40 Best ChatGPT Prompts for Customer Service (2026)
40 of the best ChatGPT prompts for customer service, ready to copy: de-escalation, refunds, delays, replies, macros, ticket summaries, and feedback.
Researched with AI assistance, reviewed and edited by Tapabrata Biswas.
In this article
- 01A good prompt beats "reply to this complaint"
- 02Calm an upset customer
- 03Refunds, returns, and billing
- 04Orders, delays, and outages
- 05Everyday replies (email and chat)
- 06Saying no and handling tricky asks
- 07Templates and macros
- 08Work the queue faster (the agent-side prompts)
- 09Learn from customers (feedback and improvement)
- 10Draft with AI, keep a human on anything sensitive
- 11What this post does not cover
- 12Sources
79 percent of support agents say an AI assistant makes them better at their job, according to Zendesk's research. The ones who feel that way aren't letting it answer customers for them. They're using it to draft the boring half of a reply in seconds, so they can spend their attention on the part that actually needs a person: the judgement, the tone, the call on what to do.
This is a collection of 40 ready-to-use ChatGPT prompts for customer service, grouped by the job: calming an upset customer, refunds and billing, delays and outages, everyday replies, saying no, building macros, working the queue faster, and learning from feedback. Grab the one you need, paste the customer's message and your details into the brackets, and shape the result. If you want to get good at writing your own, our guide to writing prompts walks through it; below, the drafts are already done.
ChatGPT is free to start, running GPT-5.5 by default as of June 2026, and every prompt here works just as well in Claude or Gemini, so use whichever tool your team already has open.
A good prompt beats "reply to this complaint"
A good support prompt gives ChatGPT three things it can't guess: the customer's actual message, the policy or facts that apply, and the voice your team uses. Ask vaguely and you get a generic, slightly robotic reply:
Reply to this customer complaint.
Give it the context and a clear job, and you get a draft you can almost send:
A customer wrote: 'My order is two weeks late and nobody has replied to my emails.' Write a calm, empathetic reply that apologises for the delay and the silence, gives a realistic next step, and offers to expedite the order. Warm and human, not corporate. Don't promise a refund.
The more of that real context you paste, the less the reply sounds like it came from a stranger, which in support is the whole game. That's how all 40 below are written, each with a [bracket] or two where your specifics go, so you're always shaping a draft rather than starting from a blank box. One rule runs through all of them, and it's the last section of this page: draft with ChatGPT, but keep a person on anything sensitive before it reaches a customer.
Calm an upset customer
The replies that decide whether someone stays or leaves. The goal is to lower the temperature and move toward a fix. When someone's angry, the instinct is to explain or defend; the prompts here do the opposite, leading with acknowledgement so the customer feels heard before you get to the solution.
1. Respond to an angry email
Acknowledge first, solve second.
A customer sent this angry message: [paste]. Write a calm, empathetic reply that acknowledges their frustration first, takes responsibility where it's fair, and gives one clear next step. Don't be defensive, and don't over-apologise. Our tone is warm and professional.
2. De-escalate a live chat
Short lines that cool things down.
I'm in a live chat with a frustrated customer about [issue]. Give me three short, calm replies I can use to lower the temperature and move toward a solution, focused on what we can do rather than what went wrong.
3. Handle a repeat complaint
Own the earlier miss without excuses.
A customer is contacting us again about [issue] that wasn't fixed last time. Write a reply that owns the earlier miss without making excuses, explains what we're doing differently now, and sets a realistic expectation. Details: [paste].
4. Reply to a public complaint
Brief, genuine, and moved to private.
A customer left this public complaint on [platform]: [paste]. Write a brief, genuine public reply that acknowledges them, avoids sounding scripted, and moves the conversation to a private channel to resolve it.
5. Hold a boundary calmly
Stay respectful when they aren't.
A customer is being abusive in [channel]. Help me write a firm but polite message that asks them to keep it respectful so we can help, without escalating things further, and notes what happens if it continues. Context: [paste].
Refunds, returns, and billing
The money conversations, where the exact wording matters most. A clumsy refund reply can turn a small problem into a lost customer or a chargeback, so these aim to be clear, warm, and unambiguous about what happens next.
6. Approve a refund warmly
Say yes in a way that keeps them.
Write a friendly reply approving a refund for [reason]. Confirm the amount and timing, apologise briefly for the trouble, and end on a note that keeps them as a customer. Details: [paste].
7. Decline a refund politely
Say no without going cold.
A customer is requesting a refund that falls outside our policy: [paste the situation and the policy]. Write a kind, clear reply that says no without sounding cold, explains why in one line, and offers an alternative we can actually do.
8. Offer a goodwill gesture
Make it feel genuine, not transactional.
A customer had a bad experience with [issue] but isn't owed a refund. Help me offer a goodwill gesture (a discount, a credit, or [option]) in a way that feels genuine rather than like we're buying their silence. Details: [paste].
9. Explain a billing charge
Clear up confusion calmly.
A customer is confused about a charge: [paste]. Write a clear, reassuring explanation of what the charge is for, broken down in plain terms, and tell them exactly what to do if it's still wrong.
10. Handle a cancellation
Process it cleanly, leave the door open.
A customer wants to cancel their [subscription or order]. Write a reply that processes it without friction, confirms what happens next and to any refund, and leaves the door open, without a pushy save attempt.
Orders, delays, and outages
When something's gone wrong on your side, honesty plus a date does the heavy lifting. Customers forgive a delay far more easily than they forgive vagueness, so every prompt here pushes for a specific reason and a real next date rather than a "soon".
11. Apologise for a delayed order
Own it, give a real date.
Write a reply to a customer whose order is delayed. Acknowledge the inconvenience briefly, explain the delay is due to [reason], give the new expected date of [date], and offer [compensation if any]. Reassuring but honest.
12. Handle an out-of-stock item
Break it gently, offer the options.
A customer ordered [item] that's now out of stock. Write a reply that breaks the news clearly, lays out their options (wait, swap, or refund), and makes the swap appealing if it genuinely fits. Details: [paste].
13. Send an outage update
Calm, clear, no blame.
We're having a service outage affecting [what]. Write a short, calm status update for affected customers that says what's happening, what we're doing about it, and when we'll update them next. No jargon, no blame.
14. Answer "where is my order"
Status, date, and a backup plan.
A customer is asking where their order is: [paste tracking or status]. Write a clear reply that gives the current status, the expected delivery, and what to do if it doesn't arrive by then.
15. Follow up after a fix
Check in like you mean it.
We resolved a customer's [shipping or order issue]. Write a short follow-up a day later, checking that everything arrived and worked out, in a way that feels caring rather than automated. Details: [paste].
Everyday replies (email and chat)
The bread-and-butter messages, written to sound like a person. These are the ones you send dozens of times a day, which is exactly why a good draft saves the most time, as long as it doesn't read like a form letter.
16. Answer a how-to question
Clear steps they won't get lost in.
A customer asked how to [do something with our product]: [paste]. Write a clear, friendly step-by-step answer they can follow without getting lost, and point them to [the relevant help article] at the end.
17. Help with login or access
Most-likely fix first.
A customer can't [log in or access their account]: [paste]. Write a calm, reassuring reply that walks them through the most likely fix first, then the backup steps, and offers to escalate if none of it works.
18. Write a quick chat reply
Short, warm, and human.
Give me a short, warm live-chat reply to a customer asking [question]. Keep it under three sentences, sound human, and end by checking if there's anything else they need.
19. Confirm a resolution and close
End on a good note.
A customer's issue with [topic] is now resolved. Write a friendly closing message that confirms the fix, thanks them for their patience, and invites them back if anything else comes up.
20. Reply to a thank-you
Accept it warmly, briefly.
A customer sent a kind thank-you message: [paste]. Write a short, genuine reply that accepts the thanks warmly without overdoing it, and makes them feel like a valued customer.
Saying no and handling tricky asks
The hard messages, where being clear is kinder than being vague. A wishy-washy "no" leaves the customer hoping and you repeating yourself, so these say it plainly and always pair the no with the nearest thing you can actually do.
21. Decline a request with an alternative
No, but here's what we can do.
A customer is asking for [something we can't do]: [paste]. Write a reply that says no clearly but kindly, explains the reason in one line, and offers the closest thing we can do instead.
22. Handle an out-of-scope ask
Point them to the right place.
A customer wants help with [something outside what we support]: [paste]. Write a polite reply that explains it's outside what we can help with, and points them to where they can get the right help.
23. Respond to a feature request
Grateful, honest, no overpromising.
A customer asked for a feature we don't have: [paste]. Write a reply that thanks them genuinely, tells them honestly it isn't available now without overpromising a timeline, and explains how we'll pass the feedback on.
24. Push back on an unreasonable demand
Firm, fair, and calm.
A customer is demanding [unreasonable thing] and threatening [to leave or a chargeback]: [paste]. Help me write a calm, firm reply that holds our position, stays respectful, and offers a fair path forward.
25. Decide whether to escalate
Get a read before you reply.
Here's a customer message: [paste]. Tell me whether this should be handled by a frontline agent or escalated to a manager or specialist, and why, then draft the handoff note or the reply accordingly.
Templates and macros
Build the reusable set once, so the queue moves faster without sounding canned. The trick with templates is keeping them human: a macro that reads like a robot is worse than no macro, so each prompt here leaves room for a real person to come through.
26. Build a canned-response set
Your most common questions, answered.
Help me build a set of reusable canned responses for the [number] most common questions our [type of business] gets: [list them, or ask me]. Keep each short, on-brand, and easy to personalise, with a [bracket] for the customer's name.
27. Write a brand voice guide
Keep the whole team sounding like one.
Help me write a short customer-service voice guide for our team: three or four voice traits, a 'we sound like / we don't sound like' list, and two example replies in that voice. Our brand is [describe].
28. Create greetings and sign-offs
Human openers and closers.
Give me a set of warm, on-brand greeting and sign-off lines for customer emails and chats, a few options for each, that sound human and not corporate. Our tone is [describe].
29. Draft an FAQ answer
Short, clear, jargon-free.
Write a clear, friendly FAQ answer to the question '[paste]' for our [type of business]. Keep it to two or three sentences, link to [more help] if useful, and avoid jargon.
30. Rewrite a robotic macro
Make a stiff template human.
This canned response sounds robotic: [paste]. Rewrite it to sound like a helpful human, keep it about the same length, and cut anything that feels like corporate filler.
Work the queue faster (the agent-side prompts)
This is the half most prompt lists skip, and it's where the real time goes. None of these touch the customer; they help you get through the work. A long ticket you don't have to re-read, or a handoff note you don't have to compose from scratch, is the kind of small saving that adds up fast across a full queue.
31. Summarise a long ticket
Cut a wall of replies down to the point.
Summarise this long ticket thread into the customer's actual problem, what's already been tried, and the single next step needed. Thread: [paste].
32. Draft an escalation note
A clean handoff the next person can act on.
I need to escalate this ticket to [team or manager]. Write a concise internal handoff note: the customer's issue, what I've done, what I think is needed, and how urgent it is. Ticket: [paste].
33. Turn a resolution into a help article
Solve it once, document it for everyone.
We just solved this issue: [paste the ticket and the fix]. Turn it into a short, reusable help-centre article with a clear title, the steps to fix it, and when to contact support.
34. Spot what they're actually asking
Find the real request inside a rambling message.
This customer message is long and a little unclear: [paste]. Tell me what they're actually asking for, which details they've given that matter, and what I should clarify before I reply.
35. Triage a batch of tickets
Sort the queue and spot the risks.
Here are several open tickets: [paste]. Group them by type, flag the ones that look urgent or at risk of churn, and suggest the order I should work them in.
Learn from customers (feedback and improvement)
The same messages that fill your queue are a map of what to fix. These turn them into something useful. Support sees problems before anyone else in the company does, so a few minutes spent summarising what customers keep saying can be worth more than the individual replies.
36. Summarise feedback themes
Find the pattern in the noise.
Here's a batch of recent customer feedback and complaints: [paste]. Pull out the main recurring themes, roughly how often each comes up, and the one issue that would help the most customers if we fixed it.
37. Spot a churn risk
Catch the customer about to leave.
Read this customer's recent messages: [paste]. Tell me whether they sound at risk of leaving, the signs you're picking up on, and one thing I could do or say to keep them.
38. Draft a win-back message
Bring back someone you lost.
A customer cancelled a while ago over [reason], which we've since improved. Write a genuine, low-pressure message letting them know what's changed and inviting them back, without sounding desperate.
39. Ask for a review at the right moment
Catch them while they're happy.
A customer just had a great experience after we resolved [issue]: [paste]. Write a short, natural message asking if they'd leave a review, that doesn't feel pushy or transactional.
40. Turn complaints into product notes
Hand the team something specific.
Here are several complaints about [area of the product]: [paste]. Summarise them into clear, specific notes I can pass to the product team, with the customer impact for each.
Draft with AI, keep a human on anything sensitive
There's a line worth holding through all of this. ChatGPT is excellent at the first draft and at the internal work, but it doesn't know your policy, your account systems, or what you're actually allowed to offer unless you tell it, and even then it will sometimes promise a refund you don't give or state a rule that's slightly wrong, in a perfectly confident voice.
So treat it as a fast assistant, not an autopilot. A person should read anything that touches money, an account, a policy decision, or an upset customer before it goes out, and you shouldn't wire ChatGPT up to send sensitive replies on its own. Check the facts it states, and keep your brand voice human, because customers can tell when they're talking to a script. Used that way, it gives your team back the time the queue eats, and keeps the judgement where it belongs, with a person.
What this post does not cover
These prompts are drafting aids for a support team, not a replacement for your judgement or your policies, and nothing here is legal advice. How you handle refunds, cancellations, and disputes depends on your own terms and the law where you operate, so confirm those before you act on any draft. Model names and features change quickly, so check the current details in your tool. For the thinking behind prompts like these, see the free prompt library for more, and if you run the business too, our prompts for small business owners and prompts for writing emails cover the rest of the customer conversation.
Sources
Frequently asked questions
Written by
Tapabrata Biswas
Tech Researcher
I test AI productivity tools and research home-automation gear the way most people use them. Not in a lab, but on an ordinary desk with an ordinary internet connection. The only test that matters: does it save you time?
Connect on LinkedInShare the Post with Your Besties
Get the plain-English tech brief
One email a week on AI tools and smart-home tech. No jargon, no hype.
