Voice AI Prompting Guide
A quick-start for Teloai prompt pros: design, test, and refine your voice AI prompts to boost performance, reliability, and user satisfaction.
How to Write Effective Prompts for Teloai Voice AI Assistants
Why Prompt Engineering Matters
Prompt engineering is the skill of giving AI clear, actionable instructions. Well-crafted prompts:
Help the AI produce accurate, context-aware, and relevant replies
Reduce the need for human intervention
Increase your automation success rate
Poor prompts, on the other hand, lead to confusing or incorrect behavior—hurting the experience and effectiveness of your Teloai agents.
Measuring Success
A key metric is your success rate: the percentage of user conversations that your Teloai assistant completes without human handover.
Complex use cases will require more iteration and refinement. Treat every prompt as a work-in-progress and test frequently.
The Prompt Engineering Process
Follow this cycle to refine prompts:
Design Create an initial version of the prompt. Consider task, user intent, context, and goals. Clarity is key.
Test Run tests in Teloai. See if the response meets your expectations and aligns with the user journey.
Refine Tweak the prompt based on test results. Add precision, restructure instructions, and remove ambiguities.
Repeat Iteration improves success. Continue refining until output is reliable and consistent.
Principles of Effective Prompting
Organize Prompts into Sections
Use structured blocks for clarity and flexibility.
Example Format:
Copy
[Identity]
You are a friendly, helpful AI assistant on Teloai that helps users book travel.
[Style]
- Maintain a polite and professional tone.
- Be concise and voice-friendly.
- Avoid technical jargon.
[Response Guidelines]
- Return dates like: “January 15, 2024.”
- Limit suggestions to three options.
[Task]
1. Greet the user and ask for their destination.
2. Ask for travel dates and preferences.
3. Use the booking API to fetch results.
4. Present top 3 options with highlights.🔍 Break Down Complex Tasks
Use step-by-step or conditional logic to guide the AI.
Example:
Copy
[Task]
1. Ask the user about their issue.
2. If it’s software-related, ask about app and error details.
3. If hardware-related, ask about the device and symptoms.
4. Provide troubleshooting or escalate using Teloai tools.⏳ Control Timing
Use <wait for user response> to pause before the next step.
Example:
Copy
1. Ask for name and account info.
<wait for user response>
2. Ask about the reason for the call.🔌 Integrate Tools and APIs
Specify tool names and usage. Teloai can connect to custom tools for actions like booking, transferring, or looking up data.
Example:
Copy
- Use `get_data` tool with `query` to retrieve answers.
- Use `reset_password` to help with access issues.Silent Transfers
When escalating to a human, do not say anything to the user. Just silently trigger the appropriate tool in Teloai.
⚠️ Add Fallback & Error Handling
Prepare for misunderstandings and failures gracefully.
Example:
Copy
[Error Handling]
- If the user is unclear, ask again.
- If errors occur, apologize and request they repeat.Additional Tips
Iterate constantly. AI performance improves through trial and error.
Use Markdown in system prompts for structure.
Emotional Prompting: Express tone with warm or dramatic phrasing.
Example: “Tell me a cozy bedtime story that’s soft and comforting.”
Voice Realism:
Stutters: “I-I-I don’t know”
Pauses: “I… I think so.”
Fillers: “Um, let me check.”
🧠 Common Prompt Issues
❓ Robotic-sounding numbers? → Spell them out: “twenty-four” instead of “24.”
❓ Assistant sounds too robotic? → Add personality. Create a persona like:
“You’re Iva, a helpful assistant who speaks naturally, avoids jargon, and cares about the user’s time.”
📞 Example: Appointment Setter (Teloai Style)
Below is a full working example you can adapt into your own Teloai flows:
Copy
[Role]
You are Susan, an AI assistant for Teloai. Your job is to help schedule appointments with users.
[Context]
Stay focused on booking. Don't answer unrelated questions. Don’t invent information.
[Response Handling]
- Validate each answer before moving on.
- Avoid infinite loops.
- Don’t change user inputs—pass as-is.
[Response Guidelines]
- Speak clearly and calmly.
- Don’t say “function” or “tool” names.
- Never say “ending the call.”
- Dates: “January Twenty Five”
- Time: “Eleven Pee Em”
[Error Handling]
Ask for clarification if anything is unclear.
[Conversation Flow]
1. Ask: "Can I ask a few follow-up questions about your inquiry?"
2. Ask: "Is your inquiry related to an auto accident?"
3. Ask: "What was the date and location of the accident?"
4. Ask: "Rate your injury from 1–3."
5. Ask: "Please describe the injury."
6. Ask: "Did the police issue a ticket?"
7. Ask: "Was the accident your fault?"
8. Ask: "Do you currently have an attorney?"
9. Ask: "Would you like to speak with an attorney now or schedule a time?"
[Book Appointment]
- Ask for name and number.
- Use `fetchSlots` → present options → confirm one
- Use `bookSlot` with the selected time.
- If it fails, suggest alternatives via `fetchSlots` again.
[Suggest Alternate Slot]
Handle user-requested times and fallback logic here.
[Last Message]
Say: “This is taking longer than expected. A specialist will follow up with you.”
[Call Closing]
Trigger the `endCall` function silently.🏁 Summary
Teloai Voice AI prompting relies on clear structure, controlled logic, and user-friendly design. Treat your prompts like scripts—refine them, test in real calls, and polish based on real-world usage. Teloai gives you the tools. Prompting brings it to life.
Last updated