STRATEGY

Brand voice guide extraction

Use when good copy already exists somewhere in the company but nobody has written down the rules behind it.

brand-voice-guide-extraction.md

Download .md

You are a voice and tone editor. Extract rules from existing copy rather than inventing an aspirational voice.

Copy samples that are considered good: {{GOOD_SAMPLES}}
Copy samples considered off brand, if any: {{BAD_SAMPLES}}
Where this voice is used: {{CHANNELS}}

Produce:
1. Five voice traits. For each: trait name, one sentence definition, and two quoted phrases from the good samples that demonstrate it.
2. Observable mechanics: average sentence length, typical opening move, person and tense, contraction use, use of questions, humor, jargon level, how numbers are presented. Give the observed pattern, not a preference.
3. A do and do not table with columns: Situation | Do this | Not this | Example rewritten from the samples.
4. Vocabulary: words the brand uses, words it avoids, and any word it uses in a specific way.
5. Punctuation and formatting rules observed, including how lists and headings are handled.
6. A 60 word test paragraph on {{TEST_TOPIC}} written in the extracted voice, followed by which trait each sentence demonstrates.

Constraints: every rule must trace to a quoted sample. If you cannot find evidence for a trait, do not include it. Do not use em dashes in the guide or the test paragraph, and do not write openers about a fast changing world.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{GOOD_SAMPLES}}
{{BAD_SAMPLES}}
{{CHANNELS}}
{{TEST_TOPIC}}

Getting a better result

Include the off brand samples; the contrast produces sharper rules than good samples alone.
Use samples from more than one writer, or you will document one person rather than the brand.
Keep the mechanics section - it is the part another writer can actually follow.

Questions about this prompt

When should I use this rather than writing a voice guide from scratch?

When copy people already like exists somewhere and nobody has written down why it works. A guide written from scratch describes a voice you do not have. This only keeps a trait when two quoted phrases from the good samples support it, and drops any trait it cannot evidence.

Which samples should I feed it?

Good samples from more than one writer, or you document one person rather than the brand. Include the off brand samples if they exist, since the contrast produces sharper rules than good samples alone. Name the channels, and pick a test topic that is representative rather than the easiest thing to write about.

What comes back, and which section can a writer follow?

Five evidenced traits, an observable mechanics section, a do and do not table, vocabulary, punctuation rules and a 60 word test paragraph annotated by trait. Keep the mechanics section. Sentence length, opening move, tense, contractions and how numbers are presented are the parts another writer can actually follow.

What is the mistake that turns the guide back into a wish?

Editing the mechanics towards what you wish the voice were. The moment it stops describing the samples, the guide becomes aspiration again and the next writer has nothing to check against. Same for traits: if the evidence is not there, leave the trait out rather than adding it back afterwards.