Editorial Voice Matching

Match article output to a target publication's editorial voice using reference pairs and distilled principles.

Overview

Every publication has a voice — the combination of sentence rhythm, vocabulary choices, humor calibration, information density, and structural patterns that make it recognizably itself. This skill provides a two-layer approach to capturing and reproducing that voice: concrete reference pairs (showing the transformation) and distilled principles (the rules behind the transformation).

Core Value: Reproducible voice consistency across articles and writers, without requiring the original author.

Differentiator from brand_guidelines.md: The marketing-team's brand_guidelines.md defines brand personality archetypes (Expert, Friend, Innovator) for marketing content. This skill captures the specific editorial voice of a publication — not what personality to project, but how this particular publication writes: its sentence length distribution, its approach to humor, its vocabulary register, its information density per paragraph.

Core Capabilities

• Two-Layer Voice Capture — Reference pairs (concrete before/after) + distilled principles (abstract rules)
• Voice Extraction Process — Analyze 10+ reference editions to build a voice profile
• 6 Voice Dimensions — Sentence rhythm, vocabulary register, humor calibration, opening/closing patterns, attribution style, information density
• Prompt Patterns — Structured prompts for voice-matched generation
• Voice Comparison — Compare two texts for voice consistency

Quick Start

1. Gather 10+ reference editions of the target publication
2. Use the voice analysis template at references/voice-analysis-template.md
3. Extract reference pairs and distilled principles
4. Apply the voice profile when drafting new articles

Key Workflows

1. Voice Extraction (One-Time Setup)

1. Gather references — Collect 10+ editions of the target publication
2. Analyze each edition across 6 dimensions:
   • Sentence rhythm — Average sentence length, variation pattern, use of fragments
   • Vocabulary register — Formal/informal ratio, jargon handling, colloquialism frequency
   • Humor calibration — Type (dry, sarcastic, wordplay), frequency, placement
   • Opening/closing patterns — How articles begin and end (hook types, sign-off style)
   • Attribution style — How sources are cited ("according to," "X says," "per X")
   • Information density — Facts per paragraph, detail level, explanation depth
3. Extract reference pairs — For each dimension, capture a before (generic) and after (publication voice) example
4. Distill principles — From the reference pairs, extract 3-5 rules per dimension
5. Compile voice profile — Use the style guide skeleton at references/style-guide-skeleton.md

2. Voice Application (Per Article)

1. Load voice profile — Read the distilled principles and reference pairs
2. Draft article — Write content following the factual requirements (script-to-article)
3. Apply voice layer — Revise the draft against each dimension:
   • Adjust sentence rhythm to match distribution
   • Align vocabulary register
   • Calibrate humor (add, remove, or adjust type)
   • Match opening/closing patterns
   • Align attribution style
   • Match information density
4. Verify — Compare a sample paragraph against reference pairs for voice similarity

3. Voice Consistency Check (Adversarial)

1. Read the article and the voice profile
2. Score each dimension (0-100) against the voice profile
3. Flag deviations — Passages that break voice consistency
4. Suggest corrections — Specific rewrites to restore voice alignment

Best Practices

• Extract voice from the publication's best editions, not average ones
• Update reference pairs quarterly as the publication's voice evolves
• Voice matching is a post-transformation step — get the facts right first, then apply voice
• When voice matching conflicts with clarity, clarity wins
• Don't over-match — slight natural variation keeps writing from sounding robotic

Voice Profile (Standard Artifact)

The voice profile (.voice-profile.md) is the standard output of the extraction process and the standard input for all voice-related operations. It contains YAML frontmatter (for machine parsing) and a markdown body (for human reading and agent consumption).

Generate: /voice/extract <editions-dir> --name "Publication Name"
Consume: Pass via --voice-profile <path> to /newsletter/generate, /review/editorial-review, or any agent using this skill.

The profile is publication-scoped and reusable — extract once, use for every edition. Re-extract quarterly or when the publication's voice evolves.

See references/voice-profile-template.md for the full template structure.

Reference Guides

• voice-profile-template.md — Standard voice profile format (output of /voice/extract, input to all voice consumers)
• voice-analysis-template.md — Template for analyzing a single edition across 6 dimensions
• style-guide-skeleton.md — Template for compiling a complete voice profile (superseded by voice-profile-template for new work)

Integration

Used by editorial-writer as a post-transformation voice layer. Consumed adversarially by voice-consistency-reviewer for voice compliance checking.