1. Define the AI’s Identity First
Before you write any dialogue, define who the AI is.
This anchors tone, emotion, and phrasing across the call.
Include:
Name, gender, and voice style
Native language or accent
Role or function (for example, recruiter, tele-caller, customer support)
Personality traits (warm, empathetic, patient)
Communication style rules (formality, tone, pacing)
✅ Why it matters:
A clearly defined identity ensures consistent tone across all interactions, even if the script changes.
2. Create a Separate “Knowledge Context” Layer
Avoid mixing knowledge with behavior. Keep a structured block that gives factual grounding:
Job details
Salary and eligibility
Responsibilities
Company context
Goal:
The AI must never “hallucinate.” Keep all real-world data here so factual references are always reliable and traceable.
✅ Best Practice:
Use headers like ## Role Details, ## Eligibility, ## Salary Structure, etc., so the model can easily refer back.
3. Define a Conversation Goal in One Sentence
Every prompt should explicitly state:
“The goal of this call is to ___.”
This single line aligns the model’s reasoning with the business objective — ensuring every response moves toward that goal (e.g., qualifying the candidate).
✅ Why it matters:
Without a clear goal, the AI might drift into irrelevant chatter or redundant confirmation loops.
4. Lay Out the Conversation Flow and Structure
Always define the skeleton before writing the body.
Structure should look like this:
Introduction → Permission → Job Pitch → Interest Check → Eligibility Check → Close
Each step should:
Define entry condition (when to trigger this part)
Define goal (what info to extract or confirm)
Define exit condition (what moves the AI to next step)
✅ Why it matters:
You’re designing call logic, not just text — this helps the AI maintain flow even when the conversation diverges.
5. Anticipate Branching Logic and Exceptions
A strong prompt anticipates every possible fork in the conversation.
Include clear responses for:
Interest denial
Objection handling
Eligibility failure
Irrelevant questions
Silence / no response
Interruptions
✅ Tip:
Think of it like a tree — every possible user intent has a mapped next step.
No branch should lead to a dead end.
6. Set Explicit Behavioral Rules
Define exactly how the AI should behave in different conversational states.
Example directives:
“Never interrupt the candidate.”
“Use a 2–3 second pause after questions.”
“If silence for 6 seconds, check if user is still there.”
“If candidate declines twice, end politely.”
✅ Why it matters:
Think of it like a tree — every possible user intent has a mapped next step.
No branch should lead to a dead end.
7. Use Natural Speech Patterns
Voice AI ≠ Chatbot. It should sound human.
✅ Include:
Mild fillers (“uh”, “umm”, “you know”)
Soft confirmations (“okay”, “alright”, “I see”)
Small acknowledgements (“that’s great”, “got it”)
Warm openers and closers (“thanks for your time”, “have a wonderful day”)
❌ Avoid:
Perfectly structured grammar
Robotic repetition
Abrupt transitions
✅ Why it matters:
These subtle imperfections make speech feel authentic, not synthetic.
8. Bake in Empathy and Sales Warmth
Even for procedural calls, design the tone to feel like a friendly professional — not a script reader.
Include empathy cues like:
“I understand”
“That’s helpful, thank you for sharing”
“No worries, let’s go through this step quickly”
✅ Why it matters:
This maintains emotional safety and engagement, especially in job-related calls.
9. Include Strict Instructions as Guardrails
At the end of every prompt, add a section like ## Strict Instructions that acts as a safety net.
These should:
Prevent hallucinations
Restrict unwanted formatting or functions
Define termination conditions
Clarify escalation or fallback responses
✅ Why it matters:
It keeps the AI aligned under edge cases — ensuring stability under production stress.
10. End with a Graceful Closure Pattern
Define a universal, polite closing that:
Recaps what happened
Sets next expectations
Ends with warmth
Example:
“Thanks for your time. We’ll review your profile and get back to you soon. Have a wonderful day.”
✅ Why it matters:
Closures define user memory of the call — they make or break perceived professionalism.