Customer complaint reply prompt: free ChatGPT template
Updated · By Robert Breen
Use this when a customer writes in angry about a late arrival, a mess or a job that doesn't feel finished. You get a calm reply that apologizes once, answers each complaint in order and offers only the fix you approved. The risky part is the make-up offer: an AI trying to calm someone down will happily promise a refund or "this will never happen again," so the prompt holds it to your numbered facts, as in Reply Faster.
The prompt
Copy it into ChatGPT (or Claude, Gemini or Copilot) and replace every [BLANK] with your own details. It uses the four parts from Prompt Writing 101: Role, Context, Task and Format.
Role: You are [YOUR NAME] at [BUSINESS], replying to an upset customer. You sound like a person who owns the problem, not a call center. Context: What is true and what we can offer, numbered: [FACTS]. Only promise what is in those facts. If they ask for something the facts don't cover, say I'll get back to them by [FOLLOW UP DAY]. Task: Write my reply. Apologize once for what actually went wrong, then answer every complaint in the order they raised it. Don't argue with their version, don't blame anyone, and don't say it will never happen again. Format: Under 170 words, short paragraphs, no subject line (it's a reply in the thread). No "we value your feedback." Sign with my name and business name. Here is the customer's email: [CUSTOMER EMAIL]
Fill in the blanks
[YOUR NAME]- Who signs it. A real name reads better than "Customer Service" to someone already annoyed.
[BUSINESS]- Your business name and what you do.
[FACTS]- A numbered list: what happened, what was your fault, and exactly what you'll do (the visit, the credit, the amount). No approved discount, no discount in the list.
[FOLLOW UP DAY]- When you'll answer anything the facts don't cover, such as "tomorrow by noon." Pick a day you'll keep.
[CUSTOMER EMAIL]- The complaint itself, with card numbers or anything private taken out.
Example, filled in
A made-up example from the free lesson Build an AI Customer Follow-Up Agent for Plumbers. Blue Maple Plumbing is the made-up plumbing company in the AI Customer Follow-Up Agent lesson. Maria had a sump pump installed this week, and the visit went badly.
Role: You are Kelly, the office manager at Blue Maple Plumbing, replying to an upset customer. You sound like a person who owns the problem, not a call center. Context: What is true and what we can offer, numbered: 1. Our plumber, Ray, was held up at a burst pipe and arrived at 11:40 instead of 9:30. The office should have called Maria and didn't. 2. Ray didn't put drop cloths on her basement stairs. That's on us. 3. Ray can come back Thursday between 1 and 3 p.m., at no charge, to clean the stairs and check why the pump is loud. 4. The owner approved $50 off her invoice. Only promise what is in those facts. If they ask for something the facts don't cover, say I'll get back to them by tomorrow at noon. Task: Write my reply. Apologize once for what actually went wrong, then answer every complaint in the order they raised it. Don't argue with their version, don't blame anyone, and don't say it will never happen again. Format: Under 170 words, short paragraphs, no subject line (it's a reply in the thread). No "we value your feedback." Sign with my name and business name. Here is the customer's email: I took the morning off for a 9:30 appointment and your plumber showed up at 11:40 with no call. He tracked mud down my basement stairs and I cleaned it myself. And the new pump is louder than my old one, it clunks every time it shuts off. For what I paid I expected better. Maria
What a good answer looks like
- One apology near the top that names the real problems (no call, the mud), not "sorry for any trouble."
- All three complaints answered in Maria's order: the late arrival, the stairs, the loud pump.
- The offer exactly as approved: Thursday, 1 to 3 p.m., no charge, and $50 off.
- At most one short reason for the delay. A paragraph about the burst pipe reads as an excuse.
How to check it before you use it
An AI draft can sound right and still say something you never told it. Use the habit from the Reply Faster lesson: check every promise, name, date and number against what you gave it before anything goes out.
- Check every promise against the four facts. Delete "this won't happen again," "we've retrained our team" or "a full refund": none of them is on the list.
- Look hard at the pump answer. The facts promise a visit to check it, so "that clunk is normal" is a diagnosis nobody made.
- Match the numbers. A draft that turns $50 off into "your next visit is free" just raised your cost.
- For a second pass, run the check reply promises prompt on the draft.
Why this prompt works
The numbered Context gives the AI something real to apologize with instead of something to invent, and the Task rules out the three habits that make complaints worse: arguing, blaming and overpromising. That is grounding applied to an angry email.
Practice it in the free lesson
Reply Faster: Turn a Messy Customer Email Thread into a Clear, Friendly Reply: Turn a messy customer email thread into a clear, friendly reply: give ChatGPT the thread and your facts, check every promise, fix the tone, and send it from Gmail. You do every step yourself in a practice copy of ChatGPT, and nothing touches your real accounts.
Also useful: Build an AI Customer Follow-Up Agent for Plumbers
Terms used here
Related prompts
- Check reply promises prompt: A ChatGPT prompt that checks a draft before you send it: a table of every promise, the fact behind each, and safer wording for…
- Angry customer call talking points prompt: A ChatGPT prompt that preps you to call an upset customer: an opening line, facts to say, your real offer, answers to pushback…
- Negative review response prompt: A ChatGPT prompt for a public reply to a bad review: calm, specific, no private details, no arguing, and an easy way to continue…
- Fix reply tone prompt: A ChatGPT prompt that fixes the tone of a draft (too stiff, too defensive, too long) while keeping every fact, number and "we…