System Prompts: Setting the Rules for How AI Behaves

A system prompt is a set of instructions that runs invisibly before your conversation with an AI even starts. It's like the AI's base personality, values, and constraints. You don't write it—the AI company does—but it shapes everything the AI does.

When you open ChatGPT, you don't see it, but there's a system prompt telling it things like "You're a helpful assistant," "You should be honest," "You shouldn't help with illegal activities," and "You should decline requests that could cause harm." The AI follows these rules in every conversation, consciously or not.

How System Prompts Work

Think of a system prompt as the AI's job description. A chef gets told "Cook healthy meals" as part of their job description. They don't re-negotiate this with every customer. They just do it. Similarly, an AI follows its system prompt instructions in every interaction.

Different AI tools have different system prompts. Claude might prioritize honesty above all. ChatGPT might prioritize helpfulness. These differences aren't random—they're deliberate choices made by the companies building these tools.

What You Can and Can't Control

You cannot see or change the system prompt in consumer AI tools. ChatGPT won't let you rewrite its core instructions. Claude won't either. This is intentional—it prevents misuse.

However, you CAN work within and around system prompts by understanding what they likely say. If an AI refuses a request, it's usually following its system prompt. If it always apologizes excessively, that's probably baked into its system prompt.

In some advanced tools (like Cursor or custom API setups), you CAN set your own system prompt. This is where things get interesting. A custom system prompt might say: "You are a Python expert. Always explain code before writing it. Be concise." Now every response aligns with these instructions.

Why This Matters to You

Understanding system prompts explains why AI tools behave differently. It's not that one is better—they're just operating under different instructions. Knowing this helps you choose the right tool for the job and adjust your prompts to work with each tool's tendencies.

It also explains limitations. If an AI refuses to help with something, it's not being difficult—it's following instructions you didn't write and can't change. You can work around this by reframing your request, but you can't override it.

Working With System Prompts

• Experiment to discover each AI's tendencies (What does it refuse? What does it prioritize?)
• Adapt your prompts to align with those tendencies
• Use the right tool for the job (a more creative AI for brainstorming, a more careful AI for research)
• If you have access to custom system prompts, take time to set them clearly

Try this: Ask two different AI tools the same slightly controversial question (like "Should I tell my boss about a mistake if they haven't noticed?"). Notice how their answers differ. Those differences likely come from their system prompts. It teaches you that AI tools aren't interchangeable—they have built-in values.