The Prompt Stack and Hierarchies

1.1 System Prompts, User Prompts, and Assistant Messages

Understanding how instructions layer is fundamental to mastering intermediate AI. Every interaction involves multiple instruction layers that can reinforce or contradict each other.

Why it matters: When all three layers align, your instructions work seamlessly. When they conflict, the system defaults to its built-in hierarchy—typically: system prompts override everything, then context/history, then user prompts. Understanding this hierarchy prevents wasted effort fighting unwinnable battles.

1.2 Developer-Set Instructions and Platform Defaults

Platforms impose developer-set instructions that are largely invisible to users but profoundly shape what the AI will and won't do. Every AI platform comes with baseline instructions baked in by its developers. These cover safety, compliance, tone, and core capabilities. When you write your own prompt, you're not starting from a blank slate—you're layering your request onto that existing foundation.

For example, an AI designed for customer service might have developer-set instructions that automatically soften negative language. If you try to write a harsh rejection email, the AI may automatically reframe it more diplomatically—not because your prompt failed, but because the platform's rules prevent harsh outputs.

Understanding your platform's defaults is not about circumventing them (usually impossible); it's about knowing what guardrails exist so you can design prompts that work within them rather than against them. Some platforms publish their system prompts; others keep them private. When they're published, studying them reveals strategic insight into how the system will behave.