Gains:
- Ability to transform abstract brand adjectives into a brand voice system prompt consisting of concrete rules, examples and things to avoid.
- Ability to consistently maintain the same brand tone across web, social media, email and advertising channels
- Ability to set up a reusable audit prompt that scores brand voice output
One of the most valuable but most invisible assets of a brand is its ability to speak with the same "voice" everywhere. Whether the customer encounters it on your website, on Instagram, or in an email; He must feel the same character, the same warmth, the same confidence. This is called brand voice in marketing: it is the set of personality and tone that describes how your brand would speak if it were a person. AI can be an excellent tool to maintain this consistency; but only when you describe your brand's voice correctly. In this unit, you will learn how to translate abstract adjectives into concrete rules that the model will understand, set up a reusable system prompt (fixed instruction text added at the top before each request), and create a control mechanism that scores the output.
What is brand voice, why does the model get it wrong?
Most teams tell the model: “Write friendly copy that aligns with our brand.” Here the model interprets “sincere” according to its average definition, often producing a generic, everyone-like text. Because adjectives like "sincere", "professional", "energetic" etc. are meaningful in your mind; not in the model's head. The model produces the “average sincere” it learns from millions of texts, and the result is indistinguishable from your competitors.
The solution is to transform abstract adjectives into concrete rules and examples. Define brand voice in three layers:
- Tone rules: How do we speak? (e.g. "We address the reader as 'you'", "Sentences are short", "We use emojis, but no more than two per post")
- Sample sentences: What does it mean to talk like us? (2-3 real examples, because the model learns more from example than rule)
- Things to avoid: What we never do? (e.g. "Exaggerated promises, barrage of exclamations, technical jargon")
These three layers are a marketing adaptation of the technique called "few-shot": when you show the model a few correct examples, it catches the pattern and imitates it.
Step by step: setting up a brand voice system prompt
- Gather your best available texts. Put together 5-10 texts that you say as a team "this is exactly our voice".
- Remove common patterns. What are the sentence lengths, form of address, frequently used and never used words?
- Fill three layers. Tone, example, things to avoid.
- Type the system prompt and save it. Keep this somewhere; you will paste it at the beginning of each content request.
- Test and retouch. Experiment with different types of content and sharpen the guide.
Weak prompt / Strong prompt
Weak prompt:
Write a friendly Instagram caption that aligns with our brand. Product: coffee subscription.
This prompt produces text that is indistinguishable from thousands of competitors.
Powerful prompt:
Write an Instagram description according to the brand voice guide below.BRAND VOICE:- Tone: Warm, friendly yet knowledgeable. Like a barista friend.- Address: We address the reader as "you".- Sentences are short and breathing. No long paragraphs.- We use emojis, but no exaggeration (2 per post max).LIKE US EXAMPLE SENTENCE: "The best part of the morning is the silence you hear as you wait for that first sip." WE NEVER MAKE:- Unprovable claims like "The best coffee in the world."- A barrage of exclamations ("Great!!! Amazing!!!")- Technical jargon ("micron-adjusted grinding rate").PRODUCT: Monthly home coffee subscription. Different region every month. REQUIRED: 3 different description variants, each with a maximum of 40 words.
The difference is like night and day. In the second prompt, the model speaks with your voice, in the first prompt with everyone's voice.
Tip: Write your brand voice guide once and save it as a system prompt. When everyone on the team uses the same manual, the output sounds the same no matter who produces it. This frees the brand voice from being dependent on one person.
three mini cases
Case 1 — Tone shift is caught. The social media team of an e-commerce brand was working with three separate freelancers. The text of each carried a different "voice"; some are very formal, some are extremely humorous. By mandating the brand voice system prompt to all three, the editor's proofreading time dropped from 12 minutes per post to 4 minutes; In the three-month measurement, brand followers' "I find the brand's personality clear" survey increased from 58% to 79%.
Case 2 — New product launch. A software company (SaaS, i.e. subscription-based cloud software) has released a new module. Thanks to the guide, 20 text variants for 4 channels (blog, LinkedIn, email, in-app notification) were produced in two hours; Since they all carried the same sound, the brand team approved them in one round. In the previous launch, this process took two days.
Case 3 — Moment of crisis. There was a malfunction in the payment system of a brand. Apology texts written in times of panic are usually either too cold or too defensive. A “crisis tone” layer was added to the brand voice guide: “Take responsibility, don't get defensive, give clear solutions and dates.” The model produced a draft apology true to this tone; The team published it in 15 minutes, and the tone of negative comments on social media softened significantly.
Copiable templates
Template 1 — Brand Voice System Prompt:
You are a content expert writing for [BRAND NAME]. Stick to this tone in all outputs: TONE: [3-4 adjectives + short description] ADDRESS: [you/you + level of formality] SENTENCE STRUCTURE: [short/long, plain/literary] WORD PREFERENCES: We use: [...]. We don't use: [...]LIKE US 2 EXAMPLE SENTENCES:1. "[...]"2. "[...]"NEVER DO: [3-5 items to avoid]TARGET AUDIENCE: [who, what expects]If a request comes in that doesn't comply with the rule, suggest the version that follows the rule first.
Template 2 — Cross-channel adaptation:
Tailor this message across three channels, keeping the same brand voice:1. LinkedIn (more informative, slightly longer)2. Instagram (visually focused, short, warm)3. Email subject line (intriguing, no more than 6 words)Do not change the voice, only the length and format.
Template 3 — Brand voice audit prompt:
Score the following text from 1-5 according to my brand voice guide.Guide: [paste]Text: [paste]Give points and one-sentence justification for each criterion:- Tone harmony- Accuracy of address- Adaptation to what to avoid- General “is it our voice?” feelingThen suggest corrected version for each criterion below 3 points.
Template 4 — Extracting rule from example:
Below are 5 sample texts from our brand. From these, extract common brand voice rules: tone, address, sentence structure, frequent words, avoidances. Export the output in a system prompt format that I can reuse.TEXT: [paste]
Consistency across channels
The brand voice remains the same, but the format varies by channel. The table below shows how the same sound sounds on different channels.
channel
Sound (fixed)
Format (variable)
Typical length
Warm, knowledgeable
Instructive, providing evidence
100-200 words
Warm, knowledgeable
Visually focused, short
30-50 words
Warm, knowledgeable
Personal, single CTA
80-150 words
ad headline
Warm, knowledgeable
Striker, net benefit
5-8 words
Caution: Update the brand voice guide over time. Refresh the guide whenever there is a new campaign language, a new audience, or a change in tone; Otherwise, the model will continue to write in the old voice and the content will feel "in the past".
Common mistakes
- Be content with adjectives. It is thought that saying "be sincere" is enough; whereas the example sentence for the model is much more informative than the adjective.
- Skipping the avoidables. Saying what you won't do is as important as saying what you will do; The barrage of exclamations or clichés only stops when they are expressly prohibited.
- Write the guide once and forget it. The brand evolves; If the guide is not updated, the output will become stale.
- Rewriting the guide for each content. Writing it manually each time instead of keeping it as a system prompt wastes time and creates inconsistency.
- Dependency on one person. If the guide is not shared, the brand voice becomes the voice of "the person who wrote that text" and disappears when that person leaves.
In summary
- Brand voice is how your brand would speak if it were a person; It should be defined concretely, not abstractly.
- Write abstract adjectives in three layers: tone rules, example sentences, and avoidances.
- Write the guide once, save it as a system prompt and add it to the beginning of each content.
- The audio remains constant, the format varies by channel.
- Make consistency measurable by scoring the output with an audit prompt.
Application task
Write a brand voice system prompt for your own brand (or a fictitious brand) by filling in Template 1. Then have the same product message adapted to three different channels with this guide (Template 2). Finally, have one of the texts you produced using Template 3 scored according to your own guide and correct the criteria below 3 points. Duration: approximately 25 minutes.
checklist
- [ ] I defined the brand voice in three layers: tone, exemplar, and avoidance.
- [ ] I've included at least two actual "like us" example sentences.
- [ ] I saved the guide as a reusable system prompt.
- [ ] I adapted the same sound to at least two channels and preserved the format difference.
- [ ] I scored the output with the audit prompt and corrected the weak criteria.
- [ ] I linked the manual to the update schedule (e.g. quarterly).