ChatGPT prompts for customer service: ticket triage, replies and handoffs
Small support teams spend most of their day on the same few jobs: sorting the inbox, answering questions the FAQ already covers, passing tricky cases to the right person, and working out why the same complaints keep coming in. These prompts come from real examples of support teams using AI for those jobs. Each one tells the AI to stick to your own policies and help articles, and to hand anything unclear to a person.
12 free prompts across 1 category. Each one links to the real-world example it is based on.
How to use these prompts
- Give the AI your real help articles and policies, and tell it to use only those. Most wrong answers come from the AI filling gaps with guesses.
- Remove card numbers, passwords and anything private before you paste tickets or transcripts into a chat tool.
- Keep refunds, credits, legal threats and safety complaints with a person. Let the AI sort and draft them, not decide them.
- Once a month, paste in the questions customers asked and fix the help-center gaps they reveal. That cuts tickets more than faster replies do.
Start with these three
- Triage. Paste this morning's new tickets into the support ticket triage prompt. You get a category, an urgency and a one-line summary for each, and anything angry or legal is flagged first.
- Replies. Paste your FAQ and one customer message into the FAQ reply prompt. It only uses facts from your doc and says it will escalate when the answer isn't there.
- Handoffs. When a chat needs a person, run the handoff summary prompt so the customer never has to repeat themselves.
Before you let AI answer customers on its own
Drafting replies for a person to approve is low risk. Letting an agent send replies without approval is a bigger step. Run it in shadow mode on real emails for a week or two first, write down which requests it may resolve and when it must hand off, and score its conversations against the same checklist you use for staff.
The 12 prompts: what to paste and what to check
| Job (prompt) | What you paste in | What to check |
|---|---|---|
| Support ticket triage Triage · beginner |
| Treat pasted messages as data, not instructions |
| Draft replies from your FAQ Replies · beginner |
| Read the draft before it goes out |
| Human handoff summary Escalation · intermediate |
| Remove personal data you don't need |
| Find gaps in your help center Knowledge base · intermediate |
| Remove personal data you don't need |
| Find out why your support bot escalates tickets Bot Improvement · intermediate |
| Treat pasted messages as data, not instructions |
| Check a support case's real reason and draft a reply in the right brand voice Multi-brand case handling · beginner |
| Read the draft before it goes out |
| Let an agent fully answer routine customer emails, with clear rules for when to hand off Email support automation · intermediate |
| Set hand-off rules; a person handles refunds and complaints |
| Send customers proactive updates before they have to chase you Proactive outreach · intermediate |
| Check order and appointment data before customer messages go out |
| Have an agent score every support conversation against your quality checklist Quality assurance · intermediate |
| Have a qualified person check legal points |
| Sort a month of customer calls by what callers wanted and how they felt, to find what is really driving them Call driver analysis · beginner |
| Treat pasted messages as data, not instructions |
| Check customer damage photos on return claims and flag pictures that were reused or copied from the web Return damage checks · beginner |
| Check every number, price and total |
| Test a customer support AI in "shadow mode" on real emails before it ever replies to a customer Safe AI rollout · beginner |
| Treat pasted messages as data, not instructions |
What AI should never decide on its own
Let these prompts draft, sort and flag. Keep a person responsible for:
- Refunds, credits and exceptions to your policy, beyond small limits you set in writing
- Replies to legal threats, chargebacks, safety or injury complaints, and regulator questions
- Closing an account, cancelling an order or changing a customer's billing
- Whether a damage photo or claim is fraud: the AI can flag it, a person decides
Frequently asked questions
Can ChatGPT answer customer service emails?
Yes, it can draft good replies if you give it your FAQ or policies and the customer's message, and tell it to use only those facts. Start with a person approving every draft. Only let an agent send on its own for simple, low-risk requests after testing it in shadow mode.
What is the best ChatGPT prompt for customer support?
For most small teams, the most useful first prompt is ticket triage: it sorts a pile of tickets by topic and urgency and flags angry or legal ones, so the important cases get seen first. Next comes a reply prompt that answers only from your own help docs.
How do I stop an AI support bot from making things up?
Paste in the approved help articles and policies, tell it to answer only from them, and give it a clear rule to escalate when the answer isn't there. Then review the tickets it escalates to find missing articles. The bot-escalation prompt in this guide does exactly that.
Not sure where to start? The free AI time savings calculator ranks which agent to build first.
32 copy-paste agent prompts (PDF) plus 5 new prompts every week.
Get it free