返回 Skill 列表
extension
分类: 内容与媒体无需 API Key

Brand Voice Enforcer

根据严格的品牓声音指南分析并重写内容,为消费品和零售品牌在语气、词汇、句法模式以及人物设定方面的一致性打分。

person作者: jakexiaohubgithub

Brand Voice Enforcer

Overview

This skill enforces brand voice consistency across all consumer-facing content — PDP copy, social posts, email campaigns, packaging copy, and customer service templates. It ingests a brand's voice framework, builds a scoring rubric, and rewrites non-conforming content while preserving factual accuracy and SEO value.

Brand voice is treated as a measurable system, not a subjective opinion. Every piece of content receives a quantified Voice Adherence Score (VAS) against the brand's defined personality dimensions.

When to Use

  • Onboarding new copywriters or agencies who need guardrails.
  • Auditing existing content libraries for voice drift after rebrands or acquisitions.
  • Adapting content across channels (packaging → digital, US → international) while preserving voice.
  • Reviewing AI-generated or user-submitted content before publication.
  • Building or updating a brand voice scoring model for automated QA pipelines.

Required Inputs

| Input | Description | Example | |---|---|---| | brand_voice_guide | The complete brand voice document or structured summary | PDF, markdown, or JSON | | voice_dimensions | 3-5 personality axes with definitions | ["Warm & Approachable", "Expert but Accessible", "Playfully Confident"] | | vocabulary_rules | Preferred terms, banned words, and substitutions | { "preferred": {"utilize": "use"}, "banned": ["cheap", "chemical-free"] } | | content_samples | 3-5 exemplar pieces that embody the ideal voice | URLs or text blocks | | input_content | The content to be evaluated and rewritten | Raw text or HTML | | channel | Target channel for format-specific norms | "Instagram caption", "PDP bullet", "email subject" | | audience_segment | Primary audience persona | "Millennial parents, health-conscious" |

Methodology

Step 1 — Voice Profile Construction

Parse the brand voice guide into a structured Voice DNA Model:

  1. Personality Dimensions: Map each dimension to a 1-5 scale with behavioral anchors.
    • Example: Warmth — 1 (clinical/detached) → 5 (conversational/intimate).
  2. Sentence Patterns: Identify preferred syntax — short declarative vs. compound; active vs. passive; question leads vs. statement leads.
  3. Lexical Fingerprint: Build a preferred vocabulary set (200-500 words) and a banned vocabulary set from the guide and exemplar content.
  4. Punctuation & Formatting Style: Em dashes vs. parentheses, exclamation point frequency, emoji policy, capitalization rules.
  5. Persona Guardrails: Define what the brand is and is not (e.g., "expert but never condescending").

Step 2 — Content Analysis & Scoring

Evaluate input content against the Voice DNA Model using the Voice Adherence Scorecard:

| Dimension | Weight | Scoring Criteria | |---|---|---| | Tone Alignment | 25% | Does emotional register match target dimensions? | | Vocabulary Compliance | 25% | Preferred terms used; banned terms absent; jargon level appropriate | | Syntax Pattern Match | 20% | Sentence length, structure, and rhythm match exemplars | | Persona Consistency | 15% | Content sounds like the defined brand persona throughout | | Channel Fit | 15% | Tone and format norms match the target channel |

Calculate a composite Voice Adherence Score (VAS) from 0-100:

  • 90-100: Publication-ready. Minor polish only.
  • 70-89: Acceptable with targeted edits. Flag specific deviations.
  • 50-69: Significant rewrite needed. Multiple dimension failures.
  • Below 50: Full rewrite. Content is off-brand.

Step 3 — Deviation Identification

For each deviation detected, produce a structured finding:

deviation:
  location: "Bullet 3, sentence 2"
  dimension: "Tone Alignment"
  severity: "major"       # minor | moderate | major
  original: "This product eliminates germs using powerful chemicals."
  issue: "Word 'chemicals' is banned; tone is clinical rather than warm."
  suggestion: "This formula wipes out 99.9% of germs with plant-powered ingredients."

Step 4 — Guided Rewrite

  1. Preserve all factual claims, keywords, and regulatory language from the original.
  2. Apply voice transformations in priority order: banned word removal → tone adjustment → syntax alignment → persona tuning.
  3. Maintain or improve readability scores (never increase grade level by more than 1).
  4. Re-score the rewritten content to confirm VAS ≥ 85.

Step 5 — Cross-Channel Adaptation

When adapting across channels, apply channel-specific voice modulations:

| Channel | Modulation | |---|---| | PDP (Amazon/Walmart) | More functional, keyword-aware; caps-led bullets | | Social (Instagram/TikTok) | Shorter sentences, emoji-permitted, conversational hooks | | Email | Personalized, benefit-first subject lines, CTA-driven body | | Packaging | Concise, legal-reviewed, regulatory claim format | | Customer Service | Empathetic, solution-oriented, first-person plural ("we") |

Output Specification

output:
  voice_adherence_score: float          # 0-100 composite VAS
  dimension_scores:
    tone_alignment: float
    vocabulary_compliance: float
    syntax_pattern_match: float
    persona_consistency: float
    channel_fit: float
  deviations: list[Deviation]           # Structured deviation findings
  rewritten_content: string             # Voice-corrected content
  rewrite_changelog: list[string]       # Summary of changes made
  confidence: float                     # Model confidence in rewrite quality

Analysis Framework

The Brand Voice Consistency Matrix evaluates voice across three layers:

  1. Surface Layer (vocabulary, punctuation, formatting) — easiest to enforce, most commonly violated.
  2. Structural Layer (sentence patterns, paragraph rhythm, information hierarchy) — requires syntactic analysis.
  3. Semantic Layer (emotional tone, persona expression, cultural resonance) — requires contextual understanding.

Enforcement priority: Surface → Structural → Semantic. Surface violations are auto-corrected; structural issues are flagged with suggestions; semantic misalignments require human review.

Examples

Brand Voice Profile: "Sunny Kitchen" — a natural food brand.

  • Dimensions: Warm (5), Playful (4), Expert (3), Premium (2).
  • Banned words: "artificial," "processed," "cheap," "stuff."
  • Preferred: "wholesome," "real ingredients," "kitchen-crafted."

Input: "Our product is manufactured using all-natural processes and contains no artificial ingredients."

Analysis: VAS = 42. "Manufactured" is clinical (Warmth violation). "All-natural" is an FDA-flagged term. Passive voice mismatches playful dimension.

Rewrite: "We craft every jar in small batches with real, wholesome ingredients — nothing artificial, ever."

New VAS: 91.

Guidelines

  • Never sacrifice regulatory accuracy for voice. Approved claims must remain verbatim even if they sound off-brand.
  • Flag any rewrite that alters a factual claim for human review.
  • When brand guides conflict with channel requirements (e.g., Amazon caps style vs. brand's lowercase preference), channel rules take precedence with a logged exception.
  • Update the Voice DNA Model quarterly or after any brand refresh.
  • Treat voice dimensions as a spectrum, not a binary — partial adherence is scored proportionally.

Validation Checklist

  • [ ] Voice DNA Model is built from the provided brand guide and exemplars.
  • [ ] All five VAS dimensions are scored independently.
  • [ ] Every deviation is logged with location, severity, and a concrete suggestion.
  • [ ] Rewritten content scores ≥ 85 VAS.
  • [ ] No factual claims, certifications, or regulatory language was altered.
  • [ ] Banned vocabulary is fully removed.
  • [ ] Readability grade level did not increase by more than 1.
  • [ ] Channel-specific formatting rules are applied.
  • [ ] Rewrite changelog is complete and auditable.
  • [ ] Final output reviewed against brand "is / is not" persona guardrails.