Brand Voice Training for AI

AI defaults to a generic professional voice because that is what its training data contains most of. Directing AI toward a specific brand voice requires giving it the right inputs: exemplary documents written in that voice, explicit description of the voice's characteristics, and negative examples that illustrate what the voice is not. This is not a one-time configuration — it is a prompt engineering discipline that improves over time as the team learns what inputs produce the most on-brand outputs.

1. Create a voice brief for prompt templates Every prompt template used to generate brand-voice content should include a voice brief: three to five sentences describing the brand's writing characteristics (direct, specific, confident, no filler phrases), two to three exemplary phrases the brand uses, and two to three phrases the brand never uses. The voice brief is the human-readable style guide translated into prompt language. It should be updated as the brand voice evolves and as the team learns which descriptions produce the best AI outputs.
2. Provide exemplary documents as context When possible, include one or two exemplary documents — written by humans in the correct brand voice — as part of the prompt context. "Write the new executive summary in the voice and style of the following exemplary document" produces more on-brand output than describing the voice in the abstract. The exemplary document is the fastest way to communicate voice characteristics that are difficult to verbalize.
3. Test systematically and refine the brief After each production cycle, evaluate whether the AI output matched the brand voice. If not, identify the specific failure: was the tone too formal? Too hedged? Too generic? Did it use phrases the brand does not use? Document the failure and update the voice brief to address it. The voice brief improves through production experience — it cannot be written perfectly from first principles and should not be treated as a fixed document.

## Brand Voice Brief — Prompt Section

VOICE CHARACTERISTICS
- [Characteristic 1]: [Description and example]
  Example: Direct. State the finding before the evidence.
- [Characteristic 2]: [Description and example]
  Example: Confident. Do not hedge when the evidence is clear.
- [Characteristic 3]: [Description and example]
  Example: Specific. Use numbers, names, and dates, not vague claims.

PHRASES WE USE
- "[Phrase 1]"
- "[Phrase 2]"

PHRASES WE DO NOT USE
- "I hope this finds you well" — always cut
- "Please do not hesitate to reach out" — replace with specific invitation
- "It is important to note that..." — if it is important, state it directly
- "Synergize" / "leverage" / "value-add" — never

EXEMPLARY DOCUMENT (paste or reference):
[Paste 1–2 exemplary paragraphs from approved documents]