Prompt Engineering / Role Prompting
Convert abstract tone instructions into concrete, demonstrable examples.
Why This Prompt Exists
“Make it professional but friendly” — everyone says this, no one can define it. The model can’t either.
You get:
- tone that varies unpredictably across responses
- stakeholders saying “that’s not our brand voice” (but can’t specify why)
- inconsistent tone between team members’ prompts
- abstract tone words that the model interprets differently each time
- no way to test if the tone is correct
But tone can be calibrated:
- lexical: specific words to use and avoid
- syntactic: sentence length, complexity, punctuation
- pragmatic: use of humor, empathy, directness
- examples: before/after demonstrations
- anti-examples: what this tone is NOT
Without calibration, tone is luck.
This prompt turns abstract tone instructions into precise, example-driven specifications.
The Prompt
Assume the role of a tone calibration expert who makes abstract instructions concrete. Your task is to convert tone descriptions into demonstrable examples and rules. Generate: 1. TONE TARGET - User's description: [quote] - Interpretation: [what this likely means in practice] 2. LEXICAL SPECIFICATION - Words/phrases to use - Words/phrases to avoid - Contractions allowed? (Yes/No/Selectively) - Jargon allowed? (Level) 3. SYNTACTIC SPECIFICATION - Average sentence length (short / medium / long) - Use of sentence fragments? (Yes/No) - Paragraph length (1-2 sentences / 3-5 / 5+) - Punctuation style (minimal / standard / expressive) 4. PRAGMATIC SPECIFICATION - Directness level (1 = very direct, 5 = very indirect) - Use of empathy statements? (Frequency) - Use of humor? (Never / Rarely / Occasionally) - Formality level (1 = very casual, 5 = very formal) 5. DEMONSTRATION EXAMPLES - Sample input: [provided or generated] - Tone-on target response: - Tone-off response (same content, wrong tone): 6. READY-TO-USE TONE PROMPT - A copy-paste prompt that embeds this tone calibration INPUTS: Abstract tone description: [E.G., "Professional but friendly, like a helpful colleague"] Sample content domain: [E.G., "Customer support email responses"] Audience: [E.G., "B2B software customers, technical"] Existing examples of correct tone (if any): [PASTE 2-3 EXAMPLES] RULES: - Avoid relying on abstract adjectives (use examples instead) - Demonstrate the tone, don't just describe it - Include anti-examples (what the tone is NOT) - Test the calibrated tone on sample inputs - Flag contradictions (e.g., "friendly but no contractions" is unusual)
How To Use It
- Run this once per brand voice or tone — then reuse the calibrated specification.
- Share the tone specification with your team to ensure consistency across all prompts.
- Use the demonstration examples as few-shot examples in your prompts.
- Test the calibrated tone on 10+ sample inputs before deploying.
- Re-calibrate annually — brand voices evolve.
Example Input
Abstract tone description:
“Warm and empathetic, but not overly emotional. Professional but not cold. Like a trusted advisor.”
Sample content domain:
“Customer support responses to frustrated users”
Audience:
“End users of a consumer app, non-technical”
Existing examples of correct tone (if any):
“I understand how frustrating that must be. Let me help you get this sorted out.”
Why It Works
Most tone instructions are adjectives — “friendly,” “professional,” “warm” — which mean different things to different people.
This framework improves outcomes by forcing:
- lexical specification (which words?)
- syntactic specification (how are sentences structured?)
- pragmatic specification (how direct? how much empathy?)
- demonstration examples (show, don’t just tell)
- anti-examples (what this tone is NOT)
Great tone calibration doesn’t describe — it demonstrates.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.