30 AI Prompts for Customer Service Teams
Copy-ready prompts that assume you will attach the ticket, the thread or your tone guide — built to produce fast, empathetic, on-brand support at scale.
Each prompt gives the AI a support persona, references what you will attach, and applies a service framework (HEARD, LARA) with a defined output. These prompts are tool-agnostic and optimised to work across all major AI assistants — Claude, Microsoft Copilot and ChatGPT (and Google Gemini). Replace the [bracketed placeholders] with your own details, and always review AI output before you act on it.
Reply drafting
Solve-It Customer Reply
Replying to a customer with a straightforward, solvable issue
the customer's message
- Acknowledge the issue
- Answer the actual question with exact steps
- Confirm the resolution
- Offer one proactive next step
The reply plus a one-line internal note if follow-up is needed.
See example AI output
Hi Mei Ling, thanks for reaching out, and sorry for the delay on your refund. I checked order #4521 and see the payment was captured twice. Log into Billing then Transactions and you'll see the duplicate charge flagged for auto-reversal within 3 business days. Internal note: confirm gateway reversal posts by Friday.
Humanize a Robotic Reply
Cleaning up a stiff, template-sounding draft before sending
draft reply to rewrite
- Rewrite the reply to sound more human and friendly
- Keep every fact accurate and concise
- Add one line of genuine empathy
The rewritten reply plus a note on what changed.
See example AI output
Original: 'Your request has been received and will be processed within 5-7 business days per policy.' Rewritten: 'Thanks so much for your patience, Farah - I can see how frustrating the wait must be. Your request is confirmed and moving through processing, and you'll have an answer within 5-7 business days.' What changed: swapped the passive opening for a direct confirmation, added a named empathy line.
Three Reply Options by Tone
Giving an agent a choice of tone before responding to a query
the customer query
- Draft a brief reply option
- Draft a detailed reply option
- Draft an empathetic reply option
- Label each clearly so the agent can match customer mood
Three labelled reply versions.
See example AI output
Query: 'Why hasn't my order shipped yet?' Brief: 'Hi Daniel, your order ships tomorrow - tracking follows by email.' Detailed: 'Hi Daniel, your order was delayed by a warehouse stock check but is now cleared and ships tomorrow.' Empathetic: 'Hi Daniel, I'm sorry for the wait - I'll personally make sure you get tracking as soon as it's out.'
Explain a Policy Without Sounding Defensive
Telling a customer no without hiding behind policy language
no file - specify the policy or limitation
- State the policy or limitation clearly
- Explain the reasoning in one line, without hiding behind policy
- Offer the best available alternative, staying empathetic
The customer-ready reply.
See example AI output
Policy: no refunds after 30 days. Reply: 'Hi Vikram, your purchase is just outside our 30-day window, so I can't process a full refund - we set that limit to keep pricing fair for everyone. What I can do is offer a store credit valid for a year, or connect you with our repair team at no charge.'
Translate Tech Jargon for Customers
Turning a technical fix into steps a non-technical customer can follow
technical explanation to translate
- Turn the technical explanation into plain language, using an analogy if it helps
- Keep the steps in order and accurate, no condescension
- Confirm what success looks like
The customer-ready explanation.
See example AI output
Technical: 'Clear your DNS cache and flush the local resolver.' Customer-ready: 'Think of your device like a phone book that gets outdated. Open Settings then Network, tap Reset Network Settings, then restart your device. If the page loads normally, you're all set.'
Complaints and escalation
De-escalate with the HEARD Method
Responding to an upset customer who needs de-escalation
the upset customer's message
- Hear and empathise with the customer genuinely
- Apologise and take ownership
- Offer a concrete resolution with one clear next step
- Diagnose the root cause internally
The customer reply plus an internal diagnose note on the root cause.
See example AI output
Hi Mr. Tan, I hear how frustrating it's been to have your account locked for three days - that's on us, and I'm sorry. I've unlocked your account and applied a one-month credit. Internal diagnose note: fraud filter mis-flagged a device change after a phone upgrade.
Win Back a Churning Customer
Responding to a customer who has threatened to leave
no file - specify the issue prompting the threat
- Acknowledge the frustration and own what went wrong, no defensiveness or empty apology
- Present a specific make-good
- Give a concrete reason to stay
The reply plus a list of retention offer options.
See example AI output
Issue: repeated billing errors over three months. Reply: 'Hi Priya, three billing mistakes isn't okay. I've fixed this month's invoice and I'm crediting the last two in full. I'd also like to move you to a dedicated account manager.' Retention offers: two months free, a dedicated account manager, or a locked-in rate for 12 months.
Summarise a Long Complaint Thread
Getting up to speed on a drawn-out complaint before responding
the full complaint thread
- Extract the core issue, separating fact from emotion
- List everything the customer has been promised and flag any broken promise
- Identify what the customer actually wants and recommend a resolution with reasoning
A structured case summary plus a recommended resolution.
See example AI output
Core issue: laptop replacement delayed six weeks. Promises: replacement within 10 days (agent, wk1); refund if not received by wk5 (supervisor, wk5). Broken promise flagged: supervisor's refund commitment unmet. Recommendation: issue the promised refund immediately.
Draft a Genuine Apology Message
Apologising for a company mistake without generic 'sorry you feel' phrasing
no file - specify the mistake to apologise for
- Take clear responsibility for the mistake
- Briefly explain what happened
- State the fix and prevention plan
- Offer appropriate recompense, staying sincere and specific
The apology message.
See example AI output
Mistake: customer's order shipped to wrong address. Apology: 'Hi Ken, we sent your order to the wrong address - that's our error. It happened due to a mismatch between your new and old address on file. We've corrected the record and reshipped via express at no cost. As an apology, we've added a RM50 credit.'
Write an Internal Escalation Summary
Handing a difficult case up to a supervisor or specialist team
case thread or ticket history being escalated
- Summarise the customer and history, factually and completely
- List what has already been tried
- Note the customer's emotional state
- Recommend the next action
The internal escalation note.
See example AI output
Customer: Adrian Wong, Premium tier, 3-year account. History: reported app crashes 4 times since June; escalated to engineering, unresolved. Emotional state: calm but losing patience. Recommended action: bypass the queue, assign a senior engineer today, offer manual checkout as interim workaround.
Knowledge base and FAQs
Turn Questions into Help-Centre Articles
Converting a list of common customer questions into help-centre content
list of common customer questions
- Write a searchable title per question and lead with a one-line answer
- List the steps to resolve it, skimmable and jargon-free
- Add a 'still stuck?' line, one article per question
The set of help-centre articles.
See example AI output
Question: 'How do I reset my password?' Title: Reset Your Password in 3 Steps Steps: 1) Click 'Forgot password'. 2) Check email for reset link. 3) Set a new password. Still stuck? Check spam or contact support via live chat.
Write a First-Time-User How-To Guide
Documenting a task for someone doing it for the first time
no file - specify the task to document
- List the prerequisites
- Write numbered steps describing what the user will see, tested for clarity
- Flag the common mistake to avoid and add a success check
The step-by-step how-to guide.
See example AI output
Task: connecting a new printer. Prerequisites: printer powered on, Wi-Fi password ready. Steps: 1) Open Settings then Printers. 2) Select your printer's name. 3) Enter Wi-Fi password. Common mistake: choosing Bluetooth name instead of Wi-Fi name. Success check: a test page prints within 30 seconds.
Build a Ten-Question Product FAQ
Building an FAQ page for a product from scratch
no file - specify the product and real customer questions
- Identify the real questions customers ask and draft ten FAQ entries
- Order them by frequency
- Answer the real intent behind each question in 40 to 70 words
The ordered ten-entry FAQ.
See example AI output
1) Q: Does the subscription auto-renew? A: Yes, monthly on your billing date unless cancelled 24 hours before. You'll get a reminder email 3 days prior. 2) Q: Can I switch plans mid-cycle? A: Yes - upgrades apply immediately with a prorated charge.
Make a Help Article Scannable
Reworking a dense help article that's hard to skim
existing help article
- Add a summary box up top
- Break content into short numbered steps and add headings
- Cut anything redundant while keeping accuracy
The rewritten, scannable article.
See example AI output
Summary box: Cancel your subscription in under 2 minutes from Account Settings. How to cancel: 1. Go to Account then Subscription. 2. Tap Cancel Plan. 3. Confirm. (Cut: three paragraphs of legal boilerplate moved to a linked terms page.)
Design a Branching Troubleshooting Guide
Diagnosing a common problem with multiple possible causes
no file - specify the common problem to diagnose
- State the symptom
- Write branching diagnostic questions with logical branches
- Map the likely causes and fix at each branch
- End every branch with a contact-support fallback, no dead ends
The troubleshooting flow, written out in words.
See example AI output
Symptom: blank screen on launch. Q1: Does it happen on Wi-Fi and mobile data? If only Wi-Fi, restart router. Q2: Is the app updated? If not, update from the store. Still blank after both fixes? Contact support with device model and OS version.
Tone and templates
Build Five Core Reply Templates
Standardising replies for your five most common request types
no file - specify your five most common request types
- Identify the five most common request types
- Write a template for each in brand voice, personal not robotic
- Add placeholders and note when to use each template
Five labelled reply templates.
See example AI output
Template 1 - Shipping delay: 'Hi [Name], I checked on order [#] and it's currently [status]. Expected arrival is now [date].' Template 2 - Refund confirmation: 'Hi [Name], your refund of [amount] for order [#] is confirmed.'
Draft a One-Page Support Tone Guide
Documenting your team's voice so replies stay consistent
no file - optionally describe your brand personality
- Define three voice principles
- List do's and don'ts with example phrases to use and avoid
- Explain how to handle an angry customer on-brand, practical and specific
A one-page tone-of-voice guide.
See example AI output
Voice principles: warm, direct, human. Do: 'I can see why that's frustrating.' Don't: 'We apologise for any inconvenience caused.' Handling anger on-brand: slow down, name the emotion once, skip scripted apologies, lead with the fix.
Make Canned Responses Feel Personal
Fixing canned responses that read as robotic
existing canned responses
- Rewrite each canned response to sound less robotic while staying efficient
- Add a human line to each
- Keep them reusable
The improved templates.
See example AI output
Original: 'Your ticket has been received and will be addressed within 24-48 hours.' Improved: 'Thanks for reaching out - I've got your ticket and I'm on it. You'll hear back within 24-48 hours, sooner if it's a quick fix.'
Write Holiday and After-Hours Auto-Replies
Setting expectations while the team is offline
no file - specify holiday dates or after-hours window
- Acknowledge the message was received
- Give the real response time, honest timelines
- Offer self-service options in the meantime and reassure the customer
The auto-reply messages.
See example AI output
After-hours: 'Thanks for reaching out! Our team is offline right now (we're open Mon-Fri, 9am-6pm) but your message is in the queue and we'll reply within 12 hours.' Holiday: 'We're closed for Hari Raya from June 1-3 and will reply starting June 4.'
Decline an Out-of-Scope Request Gracefully
Turning down a request without burning the relationship
no file - specify the out-of-scope request
- Acknowledge the request
- Briefly explain why it's out of scope, no false promises
- Offer the nearest thing you can do and leave the door open
The decline message.
See example AI output
Request: customer wants a custom feature built just for them. Message: 'Thanks for the detailed suggestion - custom one-off builds aren't something we can do, but I've logged this as a feature request, and our API integration might get you 80% of the way there.'
Analysis and insight
Find Top Recurring Issues from Tickets
Spotting patterns across a batch of tickets to fix root causes
batch of customer messages or tickets
- Identify the top three recurring issues and quantify how common each is
- Trace each to its root cause, not just the symptom
- Recommend a product or process fix per theme, quoting examples
A themed findings table with a recommended fix per theme.
See example AI output
Theme 1 - Confusing checkout (34% of tickets): 'I couldn't find where to apply my promo code'. Root cause: promo field hidden. Fix: surface it by default. Theme 2 - Late shipping notifications (22%): trigger tracking email at dispatch, not delivery.
Summarise Sentiment from Customer Feedback
Reporting sentiment trends from a batch of feedback
customer feedback to analyse
- Summarise the overall sentiment
- Identify themes driving positive and negative feeling, quoting representative comments
- Suggest an action per theme, without over-reading a small sample
A sentiment summary with a suggested action per theme.
See example AI output
Overall sentiment: 68% positive, 22% neutral, 10% negative. Positive theme - support speed: 'Got a reply in under an hour'. Action: highlight response time in marketing. Negative theme - app bugs: 'App crashes every time I upload a photo'. Action: prioritise the fix.
Turn Tickets into a Product Report
Translating support tickets into a product-actionable report
support tickets to analyse
- Summarise what customers struggle with and note the frequency of each struggle
- Assess the impact, evidence-based
- Rank the top three fixes by deflection value
The short report for the product team.
See example AI output
Struggle 1: users can't find the export button (41 tickets/month, blocks a core workflow). Fix rank #1: move export to the main toolbar. Struggle 2: date filters reset on refresh (18 tickets/month).
Estimate Ticket Volume Saved by Prevention
Building a case for a fix by its ticket-deflection value
support tickets to analyse
- Identify tickets preventable by a better help article
- Identify tickets preventable by a product change
- Estimate the ticket volume each specific fix would save
A prevention opportunity table.
See example AI output
Issue: 'How do I change my billing currency?' (60 tickets/quarter). Fix: publish a dedicated help article - estimated to deflect 45 tickets/quarter. Issue: duplicate account creation confusion (25 tickets/quarter). Fix: inline warning.
Design a Five-Question CSAT Survey
Measuring customer satisfaction after a support interaction
no file - no attachment needed
- Cover resolution, effort, tone and outcome with a consistent rating scale
- Add one open comment question
- Keep it short enough to finish in under one minute, unbiased
The five-question survey.
See example AI output
1. Was your issue resolved today? (Yes/Partially/No) 2. How easy was it to get help? (1-5) 3. How satisfied are you with the outcome? (1-5) 5. Anything you'd like to add? (open text)
Self-service and automation
Design a Support Bot Conversation Flow
Scripting a chatbot flow for a common customer request
no file - specify the common request the bot should handle
- Write the bot's greeting and questions to identify intent
- Map the resolution steps with no dead ends
- Define a clean hand-off to a human with context, always offer a human
The conversation flow in words, including the hand-off trigger.
See example AI output
Request: order status. Greeting: 'Hi! I can help track your order - what's your order number?' Hand-off trigger: if the customer replies 'it's late' twice, the bot connects to a human agent with order number, status and complaint history.
Write Grounded Answers for Support Bots
Training a knowledge-agent bot to answer only from approved sources
list of top-ten customer questions
- Write a grounded answer for each question, answering only from provided knowledge
- Cite the source each answer relies on
- Default to a human hand-off when unsure
The answer set, one per question with its source.
See example AI output
Q: What's your return window? A: 30 days from delivery, unopened items only. Source: Returns Policy p.2. Q: Can I get a same-day refund? A: I don't have confirmed information - I will connect you to a person.
Guide a Customer to Self-Serve
Pointing a customer to a self-service option instead of handling it manually
no file - specify the task the customer wants to do
- Point to the exact self-service resource
- Set the expectation for how it works, empowering not dismissive
- Give a fallback if they get stuck
The customer-facing message.
See example AI output
Task: updating payment details. Message: 'You can update your payment details in under a minute - go to Account then Billing then Payment Method. If the page doesn't load, just reply here.'
Write Proactive Onboarding Tips
Pre-empting the mistakes new users typically make
no file - specify the product new users are onboarding into
- Identify the five things new users typically get wrong
- Write one-line guidance that pre-empts each mistake
Proactive tips formatted for an onboarding email.
See example AI output
1. Forgetting to verify email - verify within 24 hours or your account pauses. 2. Skipping the team invite step - invite teammates now, it takes 30 seconds. 3. Using a weak password - use 12+ characters.
Draft a Proactive Outage Notification
Communicating a live outage before customers start asking
no file - specify the outage or issue details
- Acknowledge the issue and state what is affected
- Explain what is being done, honest and calm, no jargon
- Give an ETA and where to get updates
The full notification plus a short status-page version.
See example AI output
Notification: 'We're aware that login is currently failing for some users. Our engineering team is deploying a fix now - we expect this resolved within 45 minutes.' Status-page version: 'Login issue - investigating. Fix in progress, ETA 45 min.'
More prompt packs by function
View the full Prompt LibraryFrequently Asked Questions
Each applies a service framework — HEARD or LARA for de-escalation — and assumes you will attach the ticket, thread or tone guide. That produces fast, empathetic, on-brand replies and real insight from your tickets, rather than generic responses.
For simple, low-risk queries a well-configured agent can respond directly, but keep a human in the loop for complaints and anything sensitive. Start AI-assisted (draft, human sends) before moving to AI-automated. These prompts default to drafting for your review.
Yes — the analysis prompts are built to work from your attached tickets and feedback. Use approved enterprise AI for anything with customer PII, and anonymise where you can under PDPA.
It depends on your helpdesk. Copilot works inside Microsoft tools; Claude and ChatGPT are excellent for drafting and summarising; many teams build a support agent grounded in their knowledge base. These prompts work across all of them.
Yes. AITraining2U's courses are HRD Corp SBL-KHAS claimable for eligible Malaysian employers.
AI drafts the first pass of a reply in seconds, so agents edit and personalise instead of writing from scratch - this typically cuts handling time significantly while replies stay accurate because a human still reviews before sending. The biggest gains come from templated but common request types like shipping delays, refund status and password resets. Pair the reply-drafting prompts with your macros so agents paste, review and send rather than typing from a blank screen. Track average handle time before and after to confirm the AI is actually saving time.
Yes - modern models like Claude, Copilot and ChatGPT draft fluently in Bahasa Malaysia, Mandarin, Tamil and English, and can switch tone and formality appropriately. Paste the customer's message in their own language and ask the AI to reply in the same language. For mixed-language messages, common in Malaysian support tickets, specify which language the reply should be in. Always have a native speaker spot-check tone for culturally sensitive topics, since fluency does not guarantee the right register for every audience.
Generic output usually comes from a vague prompt - give the AI your actual tone guide, real examples of past great replies, and the specific customer context rather than asking for 'a professional reply.' Use the tone-editor and template-editor prompts to explicitly ask for warmth, contractions and one genuine empathy line, and always name the customer. Reviewing and lightly editing every AI draft before sending, rather than sending it verbatim, is the single biggest lever.
Yes - feed the AI historical ticket volume and dates, such as Hari Raya, 11.11 or year-end promotions, and ask it to identify patterns and estimate expected volume for the coming period so you can staff accordingly. It can also flag which ticket categories spike, like shipping delays or payment issues. AI forecasts are a planning aid, not a guarantee - validate against at least two to three prior comparable periods and adjust for any new product launches that break the historical pattern.
AI-drafted replies keep a human in the loop - the agent reviews and sends, which suits complaints, refunds and anything needing judgement. A support chatbot answers customers directly and instantly, suited to simple, repetitive, low-risk questions like order status or opening hours. Most Malaysian SMEs start with AI-assisted drafting because it is faster to implement and lower risk, then graduate high-confidence question types into an automated chatbot once the grounded answers have been tested against real tickets.
Go beyond prompts — train your team
Prompts are the start. AITraining2U runs hands-on, HRD Corp SBL-KHAS claimable AI training for Malaysian teams — from everyday AI productivity to building agents that run customer service workflows end-to-end.