ChatGPT Prompts for Customer Service Replies: 35 Templates You Can Copy

Bottom line: These 35 ChatGPT prompts cover the customer service replies you write every single day, refunds, delays, angry emails, cancellation requests, and reviews, and every one of them is written so you paste in your specific details and get a draft you can edit in under a minute. None of them are meant to go out word for word. That's the bit people skip, and it's the reason half the "AI customer service" complaints you see online exist in the first place.

Why most ChatGPT customer service replies sound wrong

I worked with an ecommerce client last year, a home goods brand doing around £40,000 a month in sales, whose support inbox had grown from one person to three during a bad Q4. The founder had read that ChatGPT could "handle" customer service and told the team to use it for every reply. Within three weeks, two customers had screenshotted responses on Twitter because the tone was so flat and identical that people noticed a pattern. One reply to a upset customer, whose order had arrived smashed, opened with "I understand how frustrating this must be for you" followed by a bullet-pointed refund policy. Nobody talks like that. The customer wrote back saying it felt like she was being handled by a script, which, fairly, she was.

That's the uncomfortable part nobody selling AI tools wants to say out loud: pasting a ChatGPT draft straight into a reply box without changing it is often worse than a slower, plainer human answer. Customers can tell. Not because the grammar is wrong, but because the empathy is generic and the specifics are missing. The fix isn't to avoid ChatGPT, it's to use it the way I'll show you below, as a first draft you edit for 30 to 60 seconds before sending, never as the final word.

How to use these prompts

Five steps, every time, no exceptions:

• Paste the prompt into ChatGPT and fill in the brackets with the real details, order number, product name, dates, whatever applies.
• Tell it your brand's tone in one sentence the first time you use it, for example "we're warm and a bit informal, never corporate" or "we're precise and no-nonsense, this is a B2B software client."
• Read the draft out loud. If it sounds like something you'd never say to a friend, it needs editing.
• Delete at least one generic sentence and replace it with a specific fact about this customer's situation.
• Send it from your own inbox, not a canned response tool, at least for the first few months you're building this habit.

If you want a fuller breakdown of tone and phrasing for support replies specifically, I wrote a separate piece with ChatGPT prompts for customer service replies that do not sound like a robot, which pairs well with the templates below. And if you're new to prompting altogether, start with these ChatGPT prompt examples for beginners before diving into anything customer-facing.

Refunds and returns (5 prompts)

• Prompt: "Write a reply approving a refund for [product name], order [number], purchased on [date]. Confirm the amount, the refund method, and how many business days it takes. Tone: warm, no apology-heavy language, keep it under 80 words." Use when the refund is straightforward and you just need to confirm it clearly.
• Prompt: "Write a reply explaining we can't offer a full refund on [product] because [reason, e.g. it's outside the 30-day window], but offer [alternative, e.g. store credit or a repair]. Acknowledge their disappointment without sounding defensive." Use for partial or declined refunds where you still want goodwill.
• Prompt: "Write a short reply confirming we've received a returned item [product name] and the refund is being processed. Mention it will show up in [X] business days depending on their bank." Use for the confirmation step after a return arrives.
• Prompt: "Write a reply to a customer asking why their refund is taking longer than expected. Explain that bank processing can take [X] days, apologise for the wait once, and give them a specific date to check back." Use to reduce follow-up messages.
• Prompt: "Write a reply offering a refund plus a small goodwill gesture (10% off their next order) for a customer who received a damaged [product]. Keep it brief and don't over-apologise." Use for damaged goods where you want to keep the relationship.

Shipping delays (4 prompts)

• Prompt: "Write a proactive email to customers whose orders are delayed due to [reason, e.g. courier backlog]. Give a new estimated delivery date, one sentence of context, and no more than one apology." Use for batch delay notices.
• Prompt: "Write a reply to a customer asking where their order is. Order number [X], placed [date], tracking shows [status]. Give them the real next step, not a vague 'it's on its way.'" Use for individual tracking queries.
• Prompt: "Write a reply to a customer whose order is now two weeks late with no tracking movement. Offer a choice: wait with a small credit, or a full refund. Keep it direct, they're likely frustrated." Use for serious delays.
• Prompt: "Write a reply confirming an order has shipped, including the tracking link placeholder [TRACKING_LINK] and an estimated arrival window. Make it sound like a person wrote it, not an automated shipping notice." Use for standard shipping confirmations you want to feel personal.

Angry or frustrated customers (5 prompts)

• Prompt: "A customer is angry because [specific situation]. Write a reply that opens by naming the actual problem back to them in one sentence, not a generic apology, then explains what we're doing to fix it. No corporate phrases like 'we value your feedback.'" Use for high-emotion complaints.
• Prompt: "Write a reply to a customer who says our product broke after [X weeks/months]. Ask two clarifying questions we need before offering a solution, and make clear we intend to sort it, not just gather information." Use when you need more detail before resolving.
• Prompt: "Write a de-escalation reply for a customer who has used strong or rude language in their message. Stay calm and professional, don't mirror their tone, but don't grovel either." Use for the messages that make your team wince.
• Prompt: "Write a reply to a customer threatening to leave a bad review over [issue]. Address the issue directly with a real fix, and don't mention the review threat at all." Use to keep focus on solving the problem, not managing PR.
• Prompt: "Write a reply to a repeat complainer who has contacted us three times about the same unresolved issue. Acknowledge the repetition honestly, apologise once for the delay, and give a firm resolution with a deadline." Use for escalations that have dragged on too long.

Product questions (4 prompts)

• Prompt: "Write a reply answering a pre-sale question about [product feature]. Be specific and factual, avoid sales language, and end with one relevant follow-up question to help them decide." Use for pre-purchase queries.
• Prompt: "Write a reply explaining how to use [product/feature] step by step for a customer who says they're confused. Use numbered steps, keep each step under 15 words." Use for how-to support requests.
• Prompt: "Write a reply comparing [product A] and [product B] for a customer deciding between them, based on these differences: [list differences]. Be honest about trade-offs, don't oversell either one." Use for comparison questions.
• Prompt: "Write a reply to a customer reporting a bug or fault with [product/feature]. Confirm we're looking into it, give a realistic timeframe, and tell them what to do in the meantime." Use for technical issues you can't fix instantly.

Cancellations and retention (4 prompts)

• Prompt: "Write a reply processing a subscription cancellation for [customer name], confirming the end date and that no further charges will apply. Don't try to talk them out of it in this message." Use when a customer has clearly decided.
• Prompt: "Write a reply to a customer asking to cancel, offering one genuine alternative (pause, downgrade, or discount) before processing the cancellation, in a way that doesn't feel like a hard sell." Use for save-attempt replies.
• Prompt: "Write a short exit survey follow-up asking a cancelled customer for one honest reason they left, framed as wanting to improve, not as a guilt trip." Use to gather churn feedback.
• Prompt: "Write a reply to a customer who cancelled and is now asking to come back. Welcome them without over-apologising for whatever made them leave, and confirm the practical next steps." Use for win-back replies.

Saying no without losing the customer (3 prompts)

• Prompt: "Write a reply declining a refund request because [specific policy reason]. State the policy clearly once, don't repeat it three times, and offer one alternative if one exists." Use for firm refund declines.
• Prompt: "Write a reply declining a custom request (e.g. a feature, a bespoke order, a price match) that we can't accommodate. Be direct about the no in the first two sentences, then explain why." Use for requests you can't fulfil.
• Prompt: "Write a reply to a customer asking for a discount we don't typically offer. Decline clearly but warmly, and mention one thing we can offer instead if applicable." Use for price negotiation requests.

Reviews and feedback requests (4 prompts)

• Prompt: "Write a short message asking a satisfied customer to leave a review on [platform], sent after a positive support interaction about [issue resolved]. Keep it under 50 words and don't sound needy." Use after resolving something well.
• Prompt: "Write a public reply to a positive review mentioning [specific detail from the review]. Thank them specifically, not generically." Use for review responses that need to feel personal.
• Prompt: "Write a public reply to a negative review about [specific complaint]. Acknowledge the issue honestly, state what we've done or will do, and invite them to continue the conversation privately." Use for public complaint management.
• Prompt: "Write a follow-up message to a customer who left a 3-star review, asking what would have made it a 5-star experience, without sounding defensive about the rating." Use for mid-range reviews worth improving.

Social media and public complaints (3 prompts)

• Prompt: "Write a public reply to a complaint left on [platform, e.g. Twitter/X or Facebook] about [issue]. Keep it short, acknowledge the problem, and move the conversation to DMs or email." Use for the first public response.
• Prompt: "Write a private DM following up on a public complaint, referencing what they said publicly, and offering a specific resolution." Use once you've moved the conversation off the public feed.
• Prompt: "Write a reply to a viral or widely shared complaint that other customers are commenting on. Be transparent about what went wrong and what we're changing, without sounding scripted or legalistic." Use for the rare but stressful pile-on situation. If you handle a lot of public replies, my ChatGPT prompts for social media managers post has more on tone for public-facing messages.

Apologies and follow-ups (3 prompts)

• Prompt: "Write a genuine apology for [specific mistake we made], stating exactly what went wrong and what we're doing differently, without over-explaining or making excuses." Use when the fault is clearly ours.
• Prompt: "Write a follow-up message two days after resolving an issue, checking the customer is satisfied with the outcome, kept short and low-pressure." Use to close the loop and catch anything unresolved.
• Prompt: "Write a thank-you message to a long-term customer [name] marking [X years/months] with us, referencing something specific about their history with the business if known." Use for loyalty and retention touches.

What these prompts won't fix

ChatGPT can draft the sentence structure in seconds, but it can't know that your customer called in three times last month, or that the delivery driver left the package in the rain, or that this particular client hates being called by their first name. Those details have to come from your team, every time, or the reply reads exactly like the smashed-vase story I mentioned earlier. If your support volume has grown to the point where one person can't hold all that context in their head, that's usually the moment to bring in structure around it rather than more prompts, and it's a different job entirely from copying templates, that's when working with an AI implementation coach tends to save more time than another list of prompts ever will.

Worth saying too: none of this replaces a proper help centre or FAQ page. If you're answering the same question 40 times a week, that's a documentation problem, not a prompting problem. Fix the root cause first, then use these prompts for the messages that need a human judgment call, refunds, complaints, edge cases. That's also true outside support inboxes, the same discipline applies whether you're writing cold outreach emails or replying to a small business owner's own customers, which I cover in ChatGPT prompts for small business owners.

For the rest of the series, see the ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps. To write your own, try the free prompt generator.

Frequently asked questions

Can I use ChatGPT for customer service replies without customers noticing?

Not if you paste the draft straight in. Customers notice generic phrasing and missing specifics far more than they notice AI involvement, so the fix is editing every draft for at least 30 seconds and adding one detail unique to that customer's situation before you send it.

Is it safe to let ChatGPT handle refund decisions?

No, ChatGPT should draft the wording of the reply, not decide whether the refund happens. Keep policy decisions with a person who has authority to make exceptions, and use the prompts above once that decision is already made.

How many of these prompts should a small support team use?

Start with five, the refund confirmation, the shipping delay, the angry customer de-escalation, the decline template, and the review request. Those five cover the bulk of daily volume for most small businesses, and you can add the rest once the team is comfortable editing drafts quickly.

What's the biggest mistake businesses make with AI customer service replies?

Treating it as a way to send more replies faster rather than better replies with the same care. Speed without specifics is how you end up with the flat, identical-sounding messages that customers screenshot and complain about, which defeats the point entirely.

Further reading

• OpenAI's own platform documentation