Most brand voice docs are platitude collections nobody references (“we’re authentic, bold, and human”). A useful voice doc has 3-4 anchor adjectives, a few tone axes (formal ↔ casual, expert ↔ approachable), and worked examples that show the same sentence written in-voice and not-in-voice. The 12 prompts below produce exactly that artifact, then turn it into a paste-ready style prompt for Claude or ChatGPT so AI drafts come back on-voice.
TL;DR: Extract anchor adjectives from your best existing copy (prompt 1), place dots on tone axes (prompt 2), build a paired-example library (prompt 4), compress to a one-page card (prompt 8), then convert the whole thing into an AI system prompt (prompt 11). Voice (your fixed identity) stays constant; tone shifts per channel.
Who this is for
Founders codifying voice before a hire, marketing leads handing off to contractors, content teams onboarding writers, and devrel leads aligning tutorials across authors.
When not to use these prompts
Skip them for trademark or legal style guides — those need counsel, not a model. And skip them when leadership won’t commit: a voice doc with no champion is dead on arrival.
The framework these prompts follow
Two ideas from the research keep voice docs usable:
- Voice is fixed, tone flexes. Your voice is the brand’s core identity and stays constant; tone adjusts for context (more formal in a press release, more conversational on social). Both should still read as the same company.
- Position on axes beats stacking adjectives. The widely cited Nielsen Norman Group tone-of-voice model maps voice on four scales — formal/casual, serious/funny, respectful/irreverent, matter-of-fact/enthusiastic. A dot on a spectrum is more actionable than a vague label.
The strongest one-page voice doc has five parts: anchor adjectives, words you use, words you never use, before/after examples, and axis positions. Every prompt below feeds one of those five.
Prompt anatomy / structure formula
Every voice prompt should carry six elements:
- Audience: one specific reader.
- Goal: one outcome — read / click / agree / share.
- Voice: 2-3 anchor adjectives, optional sample line.
- Constraints: word count, banned phrases, must-include facts.
- Format: paragraph, bulleted, headed, table.
- Examples: 1-2 tone samples — the single best lever for matching voice.
12 copy-ready prompt templates
Paste each into Claude (Opus 4.7 for the extraction and audit prompts, Sonnet 4.6 for fast iterations), ChatGPT (GPT-5.5), or Cursor. Swap the backtick placeholders for your own text.
1. Voice extraction from samples
Read these 5 samples of our best-performing content: [paste samples]. Extract: (1) 3-4 anchor adjectives (one positive, one constraint, one differentiator), (2) sentence-length pattern, (3) word patterns we lean on, (4) words we never use. Output as a draft voice doc.Swap: the 5 samples (use your highest-engagement pieces, not your newest).
2. Tone axes
Define 4 tone axes for `[brand]`: (a) formal <-> casual, (b) expert <-> approachable, (c) serious <-> playful, (d) bold <-> measured. Place a dot on each axis (0-10) with one-sentence rationale. Pin extremes that are OFF-LIMITS (e.g., never sloppy, never preachy).Swap: [brand]
3. Anchor adjective triangulation
I think our voice is `[adj1]`, `[adj2]`, `[adj3]`. Stress-test: (1) Name a real brand for each adjective — does it match the company we want to be? (2) Replace one adjective with a sharper synonym if vague. (3) Add a fourth that constrains (e.g., "never glib").Swap: [adj1], [adj2], [adj3]
4. Worked-example library
For each anchor adjective, produce 3 example pairs: same meaning, version A is in-voice, version B is not. Annotate why A is in-voice in one line. Use real product examples, not lorem ipsum.5. Do / Don’t list
Write 8 do / 8 don't items for our voice. Each <= 12 words. Don'ts must be specific (not "don't be unclear" but "don't hedge with maybe / probably"). Pair each do with an example.6. Per-channel voice map
Define how voice shifts for: (a) marketing copy, (b) docs, (c) error messages, (d) support replies, (e) social posts. Anchor adjectives stay; tone axes can shift. Map each context to a dot on the axis.7. Voice audit on shipped content
Audit these pieces against our voice doc: [paste pieces]. For each: (a) in-voice score 1-5, (b) one line on what's off, (c) one suggested rewrite. Skip pieces already at 5.Swap: the pieces to audit
8. Voice doc one-pager
Compress our voice doc to a 1-page reference card: 3 adjectives, 4 axes (with dot positions), 5 do, 5 don't, 2 worked examples. Goal: a freelancer can match voice on the first try.9. Translating voice across languages
Our brand voice in English is `[voiceSummary]`. Define how it should adapt in `[secondLang]`: which adjectives translate directly, which need a different anchor (e.g., "playful" in English does not equal "playful" in business Japanese), which words are forbidden.Swap: [voiceSummary], [secondLang]
10. Voice for crisis / serious comms
Define voice variants for: (1) routine, (2) celebration / launch, (3) apology, (4) outage / serious. Each: tone axis position + 2 sample phrases. Apology voice must drop the "playful" anchor if any.11. Voice training prompt for AI tools
Turn our voice doc into a system prompt I can paste into Claude or ChatGPT so AI output matches our voice. <= 400 words. Include: anchor adjectives, axis positions, 3 worked in-voice/not-in-voice examples, and banned phrases.Paste the result into ChatGPT’s Custom Instructions or a Claude Project’s custom instructions. Claude’s Styles feature (2026) can also build a tone profile directly from an uploaded writing sample, useful as a second opinion against your hand-written doc.
12. Voice drift detector
Read these 10 recent pieces. Detect drift: (1) Pieces straying off-axis, (2) Author writing in their personal voice vs the brand voice, (3) Newer pieces using banned words. Output a drift report + 1 fix per piece.Which model for which step
| Step | Best fit (June 2026) | Why |
|---|---|---|
| Extraction (1), audit (7, 12) | Claude Opus 4.7 | Strongest at synthesizing a batch of samples; 1M-token context holds dozens of pieces at once |
| Fast drafts of do/don’t, examples (4, 5) | Claude Sonnet 4.6 or GPT-5.5 | Cheap, fast, good enough for iterating |
| Pasting the final style prompt | ChatGPT Custom Instructions / Claude Project / Claude Styles | Where your team actually drafts daily |
Pricing as of June 2026: ChatGPT Plus $20/mo, Claude Pro $20/mo ($17 annual), both well within range for a small content team. Cursor Pro ($20/mo) runs the same models if you draft inside the editor.
Common mistakes
- Vague audience (“anyone who…”) — output reads generic.
- No tone anchor — every variant comes back the same flavour.
- No constraints — set word count, banned phrases, and a length cap.
- Skipping examples — paired in-voice/not-in-voice samples are the strongest signal.
- Trusting the first draft — AI lands on the safe middle; push for a sharper second pass.
- Letting clichés through (“In today’s fast-paced…”, “unlock the power of…”).
- No fact-check pass — the model is sometimes confidently wrong.
How to push results further
- Always supply 1-2 tone examples; “be friendly” alone is noise.
- Constrain ruthlessly — word count, banned phrases, must-include facts.
- Read aloud before publishing — if you stumble, rewrite.
- Cut adverbs and adjectives that don’t carry weight.
- Use AI for the first two drafts, human-edit the third — the third is what ships.
- Anchor the brief to a real person from your customer list.
- Test the headline alone — does the message survive without the body?
FAQ
- How long should the voice doc be?: 1-3 pages. The one-page card (prompt 8) is what people actually use; the longer version is the appendix.
- Should voice differ per channel?: Axis position can shift (more casual on social, more measured in docs); the anchor adjectives stay constant.
- How do I get the team to actually use it?: One-page card + worked examples + one weekly audit (prompt 7) shared in the team channel.
- Can AI maintain voice?: With the paste-ready system prompt from template 11 — yes, for first drafts. Claude Styles can also learn from a sample; a human still edits before publishing.
- Which model should I use?: Claude Opus 4.7 for extraction and audits over many samples; Sonnet 4.6 or GPT-5.5 for fast drafting. See the table above.
- How do I know my voice is working?: Outsiders can read three pieces and tell they’re from the same company.
- When should I refresh it?: Yearly, or whenever the audience or market shifts significantly.
Related
- Brand story prompts
- Tone rewrite prompts
- Homepage copy prompts
- Article rewrite prompts
- How to Write a Brand Tone Guide With AI (1-Page Template)
- Writing & Copywriting Prompts hub
Tags: #Prompt #Writing #Brand voice