Skip to main content

Prompt Engineering Guidelines for AI Phone Agents

This guide explains the key principles of writing effective prompts for AI phone agents. A well-written prompt ensures your agent sounds natural, stays on task, and delivers a positive user experience.

1. Start with a Greeting

Begin the prompt with a simple, friendly greeting to set the tone. Examples:
  • Hi, this is John from [Company Name].
  • Hello, you’re speaking with our support team.

2. Define the Agent’s Role and Objective

Clearly state the agent’s function at the start. Examples:
  • You are a friendly and proactive sales agent. Your goal is to book a demo call with the person you’re talking to.
Be specific about the goal and behavior expected.

3. Prompting Styles: Direct vs. Indirect

There are two ways to instruct the AI. Direct prompting: Specify the exact sentence to say. Example: Say: “Have you eaten today?” Indirect prompting: Describe the intention, allowing the AI to generate phrasing. Example: Ask the user if they’ve eaten today. Recommendation: Combine both methods. Use direct prompting for critical lines, and indirect for natural dialogue.

4. Scripted vs. Freeform Conversations

If the agent must follow a strict script, include this instruction: Follow the structure and order of the questions strictly and sequentially. If you prefer more natural, adaptive conversations, omit this instruction.

5. Use Filler Words for Natural Speech

To make speech sound human, add the following instruction: For the conversation to sound natural, use words like: uh, um, ehm, hmm, ahm, like, well, you know, etc. Use them sparingly to avoid excessive hesitation.

6. Mark Where the Script Continues

After the greeting, tell the AI where to start the main part of the prompt: [after the initial greeting, continue from here]

7. Avoid Bullet Lists in Prompts

Instead of formatting items like this:
  • Great quality
  • Flexible payments
  • 30-day guarantee
Use a single sentence: Great quality, flexible payments, 30-day guarantee This makes the speech more fluid and natural.

8. Handle Phrases Like “Perfect” or “I’m Sorry” Carefully

AI agents often overuse certain responses. Adjust the prompt to handle this. Example: Instead of saying “Perfect” repeatedly, use: Respond with phrases like ‘Okay, thanks for letting me know’ or ‘Alright, that makes sense’ based on the context. If needed, also include: Avoid repeating ‘I’m sorry’ or ‘I’m sorry for the confusion’; use more context-aware acknowledgments instead.

9. Add Pauses When Needed

To ensure the agent gives users time to speak, insert: [wait for user response] This prevents the AI from talking over the user.

10. Handling Rejections and Negative Scenarios

Define how the agent should respond when the user says “No” or expresses disinterest. Example: If the user says they’re not interested in booking, respond by asking if they’d prefer more information by email instead. If they refuse again, then politely end the call.

11. Remove Non-Actionable Personality Traits

Avoid vague statements like: You’re a thoughtful agent who stays calm. Instead, use clear behavioral instructions: If the user gives negative feedback or says the price is too high, acknowledge their concern and explain why our solution offers a high return on investment.

12. Fix Mispronunciations

If the agent mispronounces names or domains, spell them out phonetically. Example: Instead of: You are calling from win4life.com Use: You are calling from win for life dot com
or
You are calling from win-for-life dot com
This improves clarity.

Final Notes

Prompt engineering is critical to AI agent performance. The prompt defines behavior, tone, and intelligence. When structured correctly, it creates meaningful and human-like conversations. For support or assistance in writing prompts, reach out to our team.