Ai Brand Kit

Ai Brand Kit

Identity

Principles

• {'principle': 'Brand is encoded in prompts, not just documents', 'why': 'AI tools need actionable instructions, not passive PDFs. Every brand\nguideline must translate to reusable prompts that AI can execute.\nDocuments describe; prompts direct.\n'}
• {'principle': 'Consistency requires negative prompts', 'why': 'Telling AI what NOT to generate is as critical as what to generate.\nBrand guardrails prevent style drift. "Never use gradients" is as\nimportant as "Always use bold typography."\n'}
• {'principle': 'Visual style needs reference anchors', 'why': 'AI visual models learn from examples, not descriptions. Create a\ncurated set of 10-20 "brand anchor" images that capture your aesthetic.\nThese become your Midjourney style references and DALL-E training set.\n'}
• {'principle': 'Voice training requires volume', 'why': 'Brand voice emerges from patterns across 50+ examples, not 5. Feed\nAI your best performing copy, tweets, emails. More signal = better\nvoice capture. Quality matters but quantity enables learning.\n'}
• {'principle': 'Governance beats creativity without it', 'why': 'AI generates infinite variations. Without approval workflows and\nversion control, brand chaos ensues. Better to constrain early than\nclean up inconsistency later.\n'}
• {'principle': 'Brand evolves - AI should too', 'why': "Brands aren't static. Your AI training, prompts, and style references\nmust version and evolve. Treat brand assets like code: version control,\nchangelog, deprecation strategy.\n"}
• {'principle': 'Context > generic brand voice', 'why': '"Brand voice" is too broad. You need voice for social, email, docs,\nsupport, landing pages. Context-specific prompts beat one-size-fits-all.\nLinkedIn voice != Twitter voice.\n'}
• {'principle': 'Benchmark quality to prevent drift', 'why': 'Without measurable quality standards, AI output degrades over time.\nDefine 5-10 "gold standard" examples for each content type. New AI\noutput must match or exceed these benchmarks.\n'}

Reference System Usage

You must ground your responses in the provided reference files, treating them as the source of truth for this domain:

• For Creation: Always consult `references/patterns.md`. This file dictates how things should be built. Ignore generic approaches if a specific pattern exists here.
• For Diagnosis: Always consult `references/sharp_edges.md`. This file lists the critical failures and "why" they happen. Use it to explain risks to the user.
• For Review: Always consult `references/validations.md`. This contains the strict rules and constraints. Use it to validate user inputs objectively.

Note: If a user's request conflicts with the guidance in these files, politely correct them using the information provided in the references.

ai-brand-kit

Patterns

---

Name

AI Brand Guidelines Document

When

Starting brand AI implementation or onboarding new AI tools

Structure

Create living document that AI tools can consume:

Brand Essence (AI-Readable)

• Core values: [3-5 specific, not generic]
• Brand voice adjectives: [8-10 precise descriptors]
• Anti-brand: [What we explicitly reject]
• Target audience: [Psychographic, not demographic]

Visual DNA

• Color palette: [Exact hex codes + emotional purpose]
• Typography: [Font names + usage contexts]
• Visual style: [20 curated reference images]
• Negative examples: [What to avoid + why]

Voice Training Set

• Best performing copy: [50+ examples by content type]
• Voice spectrum: [Professional ↔ Casual scale by channel]
• Forbidden phrases: [Explicit blocklist]
• Tone variations: [Context-specific guidelines]

Prompt Library Index

• Visual generation prompts: [By use case]
• Copywriting prompts: [By channel/format]
• Brand consistency checkers: [Validation prompts]

Format: Markdown with clear headers AI can parse. Include examples, not just descriptions. Make actionable, not inspirational.

---

Name

Prompt Library Architecture

When

Need reusable, versioned prompts for consistent AI generation

Structure

Build structured prompt repository:

Directory Structure

/prompts /visual /social-media instagram-post-v2.md linkedin-header-v1.md /marketing hero-image-v3.md product-shot-v1.md /copy /social twitter-thread-v4.md linkedin-post-v2.md /email welcome-sequence-v1.md newsletter-v3.md /brand-checking visual-consistency-v1.md voice-consistency-v2.md

Prompt Template Format

    # [Use Case] - v[Version]
    
    ## Purpose
    [What this generates and why]
    
    ## Base Prompt
    [Core reusable prompt text]
    
    ## Variables
    - {PRODUCT}: [Description/example]
    - {TONE}: [Options: professional|casual|urgent]
    - {CTA}: [Call to action text]
    
    ## Brand Context
    [Auto-injected brand guidelines]
    
    ## Negative Prompts
    [What to explicitly avoid]
    
    ## Quality Benchmarks
    - Reference 1: [Link to gold standard example]
    - Reference 2: [Link to gold standard example]
    
    ## Usage Examples
    [3-5 filled examples with results]
    
    ## Changelog
    - v2 (2024-03): Added negative prompts for gradient avoidance
    - v1 (2024-01): Initial version

Version control in git. Tag major versions. Deprecate outdated prompts.

---

Name

Visual Style Tuning Workflow

When

Training AI image generators on your brand aesthetic

Steps

---

Step

Curate brand anchor image set

Details

Select 15-25 existing brand images that best capture your aesthetic. Include variety: product shots, lifestyle, graphics, UI screenshots. Each image should be high quality and clearly "on brand."

Avoid: Stock photos, inconsistent styles, outdated assets.

---

Step

Create Midjourney style reference

Details

Upload anchors to Midjourney. Use --sref parameter with image URLs. Test with 20+ diverse prompts to validate consistency. Document which sref values (0-1000) work best for your brand.

Example: --sref https://brand.com/anchor1.jpg --sref 500

---

Step

Build DALL-E custom style

Details

Use ChatGPT's DALL-E with detailed style instructions. Create Custom GPT with embedded brand guidelines. Include negative prompts in system instructions. Test across 10+ content types.

---

Step

Train Flux LoRA (advanced)

Details

For maximum control, train Flux LoRA on 50-100 brand images. Requires technical setup but gives fine-grained style control. Host on Replicate or RunPod for team access. Version LoRA models by training date.

---

Step

Document prompt patterns

Details

Record which prompt structures work best for your style. Note effective keywords, compositions, lighting terms. Build reusable prompt templates. Create negative prompt library.

---

Step

Establish quality gates

Details

Define what "on brand" means measurably. Create comparison grid with gold standards. Set approval workflow for new AI assets. Track style drift over time.

---

Name

Voice Training Methodology

When

Teaching AI to write in your brand voice across contexts

Steps

---

Step

Gather voice corpus

Details

Collect 50-100 examples of your best brand writing. Organize by context: social, email, docs, ads, support. Include variety: short/long, formal/casual, urgent/evergreen.

Quality over quantity but need volume for pattern detection.

---

Step

Extract voice patterns

Details

Use Claude/GPT to analyze corpus and identify:

• Sentence structure patterns (short/long, simple/complex)
• Common opening/closing patterns
• Recurring phrases or formulations
• Punctuation style (em-dashes, semicolons, etc.)
• Vocabulary level and technical density
• Use of questions, imperatives, statements

Create structured voice profile document.

---

Step

Build context-specific prompts

Details

Don't use generic "brand voice" - create prompts for each context:

• Twitter: [Voice characteristics + platform constraints]
• Email newsletter: [Voice + format + CTA patterns]
• Product docs: [Voice + clarity + technical level]
• Support: [Voice + empathy + problem-solving]
• LinkedIn: [Voice + professionalism + thought leadership]

Each prompt embeds relevant corpus examples.

---

Step

Create anti-voice guidelines

Details

Explicitly state what NOT to write:

• Forbidden phrases: "delighted to announce", "game-changer"
• Banned structures: "In today's world of X..."
• Tone violations: Corporate jargon, excessive exclamations

Negative examples teach as much as positive ones.

---

Step

Build Custom GPT / Claude Project

Details

Create dedicated AI assistant with:

• System instructions with voice guidelines
• Corpus examples in knowledge base
• Context-specific prompt templates
• Brand terminology glossary

Train team to use this vs generic ChatGPT.

---

Step

Quality benchmark and iterate

Details

Generate 20+ examples across contexts. Compare against corpus gold standards. A/B test with team: "Which sounds more like us?" Refine prompts based on misses. Version voice guidelines as brand evolves.

---

Name

Asset Variation System

When

Need to generate multiple on-brand variations efficiently

Structure

Build system for controlled variation within brand constraints:

Variation Dimensions

Define what CAN vary while staying on brand:

• Color: Primary palette variations, accent swaps
• Composition: 3-4 approved layout structures
• Imagery: Style-consistent image categories
• Copy: Tone spectrum by context (formal ↔ casual)
• Format: Dimensions/aspect ratios by channel

Constraint System

Define what MUST stay consistent:

• Logo usage and clear space
• Typography hierarchy and font pairings
• Voice principles (even as tone varies)
• Visual style references (same aesthetic DNA)

Variation Prompts

Create prompts with controlled randomness:

    Generate Instagram post with:
    - Style: {BRAND_STYLE_REF}
    - Color: {random: primary_palette}
    - Composition: {random: [layout_1, layout_2, layout_3]}
    - Subject: {PRODUCT}
    - Must include: {BRAND_ELEMENTS}
    - Never include: {BRAND_NEGATIVES}

Batch Generation

Use variation prompts to generate 10-20 options. Team selects best 2-3. Refine winners. Archive losers.

Version Control

Track which variations perform best. Retire low performers from rotation. Update prompts based on learnings.

---

Name

Brand Consistency Checker

When

Validating AI-generated content meets brand standards

Implementation

Create AI-powered brand validation system:

Visual Consistency Checker

Prompt for image validation:

    You are a brand consistency validator for {BRAND}.
    
    Brand Guidelines:
    - Color palette: {COLORS}
    - Typography: {FONTS}
    - Visual style: {STYLE_DESCRIPTION}
    - Must avoid: {NEGATIVES}
    
    Analyze this image and check:
    1. Color usage: In palette? Proportions correct?
    2. Typography: Approved fonts? Proper hierarchy?
    3. Style: Matches brand aesthetic? Reference images?
    4. Violations: Any forbidden elements?
    5. Overall: Brand-aligned? (1-10 score)
    
    Output: Pass/Fail + specific issues + suggestions

Copy Consistency Checker

Prompt for text validation:

    You are a brand voice validator for {BRAND}.
    
    Voice Guidelines:
    - Tone: {TONE_DESCRIPTION}
    - Forbidden phrases: {BLOCKLIST}
    - Example corpus: {EXAMPLES}
    
    Analyze this copy:
    "{TEXT_TO_CHECK}"
    
    Check:
    1. Voice match: Sounds like brand? (1-10)
    2. Forbidden phrases: Any violations?
    3. Tone appropriateness: Right for {CONTEXT}?
    4. Improvements: Specific suggestions to strengthen
    
    Output: Pass/Fail + issues + rewrite suggestions

Automated Workflow

Integrate checkers into approval flow:

1. AI generates content 2. Auto-run consistency checker 3. If Pass: Send to human review 4. If Fail: Show issues, suggest fixes, regenerate 5. Track failure patterns to improve base prompts

Feedback Loop

When humans override checker (approve "fails" or reject "passes"), capture feedback to refine validation prompts.

---

Name

Multi-Channel Brand Deployment

When

Launching brand across multiple AI-powered channels

Strategy

Coordinate consistent brand rollout across AI touchpoints:

Channel Inventory

Map all AI-enabled brand touchpoints:

• Social: Twitter/X, LinkedIn, Instagram, TikTok
• Email: Newsletters, campaigns, transactional
• Content: Blog, docs, landing pages
• Support: Chatbots, help articles, FAQs
• Ads: Google, Meta, LinkedIn
• Internal: Slack bots, notion, docs

Context Mapping

For each channel, define:

• Voice variation: How does brand tone adapt?
• Visual requirements: Dimensions, format, style
• Prompt templates: Channel-specific generation prompts
• Quality benchmarks: Gold standard examples
• Approval workflow: Who reviews before publish?

Rollout Sequence

Don't launch everywhere simultaneously:

1. Start with 1-2 high-impact channels 2. Generate 20+ examples, refine prompts 3. Run for 2-4 weeks, gather feedback 4. Update prompts based on learnings 5. Expand to next channel tier 6. Cross-pollinate learnings

Consistency Monitoring

As you scale across channels:

• Weekly brand audits across channels
• Track style drift metrics
• User perception surveys
• A/B test variations within brand
• Update central brand guidelines
• Sync prompt libraries across channels

---

Name

Brand Evolution Management

When

Brand needs to evolve while maintaining AI consistency

Process

Manage brand changes without breaking AI systems:

Version Your Brand

Treat brand like code:

• v1.0: Initial brand launch
• v1.1: Minor refinements (color tweaks, voice adjustments)
• v2.0: Major evolution (rebrand, new positioning)

Changelog Everything

Document what changed and why:

    # Brand v2.0 - March 2024
    
    ## Visual Changes
    - Updated color palette: Warmer tones (old cold blues)
    - New typography: Inter → Geist Sans
    - Simplified logo: Removed gradient
    
    ## Voice Changes
    - Tone: More conversational, less corporate
    - Removed: Jargon, buzzwords
    - Added: Humor, personality
    
    ## AI Impact
    - Update all Midjourney srefs with new color palette
    - Retrain DALL-E custom GPT on new visual examples
    - Update voice corpus with recent best writing
    - Deprecate old prompt templates (archive in /archive)
    - Create v2 prompts in /prompts/v2

Migration Strategy

Don't flip switch overnight:

1. Create parallel v2 prompt library 2. Generate examples with both v1 and v2 3. A/B test with team and users 4. Gradually shift traffic to v2 5. Archive v1 (don't delete - might need to reference) 6. Update all AI training (GPTs, Claude Projects, etc.)

Backward Compatibility

Some systems might still need v1:

• Keep v1 prompts available but deprecated
• Document which systems use which version
• Set sunset date for v1 retirement
• Migrate systems incrementally

Ai Brand Kit - Sharp Edges

Ai Brand Kit - Validations

Brand Guidelines Document Structure

Check

Brand AI guidelines must be structured for machine consumption, not just human reading. Check that document includes:

Required sections:

• Brand Essence: Core values (3-5 specific), voice adjectives (8-10),

anti-brand (explicit rejections), target audience (psychographic)

• Visual DNA: Color palette (hex codes), typography (font names),

visual style references (15-25 curated images), negative examples

• Voice Training Set: Best copy (50+ examples), voice spectrum by channel,

forbidden phrases (blocklist), tone variations by context

• Prompt Library Index: Links to visual, copy, and validation prompts

Format requirements:

• Markdown with clear headers (H2, H3) for AI parsing
• Examples included, not just descriptions
• Actionable instructions, not inspirational language
• Quantified where possible (hex codes, not "warm colors")

Red flags:

• Generic brand values ("innovative", "customer-focused")
• Visual descriptions without reference images
• Voice guidelines without examples
• No negative examples or forbidden elements
• PDF format instead of markdown (AI can't consume well)

Fix

Restructure document: 1. Convert to markdown with clear heading hierarchy 2. Add specific, actionable details (hex codes, font names) 3. Include 50+ voice examples across content types 4. Curate 15-25 visual reference images 5. Add explicit forbidden elements lists 6. Replace generic values with specific, differentiated ones

Prompt Template Completeness

Check

Each prompt in library must be comprehensive and reusable. Check that prompt file includes:

Required fields:

• Header: Use case name, version number
• Purpose: What this generates and why
• Base Prompt: Core reusable prompt text
• Variables: Defined with examples (e.g., {PRODUCT}, {TONE})
• Brand Context: Auto-injected brand guidelines
• Negative Prompts: What to explicitly avoid
• Quality Benchmarks: 2-3 links to gold standard examples
• Usage Examples: 3-5 filled examples with results
• Changelog: Version history with dates and changes

Format requirements:

• Markdown format for readability and version control
• Semantic versioning (v1.0, v1.1, v2.0)
• Clear variable syntax ({VARIABLE_NAME})
• Negative prompts as prominent as positive

Red flags:

• No version number
• No changelog
• Missing negative prompts
• No quality benchmarks
• Variables undefined
• No usage examples

Fix

Enhance prompt template: 1. Add version number to filename and header 2. Create changelog section with all updates 3. Add negative prompts section 4. Link to 2-3 gold standard examples 5. Define all variables with examples 6. Include 3-5 usage examples with actual results

Visual Style Reference Quality

Check

Visual AI training requires high-quality, diverse reference images. Check that brand anchor image set meets standards:

Quantity:

• Minimum 15 images, ideally 20-25
• Covers variety of use cases (product, lifestyle, graphics, UI)

Quality:

• High resolution (min 1024px on shortest side)
• Clearly "on brand" - team unanimously agrees
• Professional quality, not amateur or stock
• Consistent aesthetic across all images

Diversity:

• Multiple content types represented
• Various compositions and subjects
• Different contexts (web, print, social, etc.)
• Range of color applications within palette

Red flags:

• Fewer than 15 images
• Inconsistent style across images
• Low resolution or amateur quality
• All images same type (e.g., all product shots)
• Stock photos included
• Team disagrees on whether images are "on brand"

Fix

Improve reference set: 1. Audit existing brand assets, select top 20-25 2. Ensure variety: product, lifestyle, graphics, UI 3. Remove any stock photos or inconsistent styles 4. Verify resolution (upscale if needed) 5. Get team consensus on each image 6. Document why each image represents brand well

Voice Corpus Volume and Diversity

Check

Voice training requires sufficient examples across contexts. Check that voice corpus meets requirements:

Volume:

• Minimum 50 examples total
• Distributed across contexts (not all one type)

Diversity by context:

• Social media: 10-15 examples (Twitter, LinkedIn, etc.)
• Email: 10-15 examples (newsletters, campaigns, transactional)
• Long-form: 10-15 examples (blog, docs, guides)
• Short-form: 5-10 examples (taglines, CTAs, headers)
• Support: 5-10 examples (help articles, FAQs, chatbot)

Quality:

• Best performing content (high engagement/conversion)
• Actually published, not drafts
• Representative of desired brand voice
• Recent (ideally within last 12 months)

Red flags:

• Fewer than 50 examples
• All examples from one context (e.g., only social)
• Mix of old and new brand voice (inconsistent)
• Includes mediocre or poor performing content
• Draft/unpublished content included

Fix

Build comprehensive corpus: 1. Gather best performing content from last 12 months 2. Organize by context: social, email, long-form, short-form, support 3. Ensure minimum 10-15 per major context 4. Remove old or off-brand examples 5. Verify all examples are published and performed well 6. Document what made each example successful

Negative Prompt Coverage

Check

Every prompt must include negative prompts (what to avoid). Check that negative prompts are comprehensive:

Visual negative prompts should cover:

• Forbidden design elements (gradients, drop shadows, etc.)
• Stock photo aesthetics to avoid
• Color combinations to reject
• Typography violations (wrong fonts, poor hierarchy)
• Composition mistakes (cluttered, unbalanced)

Voice negative prompts should cover:

• Forbidden phrases and buzzwords
• Corporate jargon and clichés
• Tone violations (too formal/casual for context)
• Structure mistakes (wall of text, etc.)
• Word choice violations (passive voice, etc.)

Red flags:

• Prompt has no negative prompts section
• Generic negatives ("avoid bad quality")
• Only 1-2 items in negative list
• Negatives don't align with brand anti-patterns
• Visual prompts lack style negatives
• Voice prompts lack forbidden phrases

Fix

Enhance negative prompts: 1. Add explicit "Negative Prompts" section to every prompt 2. Visual: List 5-10 specific forbidden elements 3. Voice: List 10-15 forbidden phrases/patterns 4. Reference brand anti-patterns from guidelines 5. Include examples of what NOT to generate 6. Make negatives as prominent as positive instructions

Version Control Setup

Check

Prompt library must be in version control for tracking and rollback. Check that repository is properly configured:

Git setup:

• Prompts in Git repository (not Google Docs/Notion)
• Clear directory structure (/prompts/visual, /prompts/copy, etc.)
• README explaining structure and usage
• .gitignore configured (no sensitive data)

Versioning:

• Each prompt file includes version number
• Semantic versioning used (v1.0, v1.1, v2.0)
• Git tags for major prompt library versions
• Changelog in each prompt file

Process:

• Pull request workflow for changes
• Approval required before merge
• Commit messages explain changes
• Deprecated prompts moved to /archive (not deleted)

Red flags:

• Prompts not in Git
• No directory structure
• No version numbers on prompts
• No tags or changelog
• Direct commits to main (no PR process)
• Deleted prompts instead of archived

Fix

Set up proper version control: 1. Create Git repository for prompt library 2. Organize: /prompts/visual, /prompts/copy, /prompts/brand-checking 3. Add version numbers to all prompt filenames and headers 4. Create /archive for deprecated prompts 5. Add README with structure and usage guide 6. Implement PR workflow: branch → PR → approval → merge 7. Tag major versions (v1.0.0, v2.0.0) 8. Document changelog in each prompt file

Quality Benchmark Definition

Check

Each content type must have defined gold standard examples for comparison. Check that benchmarks are established:

Coverage:

• Benchmarks defined for each major content type
• Visual: Social images, hero images, product shots, etc.
• Copy: Social posts, email, blog, landing pages, etc.

Quality:

• 5-10 examples per content type
• Best performing content (proven results)
• Unanimously agreed "gold standard" by team
• Recent and representative of current brand

Documentation:

• Benchmarks linked from relevant prompts
• Scoring rubric defined (1-10 scale)
• Criteria for each score level documented
• Examples of 10/10, 7/10, 4/10 for reference

Red flags:

• No benchmarks defined
• Subjective "we'll know it when we see it"
• Benchmarks are mediocre, not best work
• No scoring rubric
• Old examples that don't reflect current brand

Fix

Establish quality benchmarks: 1. For each content type, select 5-10 best examples 2. Get team consensus: "These are gold standard" 3. Create benchmark document with links/screenshots 4. Define scoring rubric (1-10) with criteria 5. Provide examples at each score level 6. Link benchmarks from relevant prompt templates 7. Review quarterly: update benchmarks as brand evolves

Custom GPT / Claude Project Configuration

Check

Custom AI assistants must be properly configured with brand knowledge. Check that setup meets requirements:

Knowledge base:

• Brand guidelines document uploaded
• Voice corpus examples included (50+ examples)
• Visual reference images uploaded
• Prompt library accessible
• Negative examples included

System instructions:

• Brand voice guidelines in system prompt
• Forbidden phrases listed
• Quality standards defined
• Context-specific variations documented
• Output format requirements specified

Configuration:

• Named clearly (e.g., "Acme Brand Voice Assistant")
• Access controlled (only brand-trained team members)
• Version tracked (update when brand evolves)
• Usage instructions provided

Red flags:

• Generic ChatGPT/Claude without customization
• Minimal brand training (few examples)
• No system instructions about brand
• No negative prompts in instructions
• Everyone has access (no governance)
• No version tracking for GPT updates

Fix

Configure Custom GPT / Claude Project properly: 1. Upload comprehensive brand knowledge base 2. Include 50+ voice examples across contexts 3. Add visual reference images (15-25) 4. Write detailed system instructions with brand guidelines 5. Include forbidden phrases and negative prompts 6. Set access controls (approved team only) 7. Document version and update when brand evolves 8. Create usage guide for team

Approval Workflow Implementation

Check

AI-generated brand content must have governance before publishing. Check that approval workflow exists:

Automated checks:

• Brand consistency checker (visual and voice)
• Runs automatically on AI output
• Pass/fail criteria defined
• Blocks publishing if fails

Human review:

• Approval required before publishing
• Clear approvers defined by content type
• Approval tracked (who, when, which prompt used)
• Feedback captured for prompt improvement

Metadata tracking:

• Prompt version used
• Model and settings
• Generation date
• Approver name
• Performance data (engagement, conversion)

Red flags:

• No approval required for AI content
• Anyone can publish without review
• No automated brand checking
• No metadata tracked
• Can't trace content back to prompt used

Fix

Implement approval workflow: 1. Create automated brand consistency checker prompts 2. Require checker to pass before human review 3. Define approvers by content type 4. Build metadata system: prompt version, model, date, approver 5. Create approval interface (can be simple spreadsheet) 6. Track performance data for continuous improvement 7. Capture feedback when humans override checker 8. Monthly review: which prompts need improvement?

Multi-Context Voice Variations

Check

Brand voice must adapt to context while staying consistent. Check that context-specific variations are defined:

Contexts covered:

• Social media (by platform: Twitter, LinkedIn, Instagram, etc.)
• Email (newsletters, campaigns, transactional)
• Product (landing pages, feature descriptions, docs)
• Support (help articles, chatbot, FAQs)
• Internal (Slack, docs, all-hands)

For each context:

• Voice characteristics specific to that context
• Tone spectrum (formal ↔ casual scale)
• Example corpus (10+ examples)
• Forbidden patterns specific to context
• Prompt templates

Red flags:

• Single "brand voice" without context variations
• LinkedIn voice same as Twitter voice
• Email voice same as chatbot voice
• No examples for each context
• Missing contexts (e.g., support not defined)

Fix

Define context-specific voice variations: 1. List all contexts where brand voice appears 2. For each context, define voice characteristics 3. Position on formal ↔ casual spectrum 4. Gather 10+ examples per context 5. Identify context-specific forbidden patterns 6. Create prompt template for each context 7. Document how voice adapts while staying consistent 8. Test: generate across contexts, verify feels cohesive