Why this support reply prompt works
A good support reply does three things: it shows the customer they were heard, it explains what happens next, and it promises only what the company will actually do. Generic AI replies often fail the last point. Asked to "reply to an angry customer", a model may offer refunds, discounts or deadlines that nobody approved, and the customer will hold you to them.
This builder separates the inputs a support agent needs: the customer's own words, the facts of the case, what you can offer and what you cannot. The prompt tells the model to use only these, never to blame the customer or a named colleague, and to mark anything missing with [brackets] instead of inventing policy.
The channel matters too. A live chat reply should be short and conversational; a public review response is read by future customers, so it should stay polite, avoid private details and move the conversation to a private channel. The prompt ends with an internal note listing assumptions, so you can check before sending. For other customer emails, see the email prompt generator.
Worked example
Say a repeat customer's furniture order was lost by the courier and they have written in, frustrated. These are the inputs:
- Channel
- Customer's message
- This is the second time my order is late. I ordered the oak side table 12 days ago and tracking hasn't moved in a week. I'm really disappointed and want to know what's going on.
- Company and product
- Loomly Home, an online furniture shop
- Facts of the case
- Order shipped 12 days ago. The courier has confirmed the parcel is lost. The table is in stock. The customer had a late delivery once before in March.
- What you can offer
- Ship a replacement today with express delivery (2-3 working days), or a full refund within 5 working days - customer chooses
- What you cannot offer
- We cannot offer a discount code on top
- Tone
- Empathetic and calm
- Length
- Short (under 120 words)
- Sign as
- Sam from the Loomly team
The builder turns these inputs into the following prompt. Load it into the form above with the Lost parcel example chip and change anything you like.
You are an experienced customer support specialist at Loomly Home, an online furniture shop. You write replies that calm people down and solve the problem.
## Task
Write a reply to the customer message below. Channel: email.
## Facts of the case (use only these)
Order shipped 12 days ago. The courier has confirmed the parcel is lost. The table is in stock. The customer had a late delivery once before in March.
## What we can offer
Ship a replacement today with express delivery (2-3 working days), or a full refund within 5 working days - customer chooses
## What we cannot offer
We cannot offer a discount code on top
## Requirements
- Tone: Empathetic and calm
- Length: Short (under 120 words)
- Acknowledge the customer's specific problem in the first sentence, without over-apologising.
- Explain clearly what happens next and when.
- Offer only what is listed above. Do not invent policies, discounts, deadlines or compensation; mark anything I must fill in with [brackets].
- Do not blame the customer, a colleague or a partner by name.
- Sign off as: Sam from the Loomly team
## Output format
1. The reply, ready to send.
2. One line of internal notes listing assumptions or facts to double-check.
## Customer message
"""
This is the second time my order is late. I ordered the oak side table 12 days ago and tracking hasn't moved in a week. I'm really disappointed and want to know what's going on.
"""This page shows only the prompt. The answer depends on the AI tool you paste it into, so always review the result before you use it.
Tips to improve your results
- Paste the customer's exact words. Tone and details in the original help the AI respond to the real concern.
- Write your options in the offer field precisely, including timeframes. The AI will not improve on them.
- Use the "cannot offer" field for limits that agents often get pushed on.
- Remove order numbers, addresses and other personal data unless your company allows them in AI tools.
- Save the prompt for common case types so replies stay consistent across your team.
Frequently asked questions
Can I use AI for customer service replies?
Many teams use AI to draft replies faster. The agent should always check facts, policy and tone before sending. This prompt keeps the model inside the facts and offers you provide.
How do I reply to an angry customer?
Acknowledge the specific problem, take responsibility where the company is at fault, explain what will happen and when, and offer a clear next step. Avoid long apologies without action. The prompt follows this structure.
How should I respond to a negative public review?
Keep it short, polite and specific, thank the reviewer, address the issue without arguing, and invite them to continue privately. Choose "public review response" as the channel.
Is customer data safe?
This site never receives what you type; the prompt is built in your browser. Once you paste it into an AI tool, that tool's terms apply, so remove personal data you do not need and follow your company's data rules.
Can it reply in the customer's language?
Yes. Add "Reply in Spanish" (or any language) to the facts field. Ask a native speaker to check important replies if possible.