Gains:
- Ability to define brand voice and tone with concrete adjectives, example sentences and a do/don't do list and teach it to artificial intelligence
- Ability to align AI output to a consistent brand voice with sample texts (few-shots) and style guide
- Understanding that the ultimate guardian of the brand voice is the author, and that control of consistency belongs to humans.
Two brands may sell the same product but “speak” completely differently. Someone says to you like a friend on the street, "Come on, try this, you won't regret it"; the other says, “Our product is built with engineering that redefines industry standards.” That difference is the brand voice: a brand's consistent editorial personality. Tone is that persona's emotional setting in a given text—the same brand can be enthusiastic in a congratulatory post or serious in an apology; but the same "person" speaks in both cases. Brand voice is a brand's most valuable but most perishable asset. And that's where the biggest risk in the age of AI lies: AI speaks "average" by default, meaning it tends to erase your brand's voice and reduce it to a tone that sounds like everyone else.
The purpose of this unit is to teach AI to speak with your brand's voice. Let's lay out the basic principle from the beginning: The ultimate guardian of brand voice is the author. AI writes in your brand's voice as much as you teach it; Consistency checking and final adjustment always belong to the human.
Embodying the brand voice: from abstract to exemplary
Most brands define their voice with two abstract words like “friendly and professional.” This is inadequate for AI (and for most humans); Because "sincere" is different for everyone. There are three concrete tools for teaching brand voice to AI:
1. Descriptive adjectives (3-5) + opposite. Add "but not that" next to each adjective. This limit sharpens the sound.
- Sincere, but not cavalier.
- Knowledgeable, but not patronizing.
- Short and clear, but not rude.
2. Example sentences (few-shot). Give the AI 3-5 real sentences written in the brand's voice. The "few-shot" (few examples) technique provides the model with a concrete pattern rather than an abstract description and significantly increases consistency. AI learns rhythm, word choice and sentence length from examples.
3. To-do/don't-do list. Write clearly what you will use and what you will never use:
- Use: "you" language, short sentences, everyday examples.
- Using: corporate clichés like "we are proud to serve our customers", barrage of exclamations, English jargon.
Tip: Define your brand voice once and save it as a “style guide.” Paste this guide at the top of the prompt for each new content session. Thus, every text carries the same voice; you don't describe the sound anew each time.
Brand voice card: copyable template
Fill out the template below and attach it to the beginning of each prompt. This is the single document that gives AI the “sound identity” of your brand.
BRAND VOICE CARDBrand: [noun] — [single sentence description]Personality adjectives: [adjective 1] (but not [opposite]), [adjective 2] (but not [opposite]), [adjective 3] (but not [opposite])Address: [e.g. "you" language, friendly / "you" language, respectful]Sentence style: [e.g. short, direct / fluent, descriptive]Example sentences (write in this voice):1. "[actual phrase of the brand]"2. "[second example]"3. "[third example]"USE: [words/phrases]DO NOT USE: [prohibited words/clichés]
Sound application prompt:
Write exactly following the BRAND VOICE CARD below.[paste card here]Task: Write [content]. After writing, explain in one sentence why the 3 words/phrases you used fit the voice of this brand (self-check).
The last line (“self-monitor”) forces the AI to justify its choices; This makes it easier to spot deviations from the sound.
Tone: same sound, different tuning
The brand voice is constant; The tone varies depending on the situation. Same friendly brand:
- Celebration (new product): enthusiastic, energetic.
- Crisis/apology (after mistake): serious, taking responsibility, calm.
- Tutorial (how-to): patient, explanatory.
- Selling (campaign): persuasive but not pushy.
Tell the AI the tone separately with each text: “This is an apology email; keep the tone serious and sincere, no jokes.” The sound card + situation-specific tone instruction work together.
The table below shows the same message in three different brand voices (message: “a new feature is out”):
brand voice
example sentence
friendly / young
"Our new feature is here, be the first to try it!"
Corporate / reliable
"We present to you our new feature that improves the user experience."
Expert/technical
"In the new version, we added an automation layer that speeds up your workflow."
Weak prompt / Strong prompt
Weak prompt:
Write a post with the voice of our brand, our voice is sincere.
This claim is erroneous: "intimate" alone does not yield patterns. The AI maintains an overall “friendly” tone; Whether it coincides with the true voice of the brand remains a coincidence.
Powerful prompt:
BRAND VOICE CARDBrand: a yoga and breathing practice—offering a “judgment-free” space for beginners.Personality adjectives: warm (but not stilted), calm (but not sleepy), encouraging (but not forceful).Address: “you” language, soft.Sentence style: short, breathing, simple.Example sentences:1. "If you have five minutes today, that's enough."2. "There's no such thing as getting it right; just get started."3. "Listen to what your body is telling you." USE: "you", "today", "enough", breath/moment emphasis. USE: "must", "should", performance language, barrage of exclamations. Task: Write a 3-sentence Instagram caption announcing a new "night relaxation" session. Then explain why the 2 words you used fit the sound.
The difference is clear: the second prompt gives the AI a concrete world of rhythms and words to imitate; Limits such as the "should" ban protect the voice.
three mini cases
Case 1 — Voice coherence gain. A five-person social media team was producing 120 posts each month; The brand voice was messy because each writer's understanding of "sincere" was different. They created a common brand voice card and added it to each prompt. A month later, in a reader survey, perception that “the tone of the brand is familiar and consistent” increased significantly; the voice was no longer connected to the person but to the card.
Case 2 — The power of the few-shot. One corporate brand was just calling AI “professional writing” and the output was coming out cold. They added 4 actual paragraphs written in the brand voice as examples. From the examples, YZ captured the brand's unique "we solve it with you" tone; editing time halved per text.
Case 3 — Tone error. A brand had AI write its apology email after a system crash, without specifying the tone. The AI uses the brand's default "cheerful" voice and says, "There was a little mishap, but don't worry!" He produced a light sentence like: This would further anger aggrieved users. The editor noticed and reprinted the tone as "serious, taking charge." Lesson: voice is fixed, but tone is specifically stated in each text.
Common mistakes
- Describing sound with a single abstract adjective. It does not give "intimate" patterns; Example sentences and boundaries are required.
- Not giving example sentences (few-shots). The model cannot find a concrete rhythm to imitate.
- Not adjusting the tone according to the situation. The apology and the congratulation should have the same voice but different tone.
- Rewriting the sound card every session. Prepare and save once, paste over and over again.
- Not putting a ban list. Without a “do not use” list, AI slides into clichés and jargon.
Caution: AI may slowly drift from the voice over a long text; The first paragraph fits the brand, the following paragraphs gradually turn into "average". In long texts, check each section separately for audio; Don't just look at the first paragraph and approve the whole thing.
In summary
Brand voice is the consistent personality of a brand, while tone is the setting of that personality according to the situation. The AI speaks “average” by default; To speak with your brand's voice, you need to give it concrete adjectives (with its opposite), example sentences (few-shots) and a do/don't list. Collect this in a brand voice card and add it to each prompt. Specify the tone separately in each text. But remember: you are the ultimate guardian of sound; It's your job to detect and fix the AI's audio drift in long texts.
Application task
For a brand (your own brand, your customer's, or a fictitious brand), complete the brand voice card above: 3 adjectives (with opposites), appeal, sentence style, 3 example sentences, use/don't use lists. Then ask the AI for two different pieces of content with this card: a congratulatory post (enthusiastic tone) and an apology message (serious tone). Evaluate whether the two texts convey the same brand voice and whether the tone has changed correctly.
checklist
- [ ] Have I embodied the brand voice with adjective + antonym + example sentence?
- [ ] Have I added few-shot example sentences to the prompt?
- [ ] Have I defined a "do not use" (prohibited) list?
- [ ] Have I specified the context-specific tone in each text?
- [ ] Have I checked the long text section by section for any deviations from the sound?