Brand voice
How the AI writes in your tone — instructions, examples, and which text types they apply to.
On this page
An AI with no instructions writes averagely — the way the internet writes. Brand voice is where you tell it how you write: in what tone, in which words, and what you avoid. It is attached to every text the AI generates.
Controls on this screen
- Use brand voice — While this is off, neither the description nor the flags are sent anywhere. Turn it on once the description is finished.
- Voice description — Free text, up to 2,000 characters: tone, style, personality. What you would tell a new copywriter belongs here.
- Character counter — The limit is 2,000. It is not a target — shorter and more specific works better.
- Green flags — Phrases and habits the AI should use. Type and confirm with Enter.
- Red flags — What to avoid. Superlatives, overblown promises, words your brand does not use.
Writing a description that changes something
The description is the only place the AI learns it is writing for your brand. The difference between one that works and one that does not is specificity.
| Instead of | Write |
|---|---|
| "Write professionally." | "Short sentences. We explain what a parameter is good for, not what it means." |
| "Be friendly." | "We address the reader directly — “you will need”, not “it is necessary”. No exclamation marks." |
| "Focus on quality." | "For every product we state the material and the origin if we know them. If we do not, we say nothing." |
Flags versus the description
The description is prose; the flags are a list. Divide them like this:
| Field | What it is for |
|---|---|
| Voice description | Stance and tone — what cannot be reduced to bullet points. |
| Green flags | Specific phrases and themes you want to see repeatedly. |
| Red flags | Specific words and claims you never want to see. |
Where the brand voice applies
To texts the AI writes: product descriptions, short descriptions, shop copy, blog articles. It does not apply to translations of existing text (there the job is to translate faithfully, not to rewrite) or to filling parameters, where the point is the value, not the style.
How do I tell whether the brand voice is really working?
Generate one description before turning it on and one after. If the difference is hard to see, the description is too general — add concrete sentences and one prohibition.
Can each language have its own voice?
There is one description and it applies to every language. If the tone should differ in another language, write that into the description as a sentence.
Does the brand voice increase token usage?
Slightly — the description goes out with every request. You can see it in the AI logs. At 2,000 characters it is negligible next to the length of the output itself.