You just invested in an AI voice agent for your business. The technology is set up, the phone number is live, and yet… calls keep going sideways. Customers hang up, appointments don't get booked, and you're starting to question the whole investment.
The problem is almost never the technology. It's the script.
An AI voice agent — whether it runs on ElevenLabs, VAPI, or Retell — is only as good as the instructions you give it. According to McKinsey research, companies that optimize their AI agent scripts see a 14% increase in issue resolution per hour and a 9% reduction in handling time. The difference between a voice agent that frustrates your customers and one that impresses them often comes down to a few well-written paragraphs.
Here's the 5-step guide our team at Agent IA Vocal uses to configure agents that convert — and that you can apply starting today.
Step 1: Define Your Agent's Identity and Personality
Before writing a single line of script, ask yourself a fundamental question: who is your voice agent? Not just its name (though that matters too), but its personality, tone, and boundaries.
A voice agent for a dental clinic doesn't talk like one for an auto repair shop. The vocabulary, pace, and even sentence length should reflect your clientele. For a clinic, you want a reassuring, professional tone. For a home renovation contractor, a direct and warm approach works better.
In practical terms, your identity script should include:
- The specific role: "You are the virtual assistant for Smile Clinic, specializing in appointment booking and answering common questions about our dental services."
- Tone and style: "Adopt a professional but warm tone. Use natural, conversational language. Keep sentences short and clear."
- Clear boundaries: "Never provide medical diagnoses. If the patient describes an emergency, immediately transfer to the emergency line."
This is something we explore in depth in our article on quality control for AI voice agents: without a well-defined identity, there's no way to measure whether your agent is performing correctly.
Step 2: Structure the Conversational Flow
The most common mistake we see from SMBs configuring their voice agent? A linear script that only accounts for one scenario. But your customers don't all call for the same reason.
A good voice script works like a decision tree, not a monologue. Each branch corresponds to a caller intent:
- Appointment booking: availability check, confirmation, reminder setup
- Information request: hours, pricing, services offered
- Complaint or issue: active listening, human transfer when needed
- Off-topic call: polite redirect to the right contact
For each branch, plan an opening phrase, 2-3 qualifying questions, and a clear conclusion. The VAPI prompting guide recommends keeping each agent response under 10-15 seconds of spoken time — roughly 30-45 words in English. Beyond that, the caller mentally checks out.
Here's a concrete example for a hair salon:
Agent: "Hi there! Thanks for calling Elegance Salon. I'm your virtual assistant. Are you calling to book an appointment, change an existing one, or do you have another question?"
Customer: "I want to book an appointment."
Agent: "Great! What service are you looking for — a haircut, coloring, or another treatment?"
Short, direct, and each response naturally guides toward the next step.
Step 3: Build in Safeguards and Transparency
This is where most online guides fall flat. They hand you a script template but forget two critical elements: legal compliance and edge case management.
Transparency isn't optional. Your voice agent should identify itself as artificial intelligence at the start of the call. It's not just a best practice — it's a matter of trust. According to JustCall's research, 88% of people still prefer speaking with a human. If your agent pretends to be human and the customer figures it out, you lose that trust for good.
Add these safeguards to your script:
- AI disclosure: "I'm the virtual assistant for [Business]. I'm an AI-powered agent."
- Human escalation: "If you'd prefer to speak with a team member, just say 'talk to someone' at any time."
- Error handling: after 2 misunderstandings, the agent should offer a transfer rather than repeating the same question in a loop.
- Knowledge limits: "I don't have that information right now. I can take your number and have our team call you back."
These safeguards aren't optional extras. They're what separate a professional voice agent from a gimmick that annoys your customers.
Step 4: Optimize for Your Local Market
If your customers speak a specific dialect or regional variant — like Quebec French in Canada — this step is absolutely critical. Most voice AI platforms are built for standard American English first, and everything else is an afterthought.
The result? An agent that sounds foreign to your own customer base. Regional expressions, local terms for common services, and natural conversational rhythm all matter. A customer in Montreal expects a different experience than one in Paris or New York.
Our recommendations for local optimization:
- Use regional terminology that your customers actually use in conversation
- Test pronunciation of local place names — some TTS engines butcher them
- Avoid overly formal or literary constructions that sound unnatural in spoken form
- Read every script line out loud — if it sounds like a legal document, rewrite it
- If serving bilingual customers, plan clear language-switching protocols
Step 5: Test, Measure, Iterate
Your first script won't be perfect. And that's fine. The advantage of an AI voice agent over a human receptionist is precisely that you can change its behavior in minutes.
The three-phase improvement process:
1. Test before you deploy. Run at least 10 test calls covering your main scenarios. Note every moment the agent hesitates, gives a wrong answer, or sounds robotic. Our guide on pre-deployment testing for AI voice agents details the 7 essential checkpoints.
2. Track the right metrics. After launch, monitor three key indicators:
- Completion rate: how many calls achieve their goal (appointment booked, info delivered)?
- Human transfer rate: above 30%, your script has gaps.
- Average call duration: too short means people are hanging up. Too long means the agent is going in circles.
3. Iterate with small adjustments. Change one thing at a time. If you modify the greeting, the decision tree, and the tone all at once, you'll never know what worked. The best SMBs we work with review their script every two weeks during the first month, then monthly after that.
The Ready-to-Use Script Template
To give you a head start, here's the base structure our team at Agent IA Vocal uses for every deployment. Adapt it to your industry:
Section 1 — Identity
"You are [Name], the virtual assistant for [Business]. You are an AI agent. Your role is [primary role]. Your tone is [professional/warm/direct]. You speak naturally and conversationally."
Section 2 — Greeting
"Hi! Thanks for calling [Business]. I'm [Name], your virtual assistant. How can I help you today?"
Section 3 — Qualification
"[Question 1 based on detected intent]." → Branch A, B, or C.
Section 4 — Safeguards
"If the caller asks X → respond with Y. If misunderstanding after 2 attempts → offer human transfer. If emergency → transfer immediately."
Section 5 — Closing
"All set! [Summary of action taken]. Is there anything else I can help with? Have a great day!"
This template covers 90% of SMB use cases. The official OpenAI guide on voice agents offers similar structures, but aimed at developers. Our version is designed for business owners who want concrete results without touching code.
The Bottom Line
Your AI voice agent's script isn't a technical detail to push off until later. It's the heart of the customer experience. A well-scripted agent books appointments, answers questions, and frees up your team. A poorly scripted one drives customers away and damages your reputation.
The 5 steps — identity, conversational flow, safeguards, local optimization, and iteration — don't take weeks. With a focused afternoon of work, you can go from a mediocre voice agent to a genuine asset for your business.
And if you'd rather have a specialized team handle it, that's exactly what Agent IA Vocal does. We configure, test, and optimize — you just answer the customers who've already booked their appointments.