Content Writing · Editing & voice
The Best Prompt for Defining a Brand Voice
The best prompt for brand voice works from evidence. This prompt has the model analyze your writing samples first, quoting the patterns it finds in sentences, word choice and attitude, and only then write a guide. Each voice trait must be backed by quotes from the samples or labeled as a suggestion. You get traits, tone guidance by channel, a word list, mechanics, rewrites and a short version to use as AI instructions.
Brand Voice Guide Prompt
Analyzes your existing writing samples and derives a usable voice guide: traits with evidence, tone by situation, word lists and before-and-after rewrites.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Intermediate
- Complexity
- Multi-part reply
- Variables
- 5
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
5 variables · fields marked * matter most
# Role You are a brand strategist and editor. You work out how a brand writes from what it has already written, and you describe it in terms another writer could follow on their first day. # Inputs - Brand: [Brand] - Who it writes for: [Who the brand writes for] - Where the guide will be used: [Where the guide will be used] - Known things to avoid: [Things to avoid] <samples> [Writing samples] </samples> # Stage 1: analyze before you describe Read all the samples and report what the writing does: - Sentences: typical length, rhythm, use of fragments, questions and lists. - Point of view: how it refers to itself and to the reader. - Word choice: plain or technical, contractions, recurring phrases, words it seems to avoid. - Attitude: how direct it is, and how it handles humor, enthusiasm, bad news and claims about itself. - Mechanics: capitalization, punctuation habits, how headings and calls to action are phrased. Quote short phrases from the samples for each observation. Separate patterns that appear in most samples from things that appear once. Where samples contradict each other, say which version you think is the intended voice and why. # Stage 2: the voice guide 1. The voice in two sentences, specific enough that it could not describe a competitor. 2. Three to five voice traits. For each: a name, what it means in practice, how it goes wrong when overdone, and two quotes from the samples that show it. 3. Tone by situation: how the writing shifts for the channels above and for one difficult moment, such as an apology or a price change. The voice stays the same; say what changes. 4. Word list: words and phrases we use and ones we do not, drawn from the samples and the avoid list. 5. Mechanics: the rules a proofreader would need. 6. Three before-and-after pairs: a generic sentence, the same sentence in this voice, and what changed. 7. A 100 to 150 word version of the guide written as instructions for an AI writing assistant. # Constraints - Label every trait and rule OBSERVED (backed by quotes from the samples) or SUGGESTED (your extrapolation). - Describe behavior, not adjectives. A trait such as friendly or professional is only useful once it says what the writer does differently. - If the samples are too few, too short or all one type of content, say what you cannot determine and which additional samples would settle it. - Do not present the avoid list as a finding unless the samples agree with it. # Output format Stage 1 as a short findings list with quotes. Stage 2 under the seven numbered headings, with tables for the traits and the word list.
Use it when you are
- Documenting a founder's or small team's writing style so new writers and freelancers can match it
- Creating voice instructions to reuse in AI writing tools
- Checking whether your website, emails and social posts sound like the same brand
- Preparing a starting draft for a formal brand guidelines project
Why this prompt works
Analysis is a separate, visible stage
If the guide is written in one pass, the model tends to name pleasant traits and then find quotes to fit. Putting a sentence-level analysis first, with quotes, reverses that order. The traits in stage 2 have to come out of findings you have already seen and can dispute.
OBSERVED and SUGGESTED keep evidence and opinion apart
A voice guide mixes what the brand does with what someone thinks it should do. Forcing a label on every trait and rule shows which parts rest on your samples and which are the model's extrapolation, so you can accept the first and decide deliberately on the second.
Traits are defined by behavior and by their failure mode
Adjectives such as friendly, bold or professional describe most brands and guide no one. Requiring what each trait means in practice, plus how it goes wrong when overdone, gives a writer limits on both sides: enough direction to act and a warning about caricature.
Contradictions and thin evidence are reported
Real samples disagree with each other, and a few short samples cannot support a full guide. The prompt asks the model to name contradictions, choose a side with a reason, and state what it cannot determine. That is more useful than a confident guide built on one paragraph.
The guide ends in forms people will use
Before-and-after pairs teach a voice faster than description, and the short version written as assistant instructions can be pasted straight into an AI tool. Both turn the guide from a document into something a writer applies on the next piece of copy.
How to use it
- Collect three to six pieces of writing you consider on-brand, from at least two channels. Leave out pieces you are not happy with.
- Label each sample by type and paste them all into the samples field.
- List the channels where the guide will be used so the tone section covers them.
- Read stage 1 and check that the quoted patterns are ones you recognize. Tell the model which contradictions it resolved wrongly.
- Go through the SUGGESTED items and accept, change or delete each one.
- Test the guide: give the section 7 instructions to a model in a fresh conversation, ask for a short piece, and compare it with your samples.
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | Paste the 100 to 150 word version from section 7 into custom instructions or a project if you want later drafts to follow the voice without repeating the full guide. |
| Claude | Expected to work | Samples go inside XML-style tags. Anthropic's guidance suggests tagging each document separately when you supply several, so consider wrapping each sample in its own numbered tag within the block. |
| Gemini | Expected to work | If you supply many long samples, state how many there are at the top of the samples block so the analysis can refer to them by number. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: A fictional subscription coffee roaster with a small team wants a guide for a new freelance writer. Samples are shortened and illustrative.
# Role You are a brand strategist and editor. You work out how a brand writes from what it has already written, and you describe it in terms another writer could follow on their first day. # Inputs - Brand: Fieldnote, a subscription coffee roaster - Who it writes for: Home coffee drinkers who care about quality but are not hobbyists - Where the guide will be used: Website, weekly email, Instagram captions, support replies - Known things to avoid: Coffee snobbery, exclamation marks <samples> [Homepage] Good coffee, sent when you need it. We roast on Mondays and post on Tuesdays. That's most of the plan. [Email] This week's bag is from a farm we've bought from for four years. It tastes like plums, a bit. You don't have to taste the plums. [Product page] Medium roast. Works in whatever you already own. If you're not sure how to grind it, ask us and a person will answer. [Support reply] Sorry about that. Your bag went out late, which is on us. A replacement is in the post today and this month is free. </samples> # Stage 1: analyze before you describe Read all the samples and report what the writing does: - Sentences: typical length, rhythm, use of fragments, questions and lists. - Point of view: how it refers to itself and to the reader. - Word choice: plain or technical, contractions, recurring phrases, words it seems to avoid. - Attitude: how direct it is, and how it handles humor, enthusiasm, bad news and claims about itself. - Mechanics: capitalization, punctuation habits, how headings and calls to action are phrased. Quote short phrases from the samples for each observation. Separate patterns that appear in most samples from things that appear once. Where samples contradict each other, say which version you think is the intended voice and why. # Stage 2: the voice guide 1. The voice in two sentences, specific enough that it could not describe a competitor. 2. Three to five voice traits. For each: a name, what it means in practice, how it goes wrong when overdone, and two quotes from the samples that show it. 3. Tone by situation: how the writing shifts for the channels above and for one difficult moment, such as an apology or a price change. The voice stays the same; say what changes. 4. Word list: words and phrases we use and ones we do not, drawn from the samples and the avoid list. 5. Mechanics: the rules a proofreader would need. 6. Three before-and-after pairs: a generic sentence, the same sentence in this voice, and what changed. 7. A 100 to 150 word version of the guide written as instructions for an AI writing assistant. # Constraints - Label every trait and rule OBSERVED (backed by quotes from the samples) or SUGGESTED (your extrapolation). - Describe behavior, not adjectives. A trait such as friendly or professional is only useful once it says what the writer does differently. - If the samples are too few, too short or all one type of content, say what you cannot determine and which additional samples would settle it. - Do not present the avoid list as a finding unless the samples agree with it. # Output format Stage 1 as a short findings list with quotes. Stage 2 under the seven numbered headings, with tables for the traits and the word list.
What a good response should contain
- Stage 1 findings that quote the short declarative sentences, the contractions, and the understated humor such as 'That's most of the plan'
- Traits named by behavior, for example saying less than it could or admitting fault plainly, each with two sample quotes
- A note that four short samples limit confidence, with a request for longer pieces
- A tone section showing how the support reply keeps the same plainness while dropping the humor
- The avoid list items marked OBSERVED only where the samples back them up, such as the absence of exclamation marks
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- The guide describes the samples you chose. Unrepresentative or off-brand samples produce an unrepresentative guide.
- It captures how the brand writes, not positioning, values or strategy, which need input the samples do not contain.
- A few short samples support only tentative conclusions; look for the model's own statement of what it could not determine.
- A voice guide needs agreement from the people who write. Treat the output as a draft to discuss, not a finished standard.
Alternative prompts
Content Writing
Editing and Proofreading Prompt
Edits a draft at the level you choose, protects the author's voice, and returns a change log plus queries instead of silent rewrites.
- ChatGPT
- Claude
- Gemini
Graphic Design
Brand Identity System Prompt
Drafts a written identity system: color roles with contrast pairings, a type hierarchy, logo usage rules and imagery direction, tied to your positioning.
- ChatGPT
- Claude
- Gemini
Marketing
Evidence-Based Customer Persona Prompt
Builds personas from the interviews, reviews or survey answers you paste in, tagging every line as evidence, inference or assumption.
- ChatGPT
- Claude
- Gemini
Content Writing
Conversion Copywriting Prompt
Works out what a page must make the reader believe, then writes the copy in that order using only the proof you supply.
- ChatGPT
- Claude
- Gemini
Related resources
Technical guide
How to Use System Prompts
What a system prompt is, what belongs in it versus the user message, a worked example, and the limits to plan for when building on one.
Technical guide
Few-Shot Prompting: How to Use Examples in a Prompt
How to include worked examples in a prompt so the model follows a format, label set or style, and how to choose examples that generalize.
Frequently asked questions
How many writing samples do I need?
Three is a workable minimum, and they should come from different channels so the analysis can separate the constant voice from the changing tone. More and longer samples give firmer conclusions. If you have only one piece, the prompt will still run but should report low confidence.
What is the difference between voice and tone?
Voice is the consistent character of the writing: its vocabulary, rhythm and attitude. Tone is how that character adjusts to the situation, such as a product launch versus a service apology. The guide treats voice as fixed traits and covers tone in its own section by channel and situation.
We do not have a consistent voice yet. Can this still help?
Yes, as a diagnosis. Supply samples from different writers or channels and the stage 1 analysis will show where they diverge. Pick the samples closest to what you want, run the prompt again on those alone, and use the SUGGESTED items as a starting point for discussion.
How do I use the result with AI writing tools?
Take the short version from section 7 and place it in the tool's system prompt or custom instructions, then add two or three of your real samples as examples. Instructions plus examples generally hold a voice better than either alone.
Sources and references
- Prompt engineering overview — Anthropic
- Prompt engineering guide — OpenAI
Was this prompt useful?
No account needed. Do not include personal information.