How to write AI prompts for automations that actually work?

You know how to chat with ChatGPT. That’s not what this is about.

When AI runs inside an automation, nobody reads the output. A machine does. The next module in your workflow takes that output and processes it, sends it, saves it, or feeds it into another step. That changes everything about how you write prompts.

In a chat, if the answer isn’t great, you ask again. In an automation, the prompt runs once, unattended, and the output has to be right the first time. There’s no “try again.” There’s no human checking.

In this tutorial, we’ll cover:

• The difference between chat prompts and automation prompts
• System prompt vs user prompt: what goes where and why
• The 5 rules that work: define the role, specify the format, say what NOT to do, give examples, one task per prompt
• Format-specific tips: getting clean HTML, valid JSON, plain text, or decision outputs
• The most common failures: markdown wrapping, chatty preambles, inconsistent output length
• Temperature settings for automation: when to use 0.3 vs 0.7 vs 1.0
• Real examples from actual workflows
  

Full guide coming soon. The key insight: an automation prompt is not creative writing. It’s an instruction set for a machine, and it needs to work reliably every single time.