Define voice with examples, not adjectives
Most brand voice docs are useless to an AI because they are lists of adjectives: confident, approachable, bold. A model cannot turn those into consistent output because it has no shared definition of them with you. Voice is far better captured as examples. Ten of your best real posts teach a model your rhythm, vocabulary, and range in a way no adjective list can.
Voice guidelines: be confident, friendly, and authoritative while staying approachable and human.
We write short. We start with the point, not a wind-up. We say I think, not it could be argued. We never use words like leverage, seamless, or unlock. Here are ten posts that sound exactly right: [paste].
The second is something a model, or a new teammate, can actually follow. It names the rules and shows the examples. The first is a mood board.
Turn your voice into a profile the AI uses
Once you have examples and a short list of rules, the AI needs to use them on every draft, not just when you remember to paste them. This is where a voice profile beats prompting from scratch each time.
- Sentence length and rhythm: short and punchy, or longer and reflective.
- Vocabulary: the words you use and the words you ban.
- Point of view: first person, we, or observational.
- Range: what you will post about and what is off-brand.
- Formatting habits: whether you use line breaks, lists, or none.
Guardrails: the nos matter more than the yeses
A voice is defined as much by what it refuses as by what it does. Give the AI a hard list of things you never say. Negative constraints are more reliable than positive ones because they are unambiguous. It is easier for a model to obey never use hashtags than to obey be witty.
| Guardrail | Why it holds the voice |
|---|---|
| No buzzwords (seamless, unlock, supercharge) | These read as generic marketing and break a human voice instantly |
| No em dashes for drama | A common AI tell; commas and short sentences read more human |
| No hashtags or emoji unless we always do | Consistency is the whole point of a voice |
| No claims we cannot back | Protects trust, which is part of the brand |
| No wrap-up moral at the end | Lets the post land instead of over-explaining |
One human owns the final read
Voice drifts fastest when responsibility is spread across a team and an AI with no single owner. The fix is not more process, it is one person who reads every post before it ships and has the authority to say that is not us. AI can get a draft 90 percent on-brand; the last 10 percent, the judgment call about whether it sounds right, is a human job.
- It sounds like our examples, not like generic marketing.
- No banned words, no drama punctuation, no hashtags we do not use.
- The point comes first, not after a wind-up.
- It says something we would actually stand behind.
- If I saw it with no logo, I would still know it was us.