Create a consistent brand voice with AI using ready-made templates, guardrails, banned phrases, and tests. Includes copy-paste prompts, examples, and JSON-LD.
Summary
This guide helps teams define and enforce brand voice in AI-generated content. You’ll get practical voice pillars, a do/don’t table, banned phrases, copy-ready prompts, and a test workflow.
Who this is for
- Content, brand, and marketing teams
- Product/docs writers, community and support leads
- Anyone standardizing AI outputs across channels
What This Prompt Does
- Distills your existing content into clear voice pillars
- Enforces tone with guardrails and banned phrases
- Adds automated compliance checks before publish
- Provides tests so editors can approve quickly
Brand Voice Pillars (Define First, Then Automate)
Pick 3–5 pillars. Example set below—replace with your own.
- Clear, not clever
- Short sentences, everyday words, concrete nouns
- Helpful, not pushy
- Shows steps, avoids hype or pressure
- Confident, not arrogant
- Evidence over adjectives; cites facts, not fluff
- Warm, not casual
- Friendly and respectful; avoids slang/jargon unless necessary
Do/Don’t Table
| Pillar | Do | Don’t |
|---|---|---|
| Clear | Use short sentences; define terms | Use jargon; stack clauses |
| Helpful | Show steps and reasons | Hard-sell; make vague claims |
| Confident | Cite data, link docs | Use hype or unverifiable superlatives |
| Warm | Use inclusive, respectful language | Use sarcasm, slang, or inside jokes |
Banned Phrases (edit to your needs)
- “Game-changing,” “revolutionary,” “world-class,” “disruptive,” “unlock unlimited,” “effortless,” “#1,” “best-in-class,” “blazing fast,” “magic,” “instant results,” “guaranteed,” “no brainer,” “crush your competition,” “killer feature”
Step-by-Step Workflow
- Collect samples
- Gather 10–30 on-brand pieces (site, docs, emails). Add 5–10 off-brand pieces (for contrast).
- Extract patterns
- Identify tone, sentence length, vocabulary, structure, formatting, CTA styles.
- Write rules
- Codify pillars, do/don’t, banned phrases, formatting norms (headings, lists, links).
- Test prompts
- Run the prompts below on varied inputs (long/short, technical/marketing). Compare to rules.
- Review
- Human editor checks tone, claims, links, accessibility, inclusivity.
- Publish
- Ship the guide; version it; embed in your CMS; train contributors; schedule quarterly reviews.
Copy-Paste Prompt Patterns (Templates)
1) Style Distillation (build your voice from samples)
You are a brand voice analyst. Analyze the PROVIDED SAMPLES and produce a concise style guide:
1) Voice pillars (3–5): each with 1–2 sentences.
2) Sentence & structure: average length, common patterns, bullets/steps usage.
3) Vocabulary: preferred words/phrases; words to avoid.
4) Tone rules: do/don’t table (3–5 rows).
5) CTA patterns: examples that match the voice.
Constraints:
- Be specific and measurable where possible (e.g., sentence length ranges).
- Avoid generic advice; use observations from the samples.
PROVIDED SAMPLES:
[Paste 10–30 short samples here. Include 3–5 off-brand as negative examples.]
2) Tone Guardrails (apply pillars, do/don’t, banned phrases)
Act as a brand voice enforcer.
INPUT: [Paste draft content]
VOICE PILLARS:
- [Pillar 1]
- [Pillar 2]
- [Pillar 3]
DO/DON’T:
[Paste table as plain text]
BANNED PHRASES:
[Paste list]
TASKS:
1) Rewrite the draft to comply fully.
2) List exact changes (before → after) with 1-sentence rationale each.
3) Confirm no banned phrases remain.
4) Return final copy only at the end.
3) Compliance Filter (pre-publish check)
You are a compliance checker.
INPUT: [Paste final draft]
CHECKS:
- Claims are specific and sourced or removed.
- No banned phrases appear.
- Tone aligns with voice pillars and do/don’t.
- Accessibility: active voice, descriptive links, no overly long sentences (target ≤22 words).
- Inclusivity: people-first language, avoid stereotypes.
OUTPUT:
- PASS/FAIL with bullet list of issues.
- If FAIL, return a compliant revision.
4) “Rewrite Without Hype” (de-fluff marketing copy)
Rewrite the following to be factual, neutral, and helpful. Remove hype and unverifiable superlatives. Preserve meaning, add concrete specifics where possible, and prefer short sentences.
TEXT:
[Paste copy]
Examples
A) Transform Dry → Friendly (still professional)
Before (dry):
“Our platform facilitates cross-functional collaboration through integrated modules.”
After (friendly):
“Work together in one place. Your teams can plan, share files, and track progress without switching tools.”
B) Transform Friendly → Formal (for policy/docs)
Before (friendly):
“Got a question? We’ll help you get set up in minutes.”
After (formal):
“If you have questions, our support team provides step-by-step setup guidance with documented procedures.”
Model-Specific Settings (recommended defaults)
- GPT (OpenAI): temperature 0.2–0.4; top_p 1.0; enable system message with pillars/do-don’t/banned phrases; use “JSON mode” for checks where available.
- Claude (Anthropic): temperature 0.2–0.4; use “constitutional” or “safety” guidance to enforce guardrails; provide examples.
- Gemini (Google): temperature 0.25–0.35; enable safety settings; provide explicit negative prompts with banned phrases.
- General: keep prompts short; pin system instruction with pillars and rules; store the do/don’t table and banned list in your prompt library.
Variations (use when content type changes)
- Docs mode: “Prefer numbered steps, code blocks, and exact commands. Avoid adjectives; include expected outputs.”
- Support mode: “Empathetic, concise answers with one actionable next step and a link to a canonical article.”
- Executive brief: “Two short paragraphs + 3 bullets; cite metrics; no marketing language.”
Guardrails & Banned Phrases (Quick Copy)
Guardrails:
- Avoid hype; show evidence.
- Keep sentences ≤22 words when possible.
- Use descriptive anchor text (no “click here”).
- Use people-first, inclusive language.
Banned: “revolutionary,” “game-changing,” “world-class,” “best-in-class,” “blazing fast,” “guaranteed,” “no brainer,” “unlock unlimited,” “killer,” “crush competitors,” “magic.”
Quality Checklist
- Voice pillars referenced and applied
- Do/Don’t table followed
- No banned phrases present
- Factual claims supported or removed
- Links descriptive; steps clear; inclusive language used
Troubleshooting
- Output still hyped? Run “Rewrite Without Hype,” then re-apply Tone Guardrails.
- Tone inconsistent across sections? Process each section separately, then merge and smooth.
- Editor disagreements? Capture decisions in the rules; update the table and banned list; re-train the model prompts.
Safety & Compliance
- Avoid unverifiable claims and rankings.
- Use inclusive, respectful language.
- Keep accessibility in mind (headings, lists, alt text, clear links).
- If legal/regulatory content: require human legal review before publish.
TL;DR
- Define pillars, rules, and banned phrases from your best samples.
- Enforce tone with guardrail prompts and a compliance filter.
- Test, review, and publish a versioned guide your AI can follow.
FAQ
Q: How many pillars should we use?
A: Three to five is ideal—clear enough to guide, simple enough to remember.
Q: Can we keep some personality without hype?
A: Yes—use concrete benefits, examples, and reader-focused phrasing instead of superlatives.
Q: How often should we update the guide?
A: Review quarterly or after major brand or product changes.
Q: Where should the guide live?
A: In your CMS/wiki with a shared shortcut. Add the rules to system prompts in your tools.
Internal Links
- AI Guide Library — Master List — https://aiupdates.news/category/guides/
External Sources
- Nielsen Norman Group: Voice & Tone
- Microsoft Writing Style Guide (free)
- GOV.UK Content Style Guide
- Google Inclusive Language Guide
