HQ Brand Review — Voice Enforcement & Drift Detection

Paired Agent

Launch brand-reviewer agent (Task tool, model: sonnet) for execution:

• Full 6-step brand review protocol
• Multi-document brand consistency audits
• AI-generated content voice verification
• Brand drift scoring and determination

Support agents (launched alongside or after):

• communication-excellence-coach — Tone refinement after brand review flags issues
• content-strategist — Voice/tone governance across content programs

Use this skill directly (without agent) for:

• Quick banned-term spot checks (scan for known violations)
• Deciding whether content needs full review vs quick check
• Reviewing and routing completed brand review reports

When to Review vs When to Skip

CONTENT READY FOR REVIEW
  |
  +-- Is it client-facing (email, proposal, landing page, social)?
  |     YES --> Full brand review (launch brand-reviewer agent)
  |     NO  --> Continue
  |
  +-- Is it generated by AI (Claude, GPT, or any LLM)?
  |     YES --> Full brand review (AI drift is the #1 source of brand erosion)
  |     NO  --> Continue
  |
  +-- Is it internal documentation or technical spec?
  |     YES --> Skip brand review (internal docs don't need brand voice)
  |     NO  --> Continue
  |
  +-- Is it going to 10+ recipients or a public channel?
  |     YES --> Full brand review
  |     NO  --> Quick check (scan for banned terms, verify tone alignment)

Voice Attributes (Quick Reference)

The full brand guide is in the companion file. For quick checks, Sharkitect's voice is:

Determination System

After the brand-reviewer agent completes its 6-step protocol:

Anti-Patterns

1. Review Everything Equally: Reviewing internal Slack messages with the same rigor as client proposals burns 4+ hours/week on content that never reaches clients. Prioritize by audience reach and content permanence. Internal docs = skip. Client-facing with 10+ recipients = full protocol.
2. Fix During Review: The brand-reviewer evaluates and scores. It does NOT rewrite. Combining review and revision in one pass produces biased scores (the reviewer rates their own fixes favorably) and doubles the time per piece. Route revision to communication-excellence-coach or content-strategist -- separate the judge from the editor.
3. Ignore AI Content: AI-generated text is the #1 source of brand drift. Unreviewed AI content introduces an average of 3-5 banned patterns per 500 words (hedge words, filler transitions, enthusiasm inflation). Every AI output that goes client-facing must pass brand review. Skipping this single check accounts for 60%+ of brand inconsistencies in teams that use AI writing tools.
4. One-Time Review: Reviewing only new content while ignoring published material creates brand decay. Published content drifts as team voice evolves, seasonal messaging changes, and service offerings shift. Schedule quarterly consistency audits across all live client-facing content. Teams that skip quarterly audits show 15-20% brand score degradation within 6 months.
5. Skip the Score: "Feels on-brand" is not a finding. Subjective reviews produce inconsistent outcomes -- two reviewers disagree 40%+ of the time without a scoring rubric. The 6-step protocol in references/review-protocol.md exists to make brand compliance measurable and repeatable. Always score. Always record the determination.
6. Review Without the Brand Guide Open: Reviewing from memory guarantees drift in the reviewer's own standards. The brand guide contains 9 absolute bans, 5 contextual bans, and 5 voice attributes with specific ranges. No one memorizes all of these accurately. Load references/brand-guide.md every time.
7. Score Inflation for Speed: Marking content "Brand-Clear" to avoid the work of writing revision notes undermines the entire system. If a piece genuinely scores 25-30, the review takes 5 minutes. If it scores 15, writing the revision notes takes 10-15 minutes -- but publishing off-brand content costs 10x that in reader trust erosion (mere exposure effect reversal).
8. Averaging Away Problems: A piece that scores Confident=9, Direct=9, Expert=2, Approachable=8, Action-Oriented=8 averages 7.2 -- but that Expert=2 means the content makes unsubstantiated claims. One critically low dimension is worse than uniformly mediocre scores. Flag any single attribute below 4 as an automatic "Revision Required" regardless of average.

Edge Cases

Bilingual content (English + Spanish): Score each language section independently using the same protocol. Spanish business communication norms are slightly more formal -- allow +1 on formality target for Spanish sections. Check Spanish sections for direct translations of English banned terms (e.g., "sinergia", "de vanguardia", "al final del dia").

Heavily technical content (API docs, architecture specs, SOWs with technical appendices): Shift Expert target to 8-10 and relax Approachable to 4-6. Technical readers expect precision over warmth. Contextual bans on "AI-powered" and "scalable" are lifted when accompanied by specific metrics or architecture details. Default audience: Technical decision-maker.

Intentionally informal content (team celebrations, casual social, internal announcements): If the author explicitly flags "intentionally casual" or the channel is internal-only: relax all voice attribute targets by -2 points, skip contextual ban scan (absolute bans still apply), and adjust determination thresholds down by 5 points. Document the relaxation in the review report for traceability.

Multi-author content (proposals with sections from different writers): Score each section independently, then score overall consistency. Flag sections where any voice attribute differs by >3 points from the document average. Tone consistency is the highest-risk dimension here -- readers notice jarring shifts between sections even when individual sections are on-brand in isolation.

Content quoting external sources (testimonials, partner copy, regulatory text): Quoted material is exempt from brand review scoring. Mark quoted sections clearly in the review. Score only the Sharkitect-authored framing around the quotes. If quotes dominate (>50% of content), note that the piece's overall brand impression depends heavily on non-controlled text.