Brand Voice and Content Style Guide

Originally published at thatdevpro.com. This framework reference is part of the 14-tier Engine Optimization stack from ThatDevPro, an SDVOSB-certified veteran-owned web + AI engineering studio. You are reading the dev.to mirror; the source-of-truth canonical version with embedded validation tools lives at the link above.

The Voice Playbook That Connects Brand Identity to Search and AI Citation Outcomes

A comprehensive installation and audit reference for documenting brand voice, codifying a content style guide, and maintaining consistent voice at scale across every content surface. This document treats voice as an SEO and AI citation signal, not only a marketing artifact. Voice attributes, tone variation rules, persona alignment, multi author workflows, AI assisted content guardrails, AI engine voice detection mechanics, voice and E-E-A-T overlap, voice and AI Overview citation patterns, common anti patterns, and a tiered audit rubric are all specified here. Dual purpose: installation manual and audit document.

Cross stack implementation note: the patterns in this framework are written in plain HTML and YAML for clarity. For React, Vue, Svelte, Next.js, Nuxt, SvelteKit, Astro, Hugo, 11ty, Remix, WordPress, Shopify, and Webflow equivalents of voice metadata, byline blocks, and editorial markers, see framework-cross-stack-implementation.md. For pure client rendered SPAs (no SSR/SSG) see framework-react.md. For Tailwind specific concerns related to voice typography (prose classes, reading width, dark mode contrast) see framework-tailwind.md.

1. Document Purpose

The canonical reference for documenting and maintaining a brand voice on a website in a way that survives both human editorial review and AI engine voice detection. Voice is the constant. Tone is the variable. A site without a documented voice produces inconsistent content that erodes brand recognition for humans and degrades signal coherence for the language models that decide whether to cite a source. This framework closes the gap that the existing content frameworks (framework-hcs.md, framework-infogain.md, framework-eeat.md, framework-contentaudit.md, framework-contentfirst.md) leave open: they specify what content contains and how it is structured but say almost nothing about how it sounds.

The omission matters more in 2026 than it did in 2020. AI engines have become voice readers. They detect formality, persuasion style, hedge density, perspective markers, and persona signals as features that feed into citation decisions; they reward authentic voice as a Trust component of E-E-A-T. A documented voice rendered consistently across surfaces compounds with every other framework in this library. A voice patched together by multiple authors and AI tools without guardrails undermines them.

1.1 What This Document Is

Installation and audit reference for brand voice as an SEO signal. Covers the four dimensions of voice (Nielsen Norman Group, 2016), the voice attributes worksheet, tone variation per surface, the brand archetype framework, complete style guide anatomy, editorial workflow at scale, AI engine voice detection mechanics, voice and E-E-A-T overlap, voice and AIO citation phrasing patterns, the top ten anti patterns, the audit rubric (per surface, site wide, first 90 days), and report templates.

1.2 Three Operating Modes

Mode A, Install Mode: Build voice playbook and style guide. Follow Sections 2 through 14 in order. Generate implementation report at end.

Mode B, Audit Mode: Score existing voice against this framework. Skip to Section 13. Document gaps. Return to Mode A for failing items.

Mode C, Hybrid Mode: Audit then install. Common when a brand has partial documentation. Run Mode B to baseline, run Mode A on failing items.

1.3 How Claude Code CLI Should Consume This Document

1. Read Section 2 to collect client variables (voice documentation state, surfaces, multi author and AI use). Do not proceed until intake is complete.
2. Read Section 3 for the four dimensions of voice (NN/g framework).
3. Run Section 4 worksheet to produce three to five voice attributes with do/don't examples.
4. Apply Section 5 to map tone variation by content surface.
5. Install Sections 6 through 8: brand persona/ICP alignment, complete style guide, editorial workflows for multi author and AI assisted content.
6. Apply Section 9 for AI engine voice detection.
7. Apply Sections 10 and 11 for E-E-A-T Trust signals and AIO citation patterns.
8. Validate using Section 13.
9. Generate the report using Section 14.

1.4 Conflict Resolution Rules

When installing into a site with partial guidance, apply in priority order:

When in doubt, flag for manual review rather than overwriting client owned brand assets.

1.5 Required Tools

• Documented intake interview with brand owner (honest, not aspirational)
• Content sample of 15 to 30 existing pieces across surfaces, dated and tagged by author
• Spreadsheet or YAML file for the voice attributes worksheet
• Markdown editor for the style guide as a living document
• Grammar and consistency checker (Grammarly Business, Acrolinx, Writer.com)
• AI voice trainer or prompt library (Writer.com voice profiles, Jasper Brand Voice, custom Claude/GPT system prompts)
• Search Console and AI Overview tracking tools (framework-aioverviews.md) to correlate voice changes with citation outcomes

2. Client Variables Intake

brand_voice_intake:

  # Business Identity (REQUIRED)
  business_name: ""
  primary_domain: ""
  business_type: ""                        # LocalBusiness | ProfessionalService | ECommerce | Media | SaaS | etc.
  business_industry: ""
  founder_or_owner_name: ""                # Whose voice the brand is closest to
  founded_year: ""

  # Current Voice Documentation State (REQUIRED)
  has_documented_voice: false
  has_complete_style_guide: false
  voice_documentation_age_years: 0
  voice_documentation_format: ""           # "none" | "google_doc" | "pdf" | "wiki" | "notion" | "markdown_repo"
  voice_documentation_owner: ""
  voice_documentation_last_review: ""

  # Voice Self Description (REQUIRED)
  brand_in_three_adjectives: []
  brand_is_NOT_in_three_adjectives: []
  brand_compared_to_a_person: ""           # One sentence
  brand_compared_to_a_celebrity: ""        # Optional
  brand_compared_to_a_competitor_we_admire: ""
  brand_compared_to_a_competitor_we_avoid: ""

  # Audience (REQUIRED)
  primary_audience_description: ""
  audience_expertise_level: ""             # "beginner" | "intermediate" | "expert" | "mixed"
  audience_emotional_state_default: ""     # "curious" | "anxious" | "skeptical" | "celebratory" | "transactional"
  audience_relationship_to_brand: ""       # "first_touch" | "shopping" | "purchased" | "advocate" | "mixed"
  audience_industry_jargon_tolerance: ""   # "high" | "medium" | "low"

  # Content Surfaces in Scope (REQUIRED): set true for any surface in play
  content_surfaces:
    homepage: false
    service_or_product_pages: false
    blog_articles: false
    long_form_guides: false
    faq_pages: false
    pricing_pages: false
    about_us: false
    case_studies: false
    email_marketing: false
    transactional_email: false
    in_app_messaging: false
    social_media_organic: false
    paid_ad_copy: false
    help_center_or_kb: false
    support_macros: false
    press_releases: false
    legal_pages: false
    error_states_and_404: false

  # Multi Author and AI Assisted Use (REQUIRED)
  number_of_active_writers: 0
  internal_writers_count: 0
  external_freelance_writers_count: 0
  agency_or_outsourced_writers_count: 0
  uses_ai_assisted_drafting: false
  ai_tools_used: []                        # ["claude", "chatgpt", "writer_com", "jasper", "custom"]
  ai_review_quality_self_rating: ""        # 1-10
  ai_voice_profile_configured: false
  editorial_review_process_documented: false
  publication_cadence: ""                  # "daily" | "weekly" | "biweekly" | "monthly" | "ad_hoc"

  # Inconsistency Symptoms (RECOMMENDED)
  internal_team_reports_voice_drift: false
  audience_feedback_mentions_voice_change: false
  support_uses_different_voice_than_marketing: false
  ceo_voice_differs_from_brand_voice: false
  old_content_sounds_like_a_different_company: false

  # AI Engine Voice Detection (RECOMMENDED)
  brand_currently_cited_by_chatgpt: false
  brand_currently_cited_by_perplexity: false
  brand_currently_cited_by_claude: false
  brand_currently_cited_by_aio: false
  brand_voice_distinguishable_from_competitors: false

  # E-E-A-T Voice Overlap (REQUIRED)
  authors_use_first_person_consistently: false
  authors_credentialed_in_topics_they_write: false
  brand_owner_voice_audible_in_content: false
  content_demonstrates_first_hand_experience: false
  content_acknowledges_limitations: false

The first two blocks establish identity and documentation maturity. Voice self description and audience define the target attributes. Surfaces define scope. Multi author and AI use define the governance problem. Inconsistency symptoms, AI engine detection, and E-E-A-T overlap define the outcomes the voice work is meant to influence.

3. What Brand Voice Is

Voice is the consistent personality of the brand expressed in language. Tone is the situational dialing of that personality up or down depending on context. Voice is the constant. Tone is the variable. The distinction is the foundational move in every credible style guide (Mailchimp, 18F, Atlassian, Slack).

The 18F Content Guide: "Voice is a constant, but tone is a variable. Voice is an organization's unique personality... tone is analogous to a person's tone of voice, varying based on context" (18F Content Guide, 2015, current 2026; https://guides.18f.gov/content-guide/our-style/voice-and-tone/). Mailchimp: "you have the same voice all the time, but your tone changes depending on context and the emotional state of who you're addressing" (Mailchimp Content Style Guide, "Voice and Tone," 2026; https://styleguide.mailchimp.com/voice-and-tone/).

The distinction makes documentation tractable. A voice playbook documents one voice with three to five attributes. A tone matrix documents how that voice flexes across situations. Without the distinction, teams produce either a fixed monolithic voice that becomes inappropriate in empathy or celebration moments, or chaotic situational voice that loses brand recognition.

3.1 The Four Dimensions of Voice (Nielsen Norman Group, 2016)

Nielsen Norman Group's original 2016 framework decomposes any piece of content's tone into four orthogonal dimensions, each a continuum between two poles. The framework is the foundational analytical tool used by every serious UX writing practice in 2026 (NN/g, "The Four Dimensions of Tone of Voice," 2016, current; sample frame analyzed across product UI, marketing site, transactional email, and support content; https://www.nngroup.com/articles/tone-of-voice-dimensions/).

Every voice is a point in this four dimensional space. Mailchimp lands near (casual, mildly funny, respectful, mildly enthusiastic). The Wall Street Journal lands near (formal, serious, respectful, matter of fact). Wendy's social media lands near (casual, very funny, irreverent, enthusiastic). The framework's value is that it forces explicit choice rather than vague tag soup.

3.2 Voice, Tone, Style: Three Distinct Terms

Voice is the brand's permanent personality. Constant across content type. "We are warm, plainspoken, and confident." Tone is situational adjustment within the voice envelope. "In error states we are calm and helpful; in announcements we are excited and clear." Style is the rules: grammar, capitalization, terminology, punctuation, formatting standards. "Sentence case in headlines. Spell ecommerce without a hyphen."

A complete style guide documents all three: voice in three to five attributes; tone in a matrix of situations to tonal moves; style in a reference appendix of rules.

3.3 Why The Distinction Matters For SEO

The voice/tone distinction is not only editorial discipline; it is the structural prerequisite that makes voice surveillable by AI engines. A brand whose voice attribute is "confident" must produce confident phrasing consistently, and confident phrasing is declarative, not hedged. AI engines preferentially cite declarative sentences (citation studies, Section 11). When voice is documented and tone variation is also documented, the same brand can be confident in launch copy and gentle in apology copy without losing its core voice. Both modes serve extraction and E-E-A-T.

4. Voice Attributes Documentation

The canonical voice documentation deliverable is the voice attributes worksheet. Three to five attributes, each with a one sentence definition and three to six concrete do/don't examples drawn from the actual content surfaces.

4.1 The Worksheet Structure

For each of the three to five attributes, document: name (one word or short phrase, e.g. "Plainspoken"); definition (one sentence, e.g. "We strip jargon and say what we mean in everyday language"); why this attribute (one sentence audience or business reason); do_examples (three to six concrete phrases or paragraph examples); dont_examples (three to six near brand phrases that fall outside voice).

Three is the minimum. Five is the maximum. More than five and authors cannot hold them in working memory. Fewer than three and the voice is too thin to differentiate.

4.2 Why Do/Don't Examples Are Non Negotiable

Oxford College of Marketing reported (Aug 2025) that 64% of content marketers have documented brand voice guidelines but only 23% use them to train their AI tools; abstract guidelines without examples are the named reason (Oxford College of Marketing, "AI Brand Voice Guidelines," 2025, citing CMI 2025 B2B survey n=980; https://blog.oxfordcollegeofmarketing.com/2025/08/04/ai-brand-voice-guidelines-keep-your-content-on-brand-at-scale/; CMI research at https://contentmarketinginstitute.com/b2b-research/b2b-content-marketing-trends-research-2025). AI tools cannot parse "be friendly but not too friendly." Human writers interpret abstract guidelines differently. Examples are the contract.

4.3 The Mailchimp Reference Example

Mailchimp's voice is plainspoken, genuine, translator. Each attribute has concrete examples in the public guide (Mailchimp Content Style Guide, "Voice and Tone," 2026, sampled; https://styleguide.mailchimp.com/voice-and-tone/).

4.4 The Slack Reference Example

Slack's voice is "clear, concise, human" with sub attributes "confident (never cocky), witty (but never silly), conversational (but always appropriate and respectful), intelligent, friendly (but not ingratiating), helpful (never overbearing)" (Slack Design, "The voice of the brand: 5 principles," 2026; https://slack.design/articles/thevoiceofthebrand-5principles/; Slack Brand Guidelines PDF September 2020, sampled).

The "X but never Y" qualifier pattern is itself a voice documentation technique. It defines voice by what it is and by what would be a violation. A writer can ask "is this cocky?" and decline to ship.

4.5 The Atlassian Reference Example

Atlassian's voice is "Bold, optimistic, and practical (with a wink)" (Atlassian Design System, "Voice and tone," 2026; https://atlassian.design/content/voice-and-tone-principles/). Atlassian built Voicify, a tool with sliding scales for calibrating each attribute per surface. Voicify is the operational realization of the four dimensions framework.

4.6 Suggested Attributes Per Business Type

Attributes should emerge from brand personality, not a template. Observed patterns:

Patterns, not prescriptions. Actual attributes emerge from interview, content audit, and competitive differentiation, not from a category template.

5. Tone Variation By Content Type

Voice is the constant. Tone is the variable. This section specifies how the documented voice flexes across content surfaces without ever drifting out of voice.

5.1 The Tone Matrix

5.2 The Mailchimp Tone Matrix Example

Mailchimp's public guide has explicit per situation tone guidance (Mailchimp Content Style Guide, "Voice and Tone," sampled; https://styleguide.mailchimp.com/voice-and-tone/):

"Mailchimp's tone is usually informal, but it's always more important to be clear than entertaining. When you're writing, consider the reader's emotional state. Are they confused? Just sign up and getting started? They want to be reassured. Are they trying to do something? They want clear directions. Are they angry about a problem? They want acknowledgement."

"Consider the reader's emotional state" is the operational instruction. The tone matrix translates emotional states to tonal moves, each a four dimensions calibration for that surface and reader state.

5.3 The Three Surface Tone Test

For every surface in scope, document: What is the reader's emotional state on arrival? (curious, anxious, frustrated, celebratory, transactional); What does the reader most need from this surface? (information, reassurance, action, acknowledgement, escape); How does the brand voice show up to meet that? (which attributes lean forward, which lean back). Output: one paragraph tone instruction per surface in the style guide. Writers reference before drafting.

5.4 Cross Reference

framework-cro.md for voice in conversion flows; framework-formoptimization.md for email and notification patterns; framework-accessibility.md and framework-ymyl.md for accessible language in legal and policy pages.

6. Brand Pillars And Persona

Voice attributes flow from the brand's pillars (strategic values), archetype (symbolic category), and ICP (the audience the voice is calibrated for).

6.1 Brand Pillars

Brand pillars are the three to five things the brand fundamentally stands for. Not voice attributes themselves; strategic positioning from which voice attributes derive.

For each pillar (three to five), document name (e.g. "Mastery"), one_sentence_meaning (e.g. "We do hard things well, not easy things passably"), and voice_implication (e.g. "Voice is precise and specific. Hedging contradicts mastery"). The voice_implication field is the bridge from abstract value to concrete voice instruction.

6.2 Brand Archetype

The brand archetype is the symbolic personality category, drawn from Jung and adapted for marketing by Pearson and Mark in "The Hero and the Outlaw" (2001). Twelve archetypes recur (Toast Studio, "Brand archetypes in the age of AI," 2025 update, current 2026; https://www.toaststudio.com/en/articles/brand-archetypes-a-new-way-to-approach-tone-and-manner-in-content-marketing/; Pearson and Mark, 2001).

Commit to one primary archetype and at most one secondary (Apple is primary Creator with secondary Magician). Three or more signals confusion, not nuance.

6.3 ICP Voice Alignment

The voice must speak to the ICP, not at them:

icp_voice_alignment:

  primary_icp_role: ""                  # e.g. "Director of marketing at 50 to 500 person B2B SaaS"
  primary_icp_seniority: ""             # "IC" | "manager" | "director" | "VP" | "C_level" | "founder"
  primary_icp_expertise_level: ""       # "novice" | "intermediate" | "expert"
  primary_icp_jargon_tolerance: ""      # "low" | "medium" | "high"
  primary_icp_skepticism_level: ""      # "low" | "medium" | "high"
  primary_icp_time_pressure: ""         # "low" | "medium" | "high"

  voice_calibration_per_icp:            # Position on each of the four dimensions
    formality: ""
    humor: ""
    respectfulness: ""
    enthusiasm: ""

  language_alignment:
    use_their_terminology: []           # Words they use; use them
    avoid_their_jargon_landmines: []    # Words they hate; avoid them
    preferred_evidence_format: ""       # "case_studies" | "benchmarks" | "first_principles" | "testimonials"

A brand whose ICP is C level executives calibrates differently from one whose ICP is small business owners in skilled trades, even with the same archetype. The voice is the same; the calibration on each of the four dimensions shifts.

6.4 The Voice Alignment Test

Voice is correctly calibrated when a target ICP member reads three paragraphs and feels "this was written for me," and a non target reader feels "this is not for me" without feeling insulted. If both feel "fine but generic," undercalibrated. If non targets feel insulted, overcalibrated. First failure costs differentiation; second costs reach.

7. Style Guide Structure

The complete guide is the operational artifact. The voice attributes worksheet is one section of it.

7.1 The Anatomy

1. About this guide: owner, last review date, update process, authority to deviate.
2. Voice: attributes worksheet (Section 4); Voice IS and Voice is NOT in three adjectives; compared to person/celebrity/admired competitor/avoided competitor; four dimensions calibration.
3. Tone: variation matrix per surface (Section 5); emotional state to tone mapping with examples.
4. Grammar and mechanics: active voice preference; sentence and paragraph length; reading level target; punctuation (Oxford comma, em/en dash policy, ellipses, exclamations); capitalization (sentence vs title case, brand name); number/date formatting, quotation marks, hyphenation, spelling conventions.
5. Terminology: brand name, product/service names, industry term preferences; terms to avoid; inclusive language; acronyms policy.
6. Formatting: headings, lists, tables, block quotes, link anchor text, image alt text.
7. Accessibility patterns: plain language commitment per Federal Plain Language Guidelines; reading level targets (Grade 6 to 9 consumer typical, higher specialist); idioms, metaphors, pronoun and identity language conventions.
8. Content types: per type templates (blog opener, service page, FAQ pattern, case study, email).
9. AI assisted content policy: when AI may be used; review process; voice profile system prompts; disclosure.
10. Examples library: annotated good and bad examples per attribute; before/after rewrites; tone matrix samples.
11. Appendix: quick reference one pager; editorial checklist; author bio template; reviewer rubric.

7.2 Plain Language Commitment

Section 7 of the guide commits to plain language explicitly. The Federal Plain Language Guidelines (Plain Language Action and Information Network, March 2011, Revision 1 May 2011, current 2026; https://www.plainlanguage.gov/guidelines/) specify: write for the audience, organize the information, choose words carefully, write short sections and sentences, design for readability, test for understanding.

Plain language is not dumbed down. It is precise language calibrated to the reader. A surgeon writing for surgeons uses plain language dense with surgical terminology; a surgeon writing for patients translates that terminology. Both are plain language.

For SEO this matters because reading level and sentence complexity correlate with extraction. AI engines extract "clean assertions more reliably than long compound sentences with embedded clauses, hedges, and qualifiers" (Stridec, "How to Structure Content for AI Citations," 2026; https://www.stridec.com/blog/how-to-structure-content-for-ai-citations/). Plain language is both accessibility commitment and citation strategy.

7.3 The Atlassian, Slack, Mailchimp Reference Trio

Three publicly accessible style guides anchor the 2026 industry baseline:

• Mailchimp Content Style Guide (https://styleguide.mailchimp.com/): most cited UX writing reference. Tone, grammar, mechanics, terminology, formatting, accessibility, content types in dedicated chapters.
• Slack Brand Guidelines and Slack Design (https://slack.design/articles/thevoiceofthebrand-5principles/): voice principles forward over mechanics. The "X but never Y" pattern is borrowed from Slack.
• Atlassian Design System (https://atlassian.design/content/voice-and-tone-principles/): voice and tone integrated with typography, color, component guidelines.

Smaller brands need Mailchimp's structure; brands with strong design systems need Atlassian's integration; brands with strong personality need Slack's principle forward approach.

7.4 Style Guide Hosting

Best to worst: Best, public sub site or sub directory on brand's own domain (mirrors the thatdeveloperguy.com/style-guide/ pattern; indexable, linkable, demonstrates editorial maturity to AI engines). Good, versioned markdown repository (GitHub, GitLab, internal git; versioned, diffed, tied to workflow). Acceptable, wiki (Notion, Confluence; findable but less version controlled). Worst, PDF or Google Doc shared by link (stale within months, no version control, hard to embed in workflows).

For SEO, a public indexable style guide is itself a Trust signal demonstrating editorial discipline. AI engines that index the site can detect alignment between the style guide and the published content.

8. Voice Consistency At Scale

A single author can produce consistent voice without a guide. Two or more authors cannot. AI assisted content cannot. The guide is necessary but not sufficient; the editorial workflow converts the guide into consistent output.

8.1 The Multi Author Problem

When a brand publishes from multiple writers, drift is inevitable without process. Sources of drift: background voice (each writer's default leaks in), interpretive variance (two writers read "warm" differently), surface specialization (email writer, blog writer, social writer drift over time), turnover (new writer inherits prior voice rather than documented brand voice).

8.2 Editorial Workflow Solutions

Onboarding new writer: read guide end to end; read 10 representative pieces with brand owner or senior editor; write a calibration piece edited line by line; sign editorial commitment (one page); first five published pieces get line level senior editor review; quarterly calibration for first year.

Per piece workflow: pre draft brief includes attribute calibration, tone for surface, audience emotional state, surface examples; during drafting writer references guide with reminders pinned; editorial review is three passes (accuracy/structure, voice alignment, grammar/mechanics); publication checklist includes voice marker, tone marker, style marker checks.

Quarterly governance: sample 5 to 10 recent pieces; score attribute by attribute; identify drift patterns by author or surface; update guide if drift reveals a gap; recalibration session with writers.

8.3 The AI Assisted Content Problem

AI assisted content is a different governance problem. Three risks:

Default LLM voice: every major LLM has a default voice that leaks through. ChatGPT defaults verbose, hedged, transitional ("In today's fast paced world..."), structurally repetitive ("not only X but also Y"). Claude defaults measured, slightly formal, occasionally cautious to the point of hedging. Detectable by humans and increasingly by AI engines trained on counter examples (Originality.AI, "Most Commonly Used ChatGPT Phrases," 2025, https://originality.ai/blog/can-humans-detect-chatgpt; DeGPT, "ChatGPT Tells List," 2025, https://www.degpt.app/blog/chatgpt-tells-phrases-list).

Boilerplate phrases: AI tools recur to specific patterns. "It's important to note," "Furthermore," "Moreover," "In conclusion," "delve," "tapestry," "realm," "navigate the landscape." Recognized as AI signatures regardless of fact accuracy.

Brand voice collapse: without a configured profile, AI assisted content collapses to LLM default. Brand attributes become invisible. Every piece sounds like every other AI piece.

8.4 AI Voice Profile Configuration

The fix is configured AI voice profiles or carefully crafted system prompts that operationally encode voice attributes:

Attributes section: voice attributes (three to five) with definitions and examples; four dimensions calibration; "never sound like" negative examples.
Audience section: primary ICP, expertise level, emotional state at first touch, jargon tolerance.
Surface calibration: specific surface; tone instruction from Section 5; word count and structural target.
Forbidden phrase list: LLM tells to block ("delve", "tapestry", "realm", "in today's fast paced"); hedge phrases to avoid in declarative contexts ("may", "might", "could potentially").
Example calibration samples: three to five paragraphs of genuine brand content as pattern reference; annotated examples explaining alignment.
Review instructions: factual claims to verify; competitor comparisons to flag; statistics and dates to check.

A configured voice profile closes the 41 point CMI gap (64% documented, 23% in AI tools; cited Section 4.2).

8.5 AI Content Review Discipline

Configured profile is not enough. Every AI assisted piece must be reviewed by a human with authority to rewrite:

• Voice check: sounds like the brand? Forbidden phrases absent? Four dimensions calibration on target?
• Accuracy check: factual claims verified? Statistics dated and sourced? No hallucinated names, dates, or studies?
• Voice attribute check: each documented attribute audibly present in the draft.
• Tone check: tone calibrated to this surface and audience emotional state.
• AI tell scan: no "It's important to note", "Furthermore", "Moreover", "In conclusion"; no "delve", "tapestry", "realm", "navigate the landscape"; no "In today's fast paced [adjective] world"; no "not only X but also Y" at every other paragraph; no three or four item list at every section break.
• E-E-A-T check: real first hand experience present? Bylined credentialed author? AI use disclosed (framework-hcs.md Section 6.5)?

Pieces that fail the AI tell scan but are otherwise solid get rewritten paragraph by paragraph. Even small density of LLM tells collapses voice into default.

8.6 Tooling

• Human authors: Grammarly Business (voice rules at scale), Acrolinx (enterprise voice and style), Writer.com (voice profiles, terminology).
• AI assisted: Writer.com brand voice profiles, Jasper Brand Voice, custom Claude/GPT system prompts with voice attributes, custom tool with full style guide in context window.
• Audit: Originality.AI (AI detection), GPTZero (AI detection), Hemingway Editor (readability), Flesch reading ease as plain language proxy.

9. AI Engine Voice Detection

The 2025/2026 reality: AI engines read voice signals into citation decisions. Documented in citation pattern studies; reproducible in side by side tests of same facts, different voice.

9.1 What AI Engines Detect

AI engines do not have a "voice attribute extractor" module. They have language models that compute embeddings of pages and queries. Voice signals surface in those embeddings as patterns: formality, hedge density, perspective markers, sentence length variance, vocabulary specificity, claim confidence, evidence backing.

9.2 Engine Specific Voice Preferences

Engines have different aggregate voice preferences reflecting training data and synthesis behavior (5W Public Relations, "AI Platform Citation Source Index 2026," consolidating six citation studies Aug 2024 to Apr 2026 covering ChatGPT, Claude, Perplexity, Gemini, Google AIO; https://www.prnewswire.com/news-releases/5w-releases-ai-platform-citation-source-index-2026-the-50-websites-that-now-decide-what-brands-are-visible-inside-chatgpt-claude-perplexity-gemini-and-google-ai-overviews-302759804.html).

9.3 Why This Matters Strategically

The voice that earns AI citation is not the voice that earns highest social engagement. Social rewards humor, brevity, emotional language; AI engines reward confidence, specificity, evidence backing, first person credentialed perspective. A voice tuned only for social wins shares and loses citations; a voice tuned only for AI wins citations and feels robotic on social.

Sustainable position: confident, specific, evidence backed core voice; tonal flex on social to lean into humor and intimacy without losing the declarative spine; tonal restraint in YMYL and pricing surfaces where hedge would damage Trust. This is the operational version of voice constant / tone variable.

9.4 The Voice Citation Audit

To verify a brand's voice wins AI citation:

1. Baseline citation state: pull citation data from ChatGPT, Perplexity, Claude, AIO for branded and category queries. Tools: Otterly, Profound, AthenaHQ, manual prompt testing. Note which engines cite, which queries, what URLs.
2. Compare cited vs uncited: cited pages have which voice signals? Uncited absent which? Pattern: cited typically higher declarative, lower hedge, clearer perspective markers.
3. Revise and measure: rewrite 5 to 10 underperformers to align with documented voice attributes. Re measure citation 30 to 90 days later.
4. Document: which voice moves measurably improved citation; update examples library.

10. Voice And E-E-A-T

E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) is Google's quality framework (framework-eeat.md). Voice overlaps E-E-A-T at multiple points, most strongly at Experience and Trust.

10.1 Voice Signals Expertise

Expertise is signaled by precise domain vocabulary, evidence backed claims, acknowledgement of complexity, willingness to disagree with surface consensus. All voice phenomena. A page that hedges every claim ("some experts believe," "it may be the case that") signals low expertise even if the author has high expertise. The voice attribute that signals expertise is some flavor of specific, confident, or precise. The behavior is declarative claim backed by specific evidence.

10.2 Voice And First Hand Experience

The Experience pillar was added in 2022 to reward content from people who actually did the thing. First hand experience is a voice phenomenon: writer's perspective, decisions, observations audible in the text. Web Castle Boston reported in 2025 that "Google is wiring its algorithms to trust the authentic human perspective... search engines wanting the seasoned voice, not the copy paste version" (Web Castle, "E-E-A-T Optimization in 2025," https://webcastle.com/blog/e-e-a-t-optimization-mastering-googles-trust-signals-in-2025/). The "seasoned voice" is the practitioner's signature: specific decisions, named tradeoffs, anecdotes with dates and outcomes.

A brand voice that admits first person, names the author, and includes specific experience scores higher on Experience than one that hides behind "the team" without provenance.

10.3 Voice As A Trust Signal

Trustworthiness is the most important E-E-A-T pillar; Google states this directly in the SQRG (Search Quality Rater Guidelines, current edition, sampled Trust chapter; https://services.google.com/fh/files/misc/hsw-sqrg.pdf; framework-sqrg.md).

Voice contributes to Trust through: honesty markers (acknowledging limitations, admitting tradeoffs); specificity (numbers, dates, named clients with permission); provenance (named author, credentials, contact); restraint (not overclaiming, no inflated promise). The opposite voice patterns damage Trust: marketing voice that overclaims, hedge dense voice signaling avoidance, generic voice without provenance, voice that pretends to certainty where uncertainty is the truth.

10.4 The Voice And E-E-A-T Audit Overlap

Several E-E-A-T audit criteria are voice criteria in disguise. Examples (cross referenced from framework-eeat.md):

The implication: improving voice improves E-E-A-T scoring even when no other content changes are made. A site whose voice is documented, calibrated, and consistently rendered will outscore a site with identical facts and structure but voice drift.

11. Voice And AI Overview Citation

AIO citation is the most extractable surface where voice meets SEO outcome. Specific phrasing patterns earn citation; others do not.

11.1 Declarative Wins. Hedged Loses.

Most replicated finding across 2025/2026 citation studies: declarative sentences earn citation; hedged sentences do not.

Stellar AEO Labs found AI engines "prioritize declarative statements" and that "hedging language ('may,' 'might,' 'could') signals uncertainty, and AI platforms interpret modal verbs as low certainty claims and downgrade confidence scores" (Stellar AEO Labs, "How AI Engines Decide What to Cite," 2026, citation pattern analysis across ChatGPT, Perplexity, Claude, AIO; https://stellar-ai.co/blog/how-ai-search-engines-decide-what-to-cite/).

Stellar's example pair:

Hedged (uncited): "Many dermatologists suggest that incorporating certain ingredients into your routine could potentially offer benefits for various skin concerns."

Declarative (cited): "Retinol increases cell turnover and stimulates collagen production."

The hedged version has no extractable fact. The declarative version is "immediately usable." Hedging is a citation killer. Voice attributes like "confident" or "specific" must operationalize as declarative phrasing.

11.2 Specific Wins. Generic Loses.

Specificity is the second strongest voice signal. Specific numbers, dates, named entities, named places, named processes are extracted as quotable claims. Generic adjectives ("many," "various," "several," "robust," "innovative") extract as nothing.

The generic versions are uncited because there is nothing to cite. The specific versions are citation magnets because every sentence contains an extractable claim.

11.3 First Person Wins For Experience Queries

For experience driven queries, first person voice with credentials earns citation; third person without provenance does not. Google's E-E-A-T 2025 guidance rewards "authentic voice and storytelling using 'I' or 'we' to convey personal expertise and involvement" (framework-eeat.md). Voice attributes that support this: candid, observational, opinionated, willing to name decisions and outcomes.

11.4 Evidence Backed Wins

Inline citations to authoritative sources lift citation probability by ~30% (Princeton GEO study, SIGKDD 2024, n=10,000 AI citations across major engines; framework-aicitations.md, framework-contentfirst.md). Voice attributes like "evidence backed" or "well sourced" operationalize as inline citations with named sources.

11.5 Restrained Wins Over Marketing Voice

Marketing voice (superlatives, exclamations, embedded CTAs) is cited at lower rates than restrained editorial voice. AI engines train on quality journalism, encyclopedia content, academic writing; none use marketing voice. Restrained editorial voice still allows brand voice attributes to be expressed. The restraint is in tone, not voice. The tone matrix allows both modes.

11.6 The Voice For AIO Citation Checklist

11.7 Cross Reference

Complete AIO spec: framework-aioverviews.md. Broader AI citation mechanics across ChatGPT/Claude/Perplexity/Gemini: framework-aicitations.md. ChatGPT Search specific: framework-searchgpt.md. This framework supplies the voice layer the others apply.

12. Common Brand Voice Mistakes

The top ten anti patterns observed across audits in 2024, 2025, and 2026.

12.1 Tag Soup Voice Documentation

"Our voice is friendly, professional, innovative, dynamic, and approachable." Five generic adjectives that describe almost any brand. No do/don't examples. No four dimensions calibration. AI tools cannot operationalize this; human writers interpret it differently; output is generic. Fix: three to five attributes drawn from the brand's specific personality, each with definition and three to six concrete do/don't examples. See Section 4.

12.2 Aspirational Voice The Brand Cannot Sustain

A local service business documents "sophisticated, witty, urbane" because the founder admires The New Yorker. Writers without that background revert to plain voice; output becomes inconsistent. The aspirational voice may not fit the audience anyway. A neighborhood plumber's customers want "shows up on time, fixes the thing, fair price." Fix: attributes drawn from what the brand actually is, calibrated to audience need. See Section 6.3 ICP alignment.

12.3 Voice Documented But Not Trained Into AI Tools

Documented style guide; AI tools running default voice. CMI 2025 reports 64% have documented voice but only 23% use it in AI tools (Section 4.2). AI assisted content collapses to LLM default voice with detectable tells. Fix: configure AI voice profiles using style guide attributes (Section 8.4). Audit AI output for tells (Section 8.5).

12.4 Voice Drift Across Authors

Three writers produce content under one brand. Over months, three distinct voice clusters emerge. Brand recognition erodes; AI engines see inconsistent signal; readers lose engagement on author handoff. Fix: editorial workflow with line level review of new writer's first five pieces; quarterly calibration; pinned voice reminders in drafting tools. See Section 8.2.

12.5 Voice That Hedges In Buying Moments

Brand voice is "confident" everywhere except the pricing page, where it becomes "Our pricing is flexible and tailored to your needs. Contact us for a custom quote." Hedge at the buying moment is a Trust violation. Buyers want specific numbers; AI engines extract zero. Fix: tone matrix (Section 5) specifies confidence at pricing. Specific numbers, named tiers, explicit inclusions and exclusions.

12.6 Voice Performed Through Stunt Phrasing

Voice attribute is "playful." Every other sentence contains a pun or forced joke. Voice becomes performance, not personality. Trust degrades because the voice signals attention seeking. AI engines deprioritize stunt density. Fix: voice attributes are calibrations of the four dimensions, not license for stunts. Atlassian's "with a wink" framing is the model: restraint, not performance.

12.7 Voice That Ignores Audience Emotional State

Brand voice is "enthusiastic and energetic." 404 page reads "Whoopsie daisy! Looks like you took a wrong turn at Albuquerque!" The audience is frustrated. The brand performs enthusiasm at a moment that requires calm helpfulness. Fix: tone matrix calibrates expression per surface and emotional state. Error states get minimal voice and high helpfulness. See Section 5.1.

12.8 Voice With No First Person

Site filled with "the team," "our experts," "many of our clients have found." No named author; no first person perspective. E-E-A-T Experience pillar gets no signal. AI engines that reward first person credentialed voice find none. Fix: author bylines on substantive content. First person where appropriate (about, blog, founder commentary). See framework-eeat.md Section 7 on author infrastructure.

12.9 Marketing Voice At All Surfaces

Every page reads like a sales page. "Leverage our cutting edge solution." "Unlock unprecedented growth." AI engines do not cite marketing voice; readers tune out; E-E-A-T penalizes overclaim; conversions drop because persistent sales voice signals desperation. Fix: restrained editorial voice for body content; marketing voice confined to clearly marked CTA blocks and ad copy. See Section 5.

12.10 Voice That Never Disagrees

Brand agrees with every prevailing view in the industry. Surface level consensus is reinforced; no contrarian perspective surfaces. Information Gain (framework-infogain.md) requires novel contribution; voice that never disagrees produces aggregator content; citation rates collapse. Fix: brand should have a documented position on at least three contested issues in its category. Voice attributes should include some flavor of "opinionated" or "willing to disagree." See Section 6.1 and Section 11.

13. Audit Rubric

Three scopes: per content surface, site wide, first 90 days post implementation.

13.1 Per Surface Audit

For each major surface (homepage, service pages, blog, FAQ, about, pricing, case studies, email, social), sample 3 to 5 pieces and score:

Score per surface: 10. World class per surface: 9+/10.

13.2 Site Wide Audit

Site score: 15. World class voice site: 14+/15.

13.3 First 90 Days Audit

Run 90 days after voice and guide installation, when enough content has been produced under the new voice that drift patterns are detectable:

90 day score: 10. World class 90 day score: 9+/10.

13.4 Tier Bands

Total possible: 35 (per surface 10 + site wide 15 + 90 days 10).

• World class (32 to 35): fully operational. Style guide current. All authors aligned. AI tools configured. Citation outcomes improving.
• Strong (27 to 31): voice documented and applied. Minor drift or gaps. Most surfaces aligned.
• Inconsistent (21 to 26): partial documentation. Drift across authors and surfaces. AI tools not configured.
• Voice chaos (below 21): no coherent voice. Multiple voices across surfaces. Attributes undocumented or unused.

14. Maintenance Schedule And Report Templates

14.1 Maintenance Schedule

• Weekly: editorial review of published pieces against voice attributes; AI assisted content reviewed with forbidden phrase scan.
• Monthly: sample 5 to 10 recent pieces across surfaces; score against rubric; flag drift patterns; brief writers on calibration.
• Quarterly: site wide voice audit (Section 13.2); AI voice profile refresh (update examples library); governance session with writers; update guide from operational learning; citation tracking review correlated with voice changes.
• Annually: comprehensive review with brand owner; voice attributes, ICP alignment, archetype still accurate?; refresh examples library; audit competitive voice landscape; reassess differentiation.
• Post algorithm / major engine update: re run citation audit after Google core updates and AIO algorithm changes; correlate voice changes with citation changes; update playbook if engine behavior shifts.

14.2 Voice Implementation Report Template

# Brand Voice Framework Implementation Report

**Project**: {{BUSINESS_NAME}}
**Implementation Date**: {{TODAY}}

## Summary
Attributes documented, tone matrix surfaces covered, style guide sections complete, authors onboarded, AI tools configured.

## Voice Attributes Finalized
{{THE_THREE_TO_FIVE_ATTRIBUTES_WITH_DEFINITIONS}}

## Brand Archetype, ICP Calibration, Style Guide
Primary {{ARCHETYPE}}, Secondary (optional) {{ARCHETYPE}}; ICP {{ROLE}}; four dimensions positions; style guide at {{URL_OR_REPO_PATH}} ({{HOSTING_TIER}}).

## AI Voice Profile Configuration
{{TOOLS_AND_PROFILE_STATUS}}

## Editorial Workflow Installed
{{WORKFLOW_STAGES_AND_REVIEW_GATES}}

## Items Flagged For Manual Review
{{LIST}}

## Sign Off
Implementation complete {{DATE}}.

14.3 Voice Audit Report Template

# Brand Voice Audit Report

**Site**: {{BUSINESS_NAME}}
**Audit Date**: {{TODAY}}

## Executive Summary
{{ONE_PARAGRAPH_ASSESSMENT}}
**Overall Score**: {{X}}/35
**Tier**: {{WORLD_CLASS | STRONG | INCONSISTENT | VOICE_CHAOS}}

## Voice Documentation State
- Style guide exists, last reviewed, hosting tier, attributes count, tone matrix presence

## Per Surface Findings
Homepage, service pages, blog, FAQ, about, pricing, case studies, email. Score X/10 with findings each.

## Voice Drift Analysis
{{DRIFT_PATTERNS_BY_AUTHOR_AND_BY_SURFACE}}

## AI Assisted Content Review
{{AI_PROFILE_STATUS_AND_AI_TELL_DENSITY}}

## Voice And E-E-A-T Overlap Findings
{{SCORING_OF_VOICE_DRIVEN_TRUST_SIGNALS}}

## Voice And AI Citation Findings
{{CITATION_TRACKING_DATA_CORRELATED_WITH_VOICE_PATTERNS}}

## Critical Voice Failures
{{LIST_WITH_REMEDIATION}}

## Recommended Remediation Order
1. Critical: {{PRIORITY_1}}
2. High: {{PRIORITY_2}}
3. Medium: {{PRIORITY_3}}

## Estimated Remediation Effort
{{HOURS}}

## Re Audit Recommendation
{{DATE_90_DAYS_OUT}}

## Sign Off

14.4 Monthly Voice Maintenance Report Template

# Voice Maintenance Report, {{MONTH}} {{YEAR}}

**Site**: {{BUSINESS_NAME}}
**Period**: {{START_DATE}} to {{END_DATE}}

## Activity Summary
Pieces published, pieces audited, average per surface score, AI assisted count, AI tell scan pass rate.

## Drift Patterns Detected
{{LIST}}

## Calibration Actions Taken
{{LIST}}

## Citation Tracking
ChatGPT, Perplexity, Claude, AIO citation deltas vs prior month.

## Style Guide Updates
{{LIST_OF_UPDATES}}

## Next Month Priorities
{{LIST}}

15. End Of Framework Document

Document version: 1.0
Created: 2026-05-14
Maintained by: ThatDeveloperGuy

Voice is the constant. Tone is the variable. Style is the rules. A documented voice that survives multi author, AI assisted, multi surface publication earns compounding citation and Trust returns from AI engines and Google's E-E-A-T scoring. An undocumented voice, or a documented voice not trained into AI tools, produces drift that erodes brand recognition and citation rates simultaneously. The voice playbook is operational SEO infrastructure, not only a marketing artifact.

Companion documents:

• framework-contentfirst.md, framework-hcs.md, framework-infogain.md, framework-eeat.md, framework-sqrg.md, framework-ymyl.md: voice rides on the content first substrate; it is the audible signal of HCS people first content; it makes Information Gain audible; it signals Expertise and Trust in E-E-A-T; it meets SQRG trust evaluation; restraint and accuracy heighten in YMYL.
• framework-aicitations.md, framework-aioverviews.md, framework-searchgpt.md: AI citation theory across engines; Google AIO citation patterns; ChatGPT Search behavior.
• framework-perplexityspaces.md, framework-ai-content-workflow.md, framework-contentbriefs.md: Phase 2 siblings scheduled (Perplexity Spaces, AI content production workflow, content brief templates that carry voice instructions).
• framework-contentaudit.md, framework-contentrefresh.md: voice is a refresh trigger; refresh realignment patterns.
• framework-schema.md, framework-entitysalience.md, framework-accessibility.md: bylines and Person schema support voice signals; named authors as entities; plain language commitment overlaps voice.
• framework-cross-stack-implementation.md, framework-react.md, framework-tailwind.md: per stack patterns for byline blocks, voice metadata, editorial markers; voice infrastructure in SPA contexts; typography patterns.
• SEO-Search-Appearance.md, SERP-Optimization.md: how voice shows up in SERP appearance; SERP level voice optimization.

This document is the canonical voice and style guide framework. Apply it before content production begins. Audit against it during ongoing content operations. Refresh annually with the brand owner.

Owner: Joseph W. Anady, ThatDeveloperGuy, SDVOSB
Contact: 505-512-3662 | joseph.w.anady@icloud.com

From the ThatDevPro Engine Optimization framework library. Studio: ThatDevPro (SDVOSB veteran-owned web + AI engineering). Sister property: ThatDeveloperGuy. Source: https://www.thatdevpro.com/insights/framework-brandvoice/.