10 AI Prompts for Customer Service Replies (Brand-Voice Ready)

Support replies after 8 pm are where good businesses sound like bad ones. You've answered the same shipping question eleven times today, customer twelve is angry, and your patience clocked out around lunch.

That's the real use case for AI prompts for customer service — not replacing you, but drafting the reply while you're tired so you only have to fix it instead of write it. The ten prompts below cover the messages small businesses actually get: delays, refund requests, angry emails, reviews, and the follow-through most owners skip.

Copy them into ChatGPT or Claude as-is, fill the brackets, and read every single draft before it goes out. That last part isn't optional.

How to use these AI prompts for customer service

The workflow is draft, review, send — always in that order, always with you in the middle. A draft that's 90% right saves you ten minutes. A draft that auto-sends with the wrong refund amount costs you a customer and possibly a chargeback.

So nothing here gets wired straight to your inbox. If you want the bigger picture of where AI safely fits in support, we laid it out in how to use AI for customer service. For now: paste, fill, review, send.

Prompts for the hard replies

1. The empathetic delay reply

You are a customer support writer for a small business. Draft a reply
to a customer whose [ORDER/SERVICE] is delayed.

Details:
- Their message: [PASTE THE CUSTOMER'S MESSAGE]
- The honest reason for the delay: [REASON]
- The new, realistic date: [DATE]
- What we're doing to make it right: [DISCOUNT / UPGRADE / NOTHING YET]

Rules: open by acknowledging the inconvenience in plain words, give the
real reason in one sentence, commit clearly to the new date, no filler
like "we apologize for any inconvenience," under 120 words, signed [NAME].

Send this the moment you know a date will slip, not when the customer chases you. Good output names the real reason in one plain sentence — people forgive honesty far faster than "unforeseen circumstances."

2. Refund request triage

A customer is asking for a refund. Help me decide, then draft the reply.

- Their message: [PASTE MESSAGE]
- Our refund policy: [PASTE OR SUMMARIZE IT]
- The order: [WHAT THEY BOUGHT, WHEN, PRICE]
- My gut read: [APPROVE / DENY / PARTIAL / UNSURE]

First, tell me where this falls under our policy and flag anything
ambiguous. Then draft two replies — one approving the refund, one
offering [STORE CREDIT / REPLACEMENT] instead. Both under 100 words,
warm but unambiguous. I'll choose which to send.

The drafts are the smaller half of the value here. The bigger half is forcing a policy check before you decide anything while annoyed — the decision stays yours, made calmly.

3. Angry-customer de-escalation

You write de-escalation replies. A customer is angry. Draft a response
that lowers the temperature without groveling or arguing.

- Their message: [PASTE THE ANGRY MESSAGE]
- What actually happened on our side: [THE FACTS]
- What I can offer: [OPTIONS, IF ANY]

Structure: one sentence showing I actually read the complaint (name
their specific issue, never "your concerns"), one sentence of ownership
for whatever is genuinely ours, then the fix or next step with a
timeframe. No defensiveness, no policy quotes, under 110 words.

The draft comes back calmer than anything you'd write in the moment, which is the entire point. If the thread is already two angry replies deep, the full complaint de-escalation workflow goes further than one prompt can.

4. The "we messed up" apology

We made a real mistake and I need to own it. Draft an apology email.

- What we got wrong: [THE MISTAKE, STATED PLAINLY]
- What it cost the customer: [THE IMPACT]
- How we're fixing it: [THE FIX + WHEN]
- What changes so it doesn't repeat: [PROCESS CHANGE, ONLY IF TRUE]

Rules: say "we got this wrong" within the first two sentences, no
passive voice ("mistakes were made"), no excuses before the apology,
offer the fix without being asked twice. Under 130 words.

Send it from the owner's address, not a support alias. And only include the "what changes" line if something genuinely changes — customers can smell a decorative process fix.

Prompts for reviews and repeat questions

5. The positive review response

Draft a reply to this positive review: [PASTE REVIEW]

My business: [NAME + WHAT WE DO]. Mention one specific thing they
praised, in your own words — don't parrot the review back. Thank them
like a person, not a brand, and invite them back with something
concrete ([NEW ITEM / SEASONAL SERVICE]). Three sentences max.
Never "we're thrilled."

Ninety seconds per review, and the customer feels seen instead of processed. Reply to the good ones too — future customers read those threads.

6. The negative review response

Draft a public reply to this negative review: [PASTE REVIEW]

Our side of it: [WHAT ACTUALLY HAPPENED]. Remember that future
customers will read this reply, not just the reviewer. Acknowledge
their specific complaint, state our side in one neutral sentence only
if the facts matter, then offer to make it right offline with a real
contact ([EMAIL / PHONE]). Four sentences max, zero sarcasm.

Your real audience is the stranger scrolling reviews next month, deciding whether to call you. When reviews become a weekly chore, the review response writer handles both kinds with your voice already baked in.

7. FAQ answer drafting

Turn these repeated customer questions into FAQ answers.

Questions: [PASTE 5-10 REAL QUESTIONS FROM YOUR INBOX]
Business context: [WHAT YOU SELL + KEY POLICIES]

For each: a plain-English answer under 80 words, written the way a
helpful staff member would say it out loud, with every number, price,
and timeframe stated exactly. Flag any question you can't answer from
the context I gave — never guess at a policy.

Run this once a quarter on real inbox questions, then save the answers as canned replies. That's the seed of a proper library of saved support replies, which is where the real time savings compound.

Prompts for the follow-through

8. The escalation summary

Summarize this support thread so a teammate can take over cold.

Thread: [PASTE THE FULL EMAIL OR CHAT THREAD]

Output: the customer's name and what they bought, the issue in two
sentences, everything promised so far (with dates), the current
emotional temperature (calm / frustrated / furious), and the one thing
the next person must do first. Ten lines max.

Handoffs are where customers get asked to repeat themselves, which is where mildly annoyed becomes furious. This takes 60 seconds and prevents that.

9. The follow-up after resolution

Draft a short check-in email to a customer whose issue we resolved
[NUMBER] days ago.

- The issue: [WHAT WENT WRONG]
- The fix: [HOW WE RESOLVED IT]

Ask whether the fix is holding up, keep it to three sentences, and
make it easy to answer in one word. No surveys, no "rate your
experience," no upsell.

Almost nobody does this, which is exactly why it works. A three-sentence check-in a week after a problem turns a complaint into a story the customer tells in your favour.

10. Tone-matching to your brand voice

Here are three replies I wrote myself that sound like us:
[PASTE THREE REAL REPLIES YOU'RE PROUD OF]

Describe the voice in five bullets: formality, warmth, sentence
length, words we use, words we'd never use. Then rewrite this draft
in that voice: [PASTE AN AI DRAFT]

Keep every fact identical. Change only the voice.

Run this once, save the five-bullet voice description, and paste it into any of the prompts above. It's the difference between replies that sound like support software and replies that sound like your shop.

Where this breaks

The AI can't see your order system, so every fact — dates, amounts, what the customer bought — has to come from you, and you have to check it survived the draft. Anything involving a legal threat, a chargeback dispute, or a safety issue gets written by a human from scratch. And auto-sending is off the table no matter how good the drafts get; the day you stop reading them is the day one goes out wrong.

Once these prompts are saving you real time, the next step is chaining them into one reusable system — that's exactly what the customer support autopilot is built to do.