AI Prompting Guide: Best Practices for Assistantz.ai

Most AI assistants sound robotic or too salesy because they’re given too little context about how to actually speak. This guide breaks down a prompting framework that makes an assistant feel like a real extension of your brand.

The wrong way to write a prompt

A prompt like “You are an AI assistant that helps book calls and sell my services. Convince people to sign up and get them to book a call” fails for several reasons:

• Too pushy. It sounds like a robotic telemarketer, which makes people feel pressured.
• No rapport-building. It skips the human element and asks for a commitment before earning trust.
• No lead qualification. The AI has no way to understand what the person actually needs.
• No brand voice. Responses come out generic and lifeless.
• No flexibility. Every conversation gets treated as a sales pitch, even off-topic questions.

The ISRT framework

Structure your prompt around four sections, in this order:

1. Identity — who is the AI?

Give it a name (optional, but it humanizes the assistant), a role and area of expertise, who it talks to, and why people are engaging with it.

Example: “You are Sara, a professional, engaging, and witty AI assistant for [Business]. Your role is to interact with leads and customers who tested a demo, responded to an offer, or downloaded a resource. Your job is to engage them naturally, pre-qualify them, and guide them to the next step while making them feel heard.”

2. Style guardrails — how should it speak?

This is the tone and humor control, and it’s the part most prompts skip. Aim for empathetic, conversational, warm, proactive, and lightly witty when appropriate — always staying professional.

Example: “Maintain a friendly, professional tone with a mix of wit and warmth. Show empathy if a user is frustrated — acknowledge their concern before guiding them to a solution. Use humor naturally, never forced.”

Phrases that soften responses: “I totally get that, [Name]…”, “That makes complete sense.”, “That’s a great question, happy to clarify!”

3. Response guidelines — how should it reply?

Keep replies clear and concise, conversational (not abrupt), and action-oriented. Break down complex topics with bullet points when needed.

Instead of: “Yes, our White Label AI is available for GHL users.” Use: “Absolutely! Our White Label AI is built for GHL users. Want a quick breakdown of how it works?”

4. Tasks — what should it actually do?

Give it a step-by-step engagement flow: greet warmly, determine what the person needs, provide answers that build trust, ask qualifying questions, and guide qualified leads to booking. Handle objections with reassurance rather than pushing past them.

Words and phrases that shape personality

• Empathy: “I totally get that, [Name].” / “I’d feel the same way in your position.”
• Conversational: “Hey [Name], great to chat with you!” / “Let me break it down real quick.”
• Light humor (use sparingly, stay on-brand): self-aware AI jokes work better than forced one-liners — keep it relevant to your audience.
• Proactive: “Want me to show you how this works?” / “I’ve got a step-by-step guide for you, want me to send it over?”

Important reminders

• Always collect and confirm contact details (name, email, phone) before booking anything, and spell back the email and phone number to check accuracy.
• Stay neutral if asked about competitors.
• Clarify that it’s an AI assistant if the user asks directly.
• Get consent before using any voice recordings for training or marketing.

Next step

Once your prompt is structured, connect it to real tools and data: how to create a tool to add and remove tags and how to create a custom tool to extract data and fill custom fields. If bookings or responses seem off after a prompt change, see fix: AI assistant not booking appointments.