How to write a good customer service system prompt
Create a concise operational prompt that guides behavior without trying to replace missing knowledge.
A system prompt should define behavior, not smuggle in an entire knowledge base. Put changing facts in sources and use the prompt for role, tone, evidence rules, boundaries and escalation.
The strongest prompt is short enough to audit and specific enough to test. Every important instruction needs a matching evaluation question.
What you will have at the end
- A structured customer service prompt
- Explicit missing-evidence and escalation behavior
- A measured A/B result on fixed questions
Before you start
- Defined support scope and tone
- Escalation contact or fallback process
- A small prompt evaluation set
Prompt for behavior; source facts separately
The role says who the assistant represents, scope says what it may answer, evidence rules govern uncertainty, style sets response form and escalation defines a safe exit.
Examples should illustrate edge cases without hard-coding many volatile details.
01–05
Set it up step by step
Open the Custom Role editor
Work in the intended chatbot configuration.
Open Configuration → Custom Role/System Prompt and save the existing text as a versioned baseline before editing.
Write role, tone and scope
State identity and permitted work in plain language.
Define company, assistant purpose, target audience, tone and supported topics in short sections. Avoid vague requests such as “always be helpful” without operational detail.
Add evidence, boundaries and escalation
Specify what happens when knowledge is missing.
Tell the bot not to invent policies, prices or account data, to acknowledge uncertainty and to escalate explicit requests, disputes or sensitive cases through the configured process.
Set answer length, format and examples
Make the desired response observable.
Define a default length, when lists are useful and how to ask one clarifying question. Add two brief examples: a normal grounded answer and a safe missing-information response.
A/B-test in Knowledge Optimizer
Compare the old and new prompt on identical questions.
Run normal, ambiguous, out-of-scope and escalation questions with both prompt versions. Score source fidelity, tone, brevity and safe fallback; publish only after regression questions remain stable.
Example & result
See the practical test and its result
Every tutorial includes a fixed input, the expected outcome and a transparent record of what was actually verified locally.
Practical example: How to write a good customer service system prompt
This exact scenario was completed with the temporary tutorial account.
Exact test input
What is Northstar Services’ refund policy for annual contracts?
Expected result
The assistant says the policy is not present instead of inventing terms.
What was actually verified
After the prompt was saved, the live assistant said it could not find a refund policy in the indexed documents and did not invent terms.
Tips & tricks
Make the setup reliable
Test with realistic examples, record your baseline and change one setting at a time. That makes real improvements visible.
Use headings inside the prompt
Role, scope, evidence, style and escalation sections are easier to audit and update than one long paragraph.
Avoid conflicting absolutes
“Always answer” conflicts with “never guess.” State which rule wins when evidence is missing.
When something does not work
Troubleshooting
Check status, permissions and test data systematically before changing the model or prompt.
The expected option is missing
Confirm the account plan, feature permissions and selected chatbot. Paid or beta features can be hidden when prerequisites are not met.
The test result is inconsistent
Reset the test conversation, keep the input identical and change one setting at a time so the cause remains measurable.
Ready for a production-style test
Version the approved prompt with date, owner and evaluation score. Re-test whenever escalation tools, policy scope or target audience changes.