Skip to main content
Prompts define how your AI agent behaves and communicates. Kakiyo uses a two-layer prompt system that gives you complete control over conversation style and strategy.

Two-Layer Prompt System

Context Prompt

Defines your AI agent’s behavior, personality, and conversation strategy. Includes:
  • Agent’s role and personality
  • Conversation goals and objectives
  • Tone and communication style
  • Limits and boundaries
  • Fallback behaviors and responses

First Message Prompt

Defines how conversations are opened after connection acceptance. Includes:
  • Opening message structure
  • Value proposition delivery
  • Personalization approach
  • Call-to-action strategy

Variable System

Variables allow you to reuse prompts across multiple campaigns and agents with different parameters.

Required Variables

Must be configured for every campaign:
  • Prospect Description: All prospect information for personalization
  • Offering Description: Knowledge about what you’re selling
Important: you don’t need to enter a value for those variables, just place them somewhere in one of your 2 prompts and let the AI Agent fill them for you.

Optional Variables

Pre-defined variables for common use cases:
  • Agent Name: Name the AI should use in conversations
  • Company: Your company name for the campaign
  • Prospect Name: How to address prospects

Custom Variables

Create your own variables for specific needs:
  • Goal: Different objectives for different campaigns
  • Industry: Specific messaging for different sectors
  • Custom messaging: Any other variable content you need
Example usage: You want to use the same prompt in 2 different campaigns. Instead of writing “Your objective is to book a meeting” in the prompt, use “Your objective is to ” and set Goal = “book a meeting” in Campaign 1 and Goal = “visit my website” in Campaign 2.

AI Model Selection

Kakiyo offers access to all major AI models from leading providers: Available Models:
  • Anthropic Models: All Claude models (Sonnet and Opus)
  • OpenAI Models: All GPT models (ChatGPT series)
  • Grok Models: All X AI models
  • Google Models: All Gemini models
Recommendation: As of February 2026, Claude Sonnet 4.5 (Anthropic) delivers excellent conversation quality for outreach use cases. Start there, then optimize based on Sandbox and live campaign metrics.
The best-performing model changes over time. Re-test your default model regularly and adjust based on actual outcomes rather than assumptions.

Writing Effective Prompts

Best Practices

  • Define clear goals: Specify exactly what you want the AI to achieve
  • Set boundaries: Define what the AI should and shouldn’t do
  • Match your tone: Write in the style you want the AI to use
  • Include fallback instructions: Tell the AI how to handle unexpected situations
  • Use proper formatting: Structure prompts clearly with sections

Testing and Optimization

  • Use the Sandbox: Test prompt effectiveness before launching campaigns
  • Iterate based on results: Refine prompts based on conversation performance
  • Monitor conversations: Review actual AI responses and adjust accordingly

Starting Point

Kakiyo provides a default prompt template to help you launch quickly. Use it as a baseline, then iterate based on your own audience and campaign results.
Important: Reply performance depends on targeting quality, profile quality, and prompt quality. Validate improvements with controlled tests instead of assumptions.

Well-crafted prompts are essential for successful AI conversations. Start with the default template and refine based on your specific audience and goals.
Last modified on August 12, 2026