Unique & Useful: Does it change how they think, feel, or act?
C - Clear, Curious, Conversational: Does it read like talking to a friend?
K - Kept Simple & Structured: Simple ideas, clear structure?
S - Sticky: Are there memorable phrases they'll repeat?
Sticky Sentence Techniques
When identifying signature sentences in Phase 2, look for these techniques:
Alliteration — Same starting sounds
- "Specificity is the secret"
- "The best jobs are neither decreed nor degreed"
Symmetry — Parallel structure
- "Read for awareness. Write for understanding."
- "It's not 10,000 hours. It's 10,000 iterations."
Contrast — Opposing ideas
- "To be everywhere is to be nowhere."
- "Be clear, not clever. Concise, not complex."
Rhythm — Pleasing cadence
- Sentence length variation for effect
- Short sentences for emphasis
- Longer sentences for flow
Include the writer's use of these techniques in your voice skill.
AI Tells to Eliminate
When using your voice skill, watch for these patterns that signal AI involvement:
The Correlative Construction (most common):
- ❌ "X aren't just Y - they're Z"
- ❌ "It's not about X, it's about Y"
Forbidden Openers:
- ❌ "In the ever-evolving world of..."
- ❌ "Gone are the days when..."
- ❌ "Let that sink in"
Hedging Language:
- ❌ "This might help you" → "This will help you"
- ❌ "It could be argued..." → state it directly
Overused Softeners:
- Too much "just" and "actually"
- Passive voice ("was determined")
- Corporate jargon
Test: For each sentence, ask: Would this appear in ChatGPT output? If yes, rewrite it.
Using Your Voice Skill
Save to: .claude/skills/voice-[name]/SKILL.md
Invoke alongside other skills:
voice-[name] + anti-ai-writing = Authentic, humanized contentvoice-[name] + ghostwriter = Long-form pieces in voicevoice-[name] + social-content-creation = Platform-specific posts
Update as needed:
Your voice evolves. Revisit the skill quarterly or when something feels off.
Related Skills
- anti-ai-writing — Core humanization engine (use with all voice work)
- ghostwriter — For long-form content in someone else's voice
- transcript-polisher — For interview-based content
One well-crafted voice skill compounds across everything you create.
