When you first launch a new agent on Synap, it arrives like a brilliant, freshly hired university graduate. It is incredibly smart, it has read all your documents, and it is eager to please.
But it has one major problem: It doesn’t know who it is.
Unless you tell it otherwise, it will just act like a generic AI assistant. It might be too chatty, too formal, or it might try to answer questions about the weather when it should be selling your software.
This is where the System Prompt comes in.
Think of the System Prompt as the "God Mode" instruction. It is a set of invisible instructions that sits above every conversation, directing the agent's behaviour, tone, and rules.
If you want to move beyond a basic chatbot and build a powerful business tool, you need to master the art of the System Prompt. Here is the framework we use.
The 4-Step Formula: R.C.G.G.
You don’t need to be a coder to write a great prompt. You just need to be clear. We recommend breaking your instruction down into four parts: Role, Context, Goal, and Guardrails.
1. Role (Who am I?)
First, tell the AI exactly who it is. This sets the baseline for its intelligence.
-
Weak: "You are a helpful assistant."
-
Strong: "You are a senior Customer Success Manager for a boutique architectural firm in Melbourne."
2. Context (Who am I talking to?)
The AI needs to know its audience so it can adjust its vocabulary.
-
Weak: "Answer the user."
-
Strong: "You are speaking to non-technical small business owners who are often stressed and time-poor. Avoid jargon."
3. Goal (What is a 'win'?)
What constitutes a successful conversation? Is it answering a question, or getting a sale?
-
Weak: "Help them."
-
Strong: "Your primary goal is to troubleshoot their technical issue within 3 replies. If you cannot solve it, instruct them to email support@synap.au."
4. Guardrails (What must I NOT do?)
This is the most important section for business safety. You must tell the agent what is off-limits.
-
Example: "Do not mention competitor products. If asked about pricing not in your documents, do not make up a number—state that you will have a human follow up."
Putting it all together: A Real-World Example
Let’s say you run a plumbing business. Here is how you would combine those four steps into one killer prompt to paste into your Synap dashboard:
[Role] You are a friendly, knowledgeable receptionist for 'Bayside Plumbing'.
[Context] You are chatting with local homeowners who usually have an emergency (leaks, blocked drains). They are likely anxious.
[Goal] Your goal is to identify their problem and get their name and phone number so a plumber can call them back immediately.
[Guardrails] Keep your answers short. Do not attempt to give DIY plumbing advice or explain how to fix the leak themselves. Do not promise specific arrival times (e.g., "We will be there in 10 minutes").
Why "Negative Constraints" are vital
Notice in the example above, we told the agent what not to do ("Do not attempt to give DIY advice").
AI agents generally want to be helpful. If a user asks "How do I fix a burst pipe?", a standard AI will try to explain the steps. For a plumbing business, this is bad—you want to send a plumber, not teach the customer to do it themselves!
By explicitly adding a Negative Constraint (Guardrail), you force the agent to say: "I can't advise on DIY fixes for safety reasons, but I can have a plumber call you in 5 minutes to come take a look."
Iteration is Key
You rarely get the prompt perfect on the first try.
-
Write your prompt in Synap.
-
Open the "Test Chat" window.
-
Try to trick the agent. Ask it things it shouldn't know.
-
If it fails, go back to the prompt and add a new Guardrail.
Ready to take control? Log in to your dashboard, open your Agent Settings, and try rewriting your System Instruction using the R.C.G.G. formula.