Good chat examples for customer service are short, one-question-at-a-time messages that greet the customer, ask for the order number once, say how long a hold will take, hand over with a summary so nobody repeats themselves, answer product questions from real catalogue facts and close with a clear next step. Below are 14 copy-ready lines and 3 annotated transcripts.
They are written for an online store: orders, deliveries, returns and product questions. Use them for your own team, or as a yardstick when you review what an AI says in your chat. For the wider set of reply formats, see our templates hub.
Chat examples customer service teams can copy: 14 scripts
These customer service chat scripts cover the six moments of almost every store chat: the greeting, asking for the order number, putting someone on hold, transferring to a colleague, answering a product question and closing. Each line sits in a text block with [placeholders]. Keep every message to two or three short sentences.
If you draft with ChatGPT, our ChatGPT prompts page has the prompts. This page has the finished lines.
Greeting
1. Hi [first name], you're chatting with [agent name] at [store]. What can I sort out for you today?
2. Hi [first name], I can see your order [#order number] from [date]. Is this about that order?
3. Hi, I'm [store]'s AI assistant. I can check orders, returns and product questions. Type "person" at any time to reach the team.Line 3 discloses the AI and says how to reach a person. Salesforce reports that 72% of customers say it is important to know if they are communicating with an AI agent, and the EU AI Act sets a duty from 2 August 2026 to inform people that they are interacting with an AI system, unless that is obvious. That is not legal advice, so check it against your own set-up.
Asking for the order number
1. Could you share your order number? It starts with # and is in your order confirmation email. I'll use it to check where your parcel is.
2. No problem if you can't find it. Could you give me the email address you used at checkout instead?Ask once, say why you need it and say where to find it. Never ask for card details (see the mistakes table below).
Putting the customer on hold
1. I'm checking this with our warehouse now. It will take about [2] minutes. I'll message you here as soon as I have it.
2. Thanks for waiting. I'm still waiting on [carrier / warehouse]. I'll update you by [time], or I can email you the answer if you'd rather not wait.The second line is the check-back for when the answer takes longer than you promised. For targets on how fast a first reply should be, see our guide to response times.
Transferring to a colleague
1. I'm passing you to [name] from our returns team. I've shared your order number and the photos, so you won't need to explain again.
2. Our team is back at [time]. I've added your order number and the issue to your conversation, and [name] will reply here or by email.Nielsen Norman Group’s chatbot study found that bots were generally not able to take advantage of previously entered information when a new task started, and that users viewed an “escape hatch” favourably. A transfer with a summary gives you both.
Answering a product question
1. The [product] runs [true to size / slim]. At [height / usual size], most customers pick [size]. Want me to check stock in that size?
2. The difference between [A] and [B] is [fabric / feature]. If you mainly need [use], I'd go for [A].
3. [Product] in [size] is out of stock right now. [Alternative] is the closest match and is in stock in your size.Fill these placeholders only from your catalogue: the size guide, the product description and live stock per variant.
Closing the chat
1. So that's your exchange to [size] booked. You'll get a return label by email within [time]. Anything else I can help with?
2. I haven't heard back, so I'll close this chat for now. Message us any time and we'll pick it up from here.Recap the outcome and the next step, then invite a reply. If the customer has gone quiet, check once, politely, then close and say how to pick the chat up again.
What do chat support examples look like from start to finish?
Single lines are only half the job. These three short transcripts show a full chat from greeting to close. Each one comes from a fictional store and is written as an illustration, not as a real customer conversation. The notes after each transcript explain why the lines work, so you can reuse the pattern. For more phrasing across channels, see our conversation examples.
Transcript 1: where is my order (partly shipped)
Customer: Hi, where is my order? I ordered two jackets.
Agent: Hi Sanne, I'm Maya. Could you share your order number? It's in your confirmation email and starts with #.
Customer: #1042.
Agent: Thanks. One jacket has shipped and the other is still in our warehouse. Here is the tracking link for the first one: [link].
Agent: The second jacket is due to leave on [date]. I'll email you the tracking link as soon as it does.
Customer: OK, thanks.
Agent: You're welcome. Anything else I can help with?| Line | Why it works |
|---|---|
| ”Could you share your order number?” | Asked once, with where to find it |
| ”One jacket has shipped and the other is still in our warehouse” | States the real status, not a blanket “your order has shipped” |
| Tracking link plus a dispatch date for the rest | Gives what is known and a date for what is not |
| ”I’ll email you the tracking link” | Ends with a next step and who owns it |
Shopify’s order statuses define a partially fulfilled order as one where only part of the items have been fulfilled, and fulfilled as all items shipped. Orders also carry more than one status (order, payment, fulfilment), so check the right one before you answer. This is the classic WISMO chat.
Transcript 2: a product question on the storefront
Customer: Which size should I take in the Bergen coat? I'm 1.78 m and usually a medium.
Assistant: Hi, I'm the store's AI assistant. The Bergen runs slim. At 1.78 m and usually a medium, most customers pick a large for layering. Want me to check stock?
Customer: Yes please. Do you have it in navy?
Assistant: Navy in large is out of stock right now. Charcoal in large is in stock and has the same cut. Would that work?
Customer: Charcoal is fine.
Assistant: Good choice. You can add it from this page. Type "person" if you'd like a colleague to help with anything else.| Line | Why it works |
|---|---|
| Opens by saying it is an AI assistant | Disclosure up front, with a way to a person at the end |
| Size advice from fit notes and the customer’s usual size | Answers from catalogue facts, not a guess |
| One alternative when navy is out of stock | One recommendation, not five |
| No discount code, no pressure | Stays in the brand’s tone and lets the shopper decide |
This transcript is an illustration of what a good assistant says, not a recorded Engaige conversation. It is the pre-sale side of chat: product advice given while the shopper is still on the page.
Transcript 3: a return request handed to a colleague
Customer: My mug arrived broken and I want to send it back.
Agent: I'm sorry about that. Could you share your order number and a photo of the damage?
Customer: #2087. Photo attached.
Agent: Thanks. You're inside our 30-day return window. Would you prefer a replacement or a refund?
Customer: A refund, please.
Agent: I'm passing you to Jonas from our returns team. I've shared your order number, the photo and your choice of a refund, so you won't need to explain again. You'll hear from him here today.
Jonas: Hi Sanne, I have your order #2087, the photo and your refund request. I'm booking the return label now.| Line | Why it works |
|---|---|
| Order number, photo and preferred outcome in one request | Collects everything once |
| The return window checked against policy | No promise your policy cannot keep |
| ”I’ve shared your order number, the photo and your choice” | A warm transfer: the customer repeats nothing |
| The colleague opens with the summary | Proves the handover worked |
| ”You’ll hear from him here today” | States when the customer will hear back |
If the outcome is a refund, name the status correctly: in Shopify, “Partially refunded” means the amount returned is less than the full amount the customer paid. Our returns page covers the policy side. For the idea behind all this, see conversational support.
What is a chat transcript in customer service?
A chat transcript in customer service is the written record of a chat between a customer and an agent or a bot, kept as a copy for the customer, as history for a colleague and as training material. For conversation-format examples across channels, see our conversation examples; this page covers the chat scripts.
Transcripts hold personal data and sometimes order details. Never let a customer paste a full card number into chat, and never ask for one. If one arrives, follow your payment provider’s guidance and do not quote it back.
For training, review real transcripts against your scripts. The places where agents drift from the line, or where customers ask again, show where your lines break. Engaige AI runs a comparison like this between what the agent suggested and what your team sent, and proposes policy changes for you to approve.
How do you adapt these chat scripts to your store?
A chat script is only as good as the facts you fill into it. Before your team uses these lines, decide where each placeholder comes from, which policy it depends on and how it should sound in your brand’s voice. The table maps each placeholder to its source and shows what goes wrong when someone guesses.
| Placeholder | Where the value comes from | What goes wrong if it is guessed |
|---|---|---|
| [order number] | Order confirmation email or the order in your store admin | You discuss the wrong order |
| [fulfilment status] | The order’s status: unfulfilled, partially fulfilled or fulfilled | You promise a parcel that has not left |
| [return window] | Your returns policy | You promise a return outside policy |
| [size advice] | Product page size guide and fit notes | Wrong size, then a second return |
| [stock in that size] | Live inventory per variant | You recommend something that cannot ship |
| [tone] | Your brand’s voice guide | The chat sounds like a different company |
Tone is a set of small choices. Decide these once and write them down:
- Whether you use first names.
- Whether you use contractions.
- How formal the sign-off is.
- Whether you use the customer’s name once or throughout.
For wider tips on reply quality, see our guide to better replies.
Which chat mistakes should you avoid?
Most bad chats fail on the same few habits: asking the customer to repeat themselves, sending walls of text, going silent during a hold, pretending a bot is a person, and promising what your policy does not allow. Each one has a script rule that fixes it. The table pairs the mistake with the fix.
| Mistake | Fix |
|---|---|
| Asking for the order number twice | Ask once and pass it on at every transfer, because bots were generally not able to reuse information users had already entered when a new task started |
| Long paragraphs | Two or three sentences per message, because chat windows show only a very small amount of information at a time |
| Silent holds | Say what you are checking and when you will be back |
| Hiding that the customer is talking to an AI | Say so in the greeting and offer a person (see the greeting section for the research and the EU rule) |
| No way out of a bot | Offer a person at any time, as in the transfer section |
| Asking for card details | Never. The PCI Security Standards Council says unprotected card numbers must not be sent over chat |
| Staying on script when the customer does not | Let the customer go off-script. The same Nielsen Norman Group study found problems occurred as soon as users deviated from the prescribed script |
The card rule comes from the PCI FAQ: chat counts as an end-user messaging technology, and PCI DSS Requirement 4.2.2 prohibits sending unprotected card numbers over it.
Where does Engaige fit?
Engaige (that’s us) isn’t a chat app or a script library: it’s the e-commerce AI support agent that works inside the help desk your chats land in. Scripts help your team write the next message; Engaige answers in the chat itself. Engaige AI learns from your team’s replies and you approve every change. At MR MARVIS, Engaige handles over 120,000 questions a year on chat and email and resolves over 60% of the conversations it’s involved in.
You go live in Agent Assist in hours. It drafts replies inside your help desk, and nothing closes without a human. Every reply your team sends becomes training data, and Engaige AI proposes policy changes for you to approve.
Then you flip ticket types to autonomous one at a time, most brands in about a week. Going autonomous from day one is also supported. The replies your team actually sends, not a static script, are what the agent learns from.
The data that fills the placeholders comes from the store. On Shopify, Engaige combines live Admin API data, pricing across markets and SKUs, inventory, orders and customers, with the marketing content from your storefront. So [fulfilment status] and [stock in that size] come from real data, not memory. For tone, Engaige reads the marketing content on your storefront for your brand’s voice.
The same agent does a second job: pre-sale chat. Engaige also gives product advice to shoppers in a chat widget on your storefront, using live catalogue data and your brand’s marketing content. We call it an AI shopping assistant.
MR MARVIS uses Engaige for personal product advice 24/7. Our product advice page has the detail.
One thing that impresses me every day is the quality of the product recommendations.
When a chat needs a person, the agent hands back to your team. At MR MARVIS, the AI gathers context before a human agent steps in. One thing to know: Engaige is e-commerce-specialised, not a generic chat tool, and text-first by design (email, chat, WhatsApp, SMS and social DMs, no voice AI).
That is the short version. The full picture is in the templates hub. For where an agent fits next to people, read AI agents, and for the cost side, support ROI. To see it on your own chats, book a demo.
Frequently asked questions
What is a good opening line for customer service chat?
Say who you are and ask how you can help: “Hi [first name], you’re chatting with [agent name] at [store]. What can I sort out for you today?” If the customer has an open order, name it up front. If it is an AI, say so and offer a person.
How do you ask for an order number in chat without sounding robotic?
Ask once, say why and say where to find it: “Could you share your order number? It starts with # and is in your order confirmation email.” If they cannot find it, ask for the email address used at checkout.
What should you say when putting a customer on hold in chat?
Say what you are checking and roughly how long it will take, then message them when you have it. If it takes longer, check back with a time, or offer to email the answer.
How do you transfer a chat to a colleague without the customer repeating themselves?
Share the order number, photos and the customer’s wish with the colleague, and tell the customer you have done so. The colleague should open with that summary.
What is a chat transcript in customer service?
It is the written record of a chat between a customer and an agent or a bot. For conversation-format examples, see our conversation examples.
Can you ask for card details in live chat?
No. The PCI Security Standards Council says unprotected card numbers must not be sent over chat, so never ask for one and never quote one back.
Should an AI chat assistant say it is an AI?
Yes. Customers expect it, and the EU AI Act sets a disclosure duty from 2 August 2026 unless it is obvious. This is not legal advice. Say it in the greeting and offer a way to reach a person.
Can an AI agent answer product questions in chat?
Yes. Engaige gives product advice to shoppers in a storefront chat widget, from live catalogue data and your brand’s marketing content. MR MARVIS uses it for personal product advice 24/7.
Do I need chat scripts if I use an AI agent?
Yes, for your team and as a yardstick for what the agent says. Engaige’s behaviour comes from your policies and from learning what your team sends, and you approve every change.
