Prompting Best Practices
A practical framework for building effective, human-like AI prompts
1. Think of Prompts as Modular (Not Scripts)
The best prompts are not rigid scripts—they’re flexible systems made of reusable parts.
Treat each section of a prompt as a building block:
- Tone strategies
- Objection handling
- Call flows
- Questions and responses
Mix and match these components depending on your use case.
Key idea:
You’re designing a conversation system, not writing a one-time script.
2. Use a Consistent Prompt Structure
Strong prompts follow a predictable framework. This keeps outputs consistent and scalable.
Recommended Structure
- Agent Information
- Objective
- Call/Conversation Goal
- Guiding Principles
- Behavior & Tone
- Introduction
- Immediate Objection Handling
- Discovery / Core Interaction
- Call-to-Action (CTA)
- Transfer or Next Step Rules
- Objection Handling Strategy
- Example Objections & Responses
- Knowledge Base (Optional)
Why it works:
- Ensures clarity
- Reduces randomness
- Improves performance over time
3. Define Clear Objectives (Don’t Skip This)
Every prompt must answer:
- What is the AI trying to accomplish?
- What does success look like?
Examples
- Book a meeting
- Transfer to a live agent
- Gather information
- Provide support and resolution
Best Practice:
Define both:
- Objective (what happens during the interaction)
- Goal (final outcome)
4. Control Tone and Personality
Tone is one of the biggest drivers of performance.
Define:
- Energy level (calm, upbeat, confident)
- Style (professional, witty, empathetic)
- Communication style (short, conversational, direct)
Guidelines
- Sound human, not scripted
- Keep responses short and natural
- Match the user’s tone
- Avoid repetition
Example Direction:
- “Speak like a helpful expert, not a script”
- “Keep responses under 2 sentences”
5. Keep It Short, Clear, and Conversational
High-performing prompts prioritize:
- Clarity over complexity
- Brevity over detail
- Flow over perfection
Rules
- Avoid long paragraphs
- Use simple language
- Limit responses to 1–3 sentences when possible
- Remove unnecessary filler
6. Handle Objections with a Proven Framework
Don’t guess how to respond to objections—use a structure.
Recommended Framework
SPARK → ACKNOWLEDGE → REFRAME → RESET
- SPARK: Keep momentum (“Totally fair…”)
- ACKNOWLEDGE: Validate their concern
- REFRAME: Offer a new perspective
- RESET: Guide back to the goal
Example
User: “I’m not interested.”
Response:
- SPARK: “Totally fair…”
- ACKNOWLEDGE: “I get that timing matters.”
- REFRAME: “Sometimes the best opportunities come unexpectedly.”
- RESET: “Would it hurt to take a quick look?”
Key Principle:
Objections = interest signals, not rejection.
7. Build Strong Call-to-Actions (CTAs)
Your prompt should clearly guide toward a next step.
Effective CTAs:
- Direct but conversational
- Timed appropriately (after interest)
- Easy to say “yes” to
Examples
- “Want me to connect you now?”
- “Should we take a quick look together?”
- “Would it help to speak with a specialist?”
Avoid
- Repeating the CTA too often
- Asking too early
- Being overly pushy
8. Use Discovery to Drive Engagement
Good prompts ask smart, simple questions to:
- Understand the user
- Keep them engaged
- Guide the conversation
Examples
- “What are you focused on right now?”
- “What’s been the biggest challenge?”
- “How are you handling this today?”
Best Practice:
- Ask one question at a time
- Keep it natural, not interrogative
9. Set Clear Guardrails
Define what the AI should NOT do.
Examples
- Do not make promises
- Do not invent information
- Do not transfer without confirmation
- Do not repeat the same phrasing
- Do not over-explain
Guardrails improve:
- Trust
- Accuracy
- Consistency
10. Optimize for Real Conversations (Not Perfect Ones)
Perfection kills performance.
Focus on:
- Natural flow
- Adaptability
- Responsiveness
Avoid:
- Overly scripted responses
- Robotic phrasing
- Trying to control every possible path
Key idea:
Flexibility beats perfection every time.
11. Improve Performance with Iteration
Great prompts are built over time.
Look for:
- Drop-off points
- Weak CTAs
- Repetitive responses
- Missed objections
Then:
- Adjust tone
- Simplify wording
- Add better examples
- Refine objection handling
12. Common Mistakes to Avoid
- Writing prompts like scripts instead of systems
- Overloading with too many instructions
- Ignoring tone and personality
- Weak or unclear CTA
- Giving up after first objection
- Long, robotic responses