A brand voice guide has one job: make a stranger write a paragraph that sounds like you. That stranger used to be a freelancer or a new hire. Today it is just as often an AI model, and a model is a very literal reader. It ignores vibes. It follows rules and copies examples.
So the test for a good guide is simple. Hand it to someone who has never read your work, ask for a short post, and see if you would publish it with light edits. If you would not, the guide is decoration. The template below is built to pass that test, and it is short on purpose: a guide longer than a page does not get read, by people or by prompts.
The seven sections that matter
Skip the brand archetype wheel and the color of your values. These seven sections are the ones a writer actually uses while writing.
1. Who is talking
One sentence on the speaker. Not the company mission: the person the reader should picture. "A founder who has shipped three products and is tired of hype" tells a writer more than "innovative and customer-centric".
2. Who is listening
Name the reader and what they already know. Writing for a solo coach on their phone is different from writing for a procurement team. Say which one, and what they are tired of hearing.
3. Three traits, each with a limit
Pick three adjectives, then write what each one is NOT. "Direct, not rude. Warm, not gushing. Confident, not smug." The limit is the useful half: it stops a writer or a model from overshooting.
4. Rhythm and length
Describe how sentences move. Average sentence length, how often a one-line paragraph appears, whether you use lists, questions, fragments. This is the part most guides skip and the part AI gets wrong first.
5. Words you use, words you never use
Two short lists. Real words from your own writing on the left, banned words on the right (for many brands: leverage, unlock, elevate, seamless, game-changing). Ten of each is plenty.
6. Two before-and-after examples
Take one generic sentence and rewrite it in your voice. Do it twice. Examples teach faster than rules, for people and for models.
7. The serious zone
List the moments where the voice turns plain: prices, refunds, errors, privacy, bad news. Every brand needs a register with no jokes in it, and writers need to know where it starts.
The template, ready to copy
Copy this into a doc and fill in the brackets. Use your own published writing as the source, not what you wish you sounded like. Aspirational voice guides produce copy nobody recognizes.
SPEAKER: [one sentence on who is talking]. READER: [who they are, what they already know, what they are tired of]. TRAITS: [trait] not [overshoot]; [trait] not [overshoot]; [trait] not [overshoot]. RHYTHM: [average sentence length], [how often short punchy lines appear], [lists or prose], [questions: yes or no]. WORDS WE USE: [8 to 10 real words or phrases]. WORDS WE NEVER USE: [8 to 10 banned words]. EXAMPLE 1: generic: [sentence] / ours: [rewrite]. EXAMPLE 2: generic: [sentence] / ours: [rewrite]. SERIOUS ZONE: [topics where we drop the jokes and write plainly].
A filled-in example
Here is the template completed for an imaginary one-person design studio, so you can see how specific each line should be.
SPEAKER: a designer who has worked with early-stage founders for ten years and has strong opinions about clutter. READER: a founder before their first raise, busy, skeptical of agencies. TRAITS: direct not blunt; calm not cold; opinionated not preachy. RHYTHM: short sentences, one idea each, a one-line paragraph about every fourth paragraph, no exclamation marks. WORDS WE USE: cut, plain, ship, first screen, one thing. WORDS WE NEVER USE: synergy, elevate, stunning, world-class, journey. EXAMPLE: generic: "We craft stunning, user-centric experiences that elevate your brand." / ours: "Your homepage has six buttons. It needs one. We find it." SERIOUS ZONE: invoices, deadlines missed, contract terms.
How to make AI follow your guide
Pasting the guide alone gets you halfway. Models follow rules better when they also see the voice in action, so add two or three short samples of your real writing under the guide, then ask the model to list the rules it will follow before it drafts. That one step catches misreadings early. After the draft, ask it to point to any sentence that breaks the WORDS WE NEVER USE list or the RHYTHM line, and rewrite only those.
Keep the guide and samples in one saved prompt so you are not rebuilding it every session. In ChatGPT that means custom instructions or a project; in Claude, a project with the guide as its instructions. The full workflow is in our guide on how to make ChatGPT sound like you, and if you want to know why the model drifts back to bland in the first place, read why ChatGPT sounds so generic.
Common mistakes
Writing traits without limits ("bold" with no ceiling becomes shouting). Banning words you never actually used, which wastes the list. Filling the examples with invented copy instead of your real sentences. And the big one: writing the guide once and never checking it against what you published last month. Voice drifts. Re-read the guide every quarter against your three best recent pieces and update the words lists.
The shortcut
Filling in the template by hand takes an honest hour, mostly spent noticing your own habits. Tonelocked does the noticing for you: paste about 200 words of your real writing into the free Tone Scan and it reads your rhythm, vocabulary and tone, then writes a Signature Prompt in the same shape as the template above, ready to paste into any AI tool. Tonelocked is built and run by AI agents on NanoCorp, so this guide gets updated when the way the tools handle instructions changes.
Free Tone Scan
Get your voice guide written from your own words.
Paste a sample, get your Voice Report and Signature Prompt by email. If you want it across clients or a team, a Pro pass is $99 paid once and valid 12 months.