From Adjectives to Rules: Making Brand Voice Machine-Executable

Your brand voice guidelines probably say something like: "Be conversational and friendly. Avoid jargon. Use a warm tone."

These are adjectives and principles. They work for humans. They don't work for AI.

When you tell an AI to "be conversational," it has millions of training examples of conversational copy. It generates something that sounds conversational. But it's not your conversational voice. It's a generic approximation. It hallucinates.

Why Principles Fail

Principles are interpretations. They are subjective. Different people interpret them differently. Different AI models interpret them differently.

"Conversational" to a human might mean: casual, friendly, uses contractions, short sentences, asks questions.

"Conversational" to an AI might mean: uses informal language, lowercase letters, emojis, slang.

"Conversational" to another AI might mean: uses dialogue, back-and-forth exchanges, narrative structure.

Without explicit rules, the AI has to guess. It generates based on probability. It hallucinates.

What Rules Look Like

Rules are explicit. They are unambiguous. They are machine-executable.

Instead of "be conversational," you need:

Conversational means:

1. Use contractions (don't, can't, it's)

2. Use active voice, not passive voice

3. Use second person ("you") not first person ("we")

4. Use short sentences (under 15 words)

5. Ask rhetorical questions

6. Avoid: formal register, passive voice, industry jargon, nominalization

Conversational does NOT mean:

1. Lowercase letters

2. Emojis

3. Slang

4. Fragmented sentences

Now the AI has explicit boundaries. It knows what conversational means in your brand. It can execute on it.

Rules Need Examples

Rules alone are not enough. Rules need scored examples.

A scored example shows the AI what good looks like and what bad looks like.

Good example (conversational):
"You can save hours every week by automating your workflow. Here's how it works."

Bad example (not conversational):
"The implementation of workflow automation protocols can facilitate significant temporal resource optimization."

The AI learns from the contrast. It understands the difference between your conversational voice and generic conversational copy.

Without examples, rules are abstract. With examples, rules are concrete.

The Semantic Layer for Tone

Here is what a semantic layer for tone looks like:

{
  "tone_of_voice": {
    "name": "Conversational and Direct",
    "principles": [
      "Friendly but professional",
      "Clear and direct",
      "Helpful, not pushy"
    ],
    "rules": {
      "do": [
        "Use contractions",
        "Use active voice",
        "Use second person",
        "Use short sentences",
        "Ask rhetorical questions"
      ],
      "avoid": [
        "Passive voice",
        "Industry jargon",
        "Formal register",
        "Nominalization",
        "Lowercase letters"
      ]
    },
    "scored_examples": [
      {
        "text": "You can save hours every week by automating your workflow. Here's how it works.",
        "score": 1.0,
        "reason": "Uses contractions, active voice, second person, short sentences"
      },
      {
        "text": "The implementation of workflow automation protocols can facilitate significant temporal resource optimization.",
        "score": 0.1,
        "reason": "Passive voice, nominalization, formal register, jargon"
      }
    ]
  }
}

{
  "tone_of_voice": {
    "name": "Conversational and Direct",
    "principles": [
      "Friendly but professional",
      "Clear and direct",
      "Helpful, not pushy"
    ],
    "rules": {
      "do": [
        "Use contractions",
        "Use active voice",
        "Use second person",
        "Use short sentences",
        "Ask rhetorical questions"
      ],
      "avoid": [
        "Passive voice",
        "Industry jargon",
        "Formal register",
        "Nominalization",
        "Lowercase letters"
      ]
    },
    "scored_examples": [
      {
        "text": "You can save hours every week by automating your workflow. Here's how it works.",
        "score": 1.0,
        "reason": "Uses contractions, active voice, second person, short sentences"
      },
      {
        "text": "The implementation of workflow automation protocols can facilitate significant temporal resource optimization.",
        "score": 0.1,
        "reason": "Passive voice, nominalization, formal register, jargon"
      }
    ]
  }
}

{
  "tone_of_voice": {
    "name": "Conversational and Direct",
    "principles": [
      "Friendly but professional",
      "Clear and direct",
      "Helpful, not pushy"
    ],
    "rules": {
      "do": [
        "Use contractions",
        "Use active voice",
        "Use second person",
        "Use short sentences",
        "Ask rhetorical questions"
      ],
      "avoid": [
        "Passive voice",
        "Industry jargon",
        "Formal register",
        "Nominalization",
        "Lowercase letters"
      ]
    },
    "scored_examples": [
      {
        "text": "You can save hours every week by automating your workflow. Here's how it works.",
        "score": 1.0,
        "reason": "Uses contractions, active voice, second person, short sentences"
      },
      {
        "text": "The implementation of workflow automation protocols can facilitate significant temporal resource optimization.",
        "score": 0.1,
        "reason": "Passive voice, nominalization, formal register, jargon"
      }
    ]
  }
}

This is semantic data. The AI can query it. It can understand what your brand voice is. It can execute on it.

The Outcome

With semantic tone data, AI generates copy that is actually your voice, not a generic approximation. Without it, AI hallucinates. It generates copy that sounds like it could be your brand, but isn't.

The difference is not small. It is the difference between on-brand and off-brand. Between authentic and generic. Between your voice and a hallucination.

Your tone guidelines are failing AI not because tone is hard to execute. They are failing because guidelines are principles, not rules. Semantic layers translate principles into machine-executable rules.