When a customer is angry, ChatGPT should not produce a long apology or invent a fix. The useful job is to separate the customer's emotion from the facts, answer what can be answered, and show when a person should take over.
Copy this prompt
Write a calm reply to an angry customer.
Customer message:
[paste message after removing private details]
What happened:
[known facts only]
What the customer wants:
[refund, fix, explanation, manager, cancellation, apology, unclear]
What we can do now:
[offer, next step, timeline, escalation, or "not sure"]
Policy or source:
[FAQ, order status, internal note, or "not available"]
What we must not say:
[fault not confirmed, no legal wording, no discount, no refund promise, no exact timeline]
Rules:
- Acknowledge the specific problem without arguing.
- Do not tell the customer to calm down.
- Do not blame the customer, staff, carrier, vendor, or system unless the facts clearly support it.
- Do not invent a solution.
- If a person should handle it, prepare a handoff summary instead of a final answer.
- Keep the reply under 130 words.
Return:
1. Main customer issue in one sentence
2. Safe reply
3. Firmer version if policy limits apply
4. What not to say
5. Whether this needs human handoff and why
6. Internal note for the next person
Example input
What happened: Customer was charged twice, but the second charge is still pending and may drop off automatically.
What the customer wants: Refund and apology
What we can do now: Explain pending charge, check again in 24 hours, escalate to billing if it posts
Policy or source: Billing note says pending authorizations usually drop within 1 to 3 business days
What we must not say: Do not promise the bank will remove it today
What good output should include
- A specific acknowledgment, not generic sympathy.
- The safe next step.
- No invented refund, credit, or timeline.
- A short internal note if the case needs escalation.
- Language that lowers tension without sounding fake.