Prompt Writing Fundamentals

The quality of your prompt has a direct impact on how the AI sounds on the phone. Strong prompts create clear, confident behavior. Weak prompts create calls that feel generic, awkward, or inconsistent.

Prerequisites

You know the campaign goal and target audience.
You have listened to real calls or reviewed transcripts from your team.
You can describe what a successful call should sound like in plain language.

What a prompt should do

A good prompt should tell the AI:

what job it is doing
how it should open the call
what information matters most
how to handle common objections
when to stop, wrap up, or route to the next step

The prompt should not try to script every sentence. The AI needs direction, not a wall of instructions.

What good prompts look like

Good prompts are:

specific: They define the outcome clearly.
practical: They sound like instructions a strong manager would give a rep.
prioritized: They focus on the few things that matter most.
natural: They use language that would sound normal in a live call.

Steps

Start with the business outcome, such as booking a meeting or qualifying interest.
Write a short description of the audience so the AI knows who it is speaking to.
Define the opening in plain language and keep it concise.
Add the most important discovery points, not every possible question.
Write objection guidance for the top recurring pushbacks.
Add voicemail behavior separately so the AI does not use the same flow for both scenarios.
Include stop conditions that explain when the AI should end the conversation respectfully.
Test the prompt on a small sample and revise only one area at a time.

What each important prompt section means

Goal

This tells the AI what success looks like. If the goal is too broad, the call usually becomes too broad.

Opening guidance

This shapes the first impression. The opening should quickly answer who is calling and why the conversation matters.

Discovery guidance

This keeps qualification focused. Strong discovery instructions help the AI learn enough to move the call forward without sounding interrogative.

Objection handling

This tells the AI how to respond when the prospect hesitates, pushes back, or says no. It should be calm, concise, and bounded.

Voicemail behavior

This controls what happens when nobody picks up. It should be shorter and simpler than your live-answer flow.

Anti-patterns to avoid

Writing one prompt for too many audiences or use cases.
Packing the prompt with marketing copy instead of practical call guidance.
Giving the AI too many equal-priority instructions.
Writing openings that are long before they get to the point.
Using objection rules that sound defensive or argumentative.
Forgetting to tell the AI when to stop pushing.

Decision criteria

Rewrite the prompt if calls sound generic or the AI keeps missing the point.
Simplify the prompt if the AI sounds robotic or overloaded.
Split one prompt into multiple versions if different audiences need different openings or qualification logic.
Keep the current version if quality and outcomes are both improving.

Troubleshooting

The AI sounds too scripted

Replace rigid phrasing with simpler intent-based instructions. Tell the AI what to achieve, not the exact sentence it must repeat.

The AI sounds too vague

Add more specificity around the goal, audience, and first question. Vague calls usually come from vague prompts.

The prompt keeps getting longer

Move only the critical instructions into the main prompt. If everything is important, nothing stands out.

Final checklist

The prompt has one clear business outcome.
The opening is concise and believable.
Discovery instructions are focused.
Objection and voicemail behavior are handled separately.
Stop conditions are explicit.
The prompt has been tested on a small sample before scaling.