20 customer support prompts that solve it on the first reply
Customer support prompts for ChatGPT, Claude and Gemini that fix the reason tickets bounce back: the first reply guessed. Each one makes the model find what the customer actually asked, check your docs and policy, and answer in one go.
Swap anything in [brackets] for the ticket, your docs or your policy.
01. Find what the customer is actually asking
A long or confused ticket.
Here is a customer ticket: [paste the ticket] Tell me: 1. What they are asking for, in one sentence. 2. Every separate question or problem in it. 3. What they have already tried. 4. How frustrated they seem, from 1 to 5, and why. 5. What I need to check before I can answer. Do not write a reply yet.
03. Reply to an angry customer without groveling
A furious ticket, or one threatening to cancel.
An angry customer wrote: [paste the ticket] What actually happened on our side: [facts] What I can offer: [options and limits] Write a reply that acknowledges the specific problem in the first line, says plainly what went wrong if it was our fault, says what we are doing now, and gives a clear next step. No "we apologize for any inconvenience", no defensiveness, no promises I cannot keep.
Unlock the other 17 prompts for free
Sign up for the Whizi newsletter to read the rest of this pack.
- Handle a refund request by your actual policy
- Turn a vague bug report into one engineers can use
- Turn your best replies into a reusable macro
- Write a help article from the tickets you keep getting
- Write an escalation another team can act on
- Write the message for an outage or incident
- Check a reply's tone before you send it
- Reply to a cancellation honestly
- Find the trends in a week of tickets
- Build an internal FAQ for new support agents
- Answer in live chat in short messages
- Reply in the customer's language without losing meaning
- Reply to a feature request without overpromising
- Follow up on a bad satisfaction score
- Rewrite a policy so customers understand it
- Score a support conversation like a QA lead
- Have a second model check the first one's reply