guide
45 Live Chat Scripts, Organised by the Moment That Goes Wrong
Chat scripts fail for reasons email scripts never do
Every list ranking for this query groups scripts the same way: greetings, sales, support, closing. That is how email is organised, and chat is not email.
Chat goes wrong at specific mechanical moments — the silence while you look something up, the transfer, the conversation that will not end. These are grouped by those.
One rule before any of them
Chat messages must be short enough to read in one glance. A six-line paragraph pasted into a chat window reads as a wall and gets skimmed; the same text in an email reads as thorough.
Two lines is the working maximum. If an answer needs more, send it as two or three separate messages a second apart — it reads as a person thinking rather than a macro firing. Everything below is written to that constraint.
Opening (5)
| Situation | Script |
|---|---|
| Standard open | Hi [Name] — I'm [Agent]. What can I help with? |
| They opened with the full problem already | Got it, thanks for laying that out. Give me one second to check the account. |
| Proactive, on a pricing page | Anything I can answer while you're comparing? Happy to just leave you to it too. |
| Proactive, after an error | Looks like something failed on that page. Want me to take a look? |
| Returning customer | Hi [Name], good to see you again. Is this about [previous topic] or something new? |
The proactive ones matter more than they look. A proactive chat that opens with can I help you today is ignored; one that names what the person is doing gets a reply, because it demonstrates the window is attached to something.
The silence (6)
The single most common failure in live chat. You go to look something up, thirty seconds pass, and the customer cannot tell whether you left. Nothing in the interface distinguishes checking from gone.
| Gap | Script |
|---|---|
| Before any lookup | Let me pull that up — about a minute. |
| At 60 seconds | Still on it. The billing record is loading slowly. |
| At 2 minutes | Sorry, this is taking longer than I said. Two more minutes — I'd rather get it right. |
| At 4 minutes | This is longer than it should be. Want me to email you when I have it, or are you happy to wait? |
| Coming back | Right, got it. [Answer]. |
| Coming back with nothing | No luck from here — I need [team] to check. Here's what happens next. |
Naming a duration is what makes these work. About a minute sets a clock the customer can hold you to, and breaking it with an update at sixty seconds costs you nothing. Silence is what makes people close the window.
Asking for things (5)
| Need | Script |
|---|---|
| Order number | What's the order number? Starts with # and it's in your confirmation email. |
| Account email | Which email is the account under? It might not be the one you're chatting from. |
| Screenshot | Could you send a screenshot of what you're seeing? Paperclip icon below. |
| Something they already gave | Sorry — I have [X] but not [Y]. Could you send that one too? |
| Something sensitive | I can't take card details here. I'll send a secure link instead — one moment. |
The last one is not optional. Chat transcripts are stored, searchable and frequently visible to more people than anyone assumes, and a customer will paste a full card number if you do not stop them first.
Not knowing (4)
- "I don't know — let me find out rather than guess." Says the quiet part and buys real credit.
- "That's outside what I can see from here. [Team] handles it and I'm bringing them in."
- "I want to check before I answer, because I've seen this go both ways."
- "Honest answer: I'm not sure that's supported. Give me two minutes to confirm before you plan around it."
Confident wrong answers cost more in chat than anywhere else, because the customer acts on them immediately. The window is open, they are at the keyboard, and by the time you correct it they have already done the thing.
Transfers and escalation (6)
| Situation | Script |
|---|---|
| Standard transfer | This needs [team] — transferring you now. They'll have the whole conversation, so no need to repeat any of it. |
| Transfer with a wait | [Team] can fix this. There's a queue of about [N] minutes — want to wait or shall I have them email you? |
| No one available | The team who handles this is offline until [time]. I'll write it up with everything you've told me and they'll pick it up first thing. |
| Customer asks for a manager | Of course. Before I bring them in — is there anything you want me to put in front of them first? |
| Second transfer | You've been passed once already and I'm sorry. I'm staying on this one until it's with the right person. |
| Escalating without being asked | This is past what I can do from here. I'm escalating it rather than making you ask. |
They'll have the whole conversation is the line that does the work, and only if it is true. Making someone re-explain after a transfer is the single most reliable way to turn an inconvenience into a complaint.
When they are annoyed (5)
- "That's a fair thing to be annoyed about. Here's where it actually stands."
- "You're right that this shouldn't have happened. I can't undo it — here's what I can do."
- "I'm not going to tell you it's fine. Let me tell you what I'm doing about it."
- "Three days without a reply isn't acceptable. I have it now and I'm not passing it on."
- "I'd be annoyed too. Give me five minutes to get you a real answer rather than an apology."
None of these begin with I understand how you feel. In chat that phrase arrives instantly, which makes it read as automated, and a customer who has been on hold twice this week recognises it as the thing that comes before nothing happening.
Saying no (5)
| Situation | Script |
|---|---|
| Outside policy | I can't do that one — it's outside what we allow. What I can do is [alternative]. |
| Feature that does not exist | We don't have that. Most people in your position use [workaround] — want me to walk you through it? |
| Refund outside window | The window closed on [date] so I can't process it as a refund. I can offer [alternative] instead. |
| Repeated ask | I've checked again and the answer is still no — I'm sorry. I'd rather be straight than keep you here. |
| No, and it is our fault | We can't do that, and the reason is on us. Here's what I'm offering instead. |
Say the no first. Chat rewards front-loading — a customer reading three lines of context before the refusal will skim to the end anyway, and the delay reads as evasion rather than care.
Ending (5)
| Situation | Script |
|---|---|
| Resolved | That's sorted. Anything else while I'm here? |
| Resolved, with a follow-up | Done. You'll get a confirmation email in about ten minutes — if it doesn't arrive, reply to this chat. |
| Not resolved, handing off | Not fixed yet, and I don't want to pretend otherwise. [Team] has it and you'll hear by [time]. |
| Customer has gone quiet | Still there? I'll leave this open for a few minutes in case you're mid-something. |
| Closing an abandoned chat | Closing this one for now — everything's saved, so replying reopens it with the full history. |
The four scripts to delete
These appear on most ranking lists for this query and all four do damage.
| Script | Problem |
|---|---|
| "Is there anything else I can help you with today?" after not helping | Reads as a close attempt. Ask it only after resolving something |
| "I completely understand your frustration." | Recognised as scripted, and it arrives too fast in chat to be believed |
| "Unfortunately, that is our policy." | Names the rule, not the reason, and offers nothing |
| "Thank you for your patience." during a long wait | Thanks them for something they had no choice about |
Where these go once you have AI
Chat is the channel automation reaches first, and the openings, the lookups and the routine noes are exactly what an AI layer will take. The silence scripts stop mattering — a machine does not go quiet to check something.
What stays human is the annoyed section, the second transfer, and every no where the reason is that you got it wrong. Those are the ones worth rehearsing, and they are the ones every ranking list treats as an afterthought.
Written to be used rather than read. Square brackets are the only things to replace. The grouping by chat mechanics is our own — we have not found another list organised this way, which is the reason this page exists.
Frequently Asked
What are live chat scripts?
Pre-written openings and responses agents adapt in real time. The useful ones are short — two lines maximum — because a six-line paste into a chat window reads as a wall and gets skimmed.
How long should a live chat message be?
Two lines. If an answer needs more, send it as two or three separate messages a second apart, which reads as a person thinking rather than a macro firing.
What do you say when you need to look something up in chat?
Name a duration — "let me pull that up, about a minute" — then update at sixty seconds. Silence is what makes people close the window, because nothing in the interface distinguishes checking from gone.
How do you transfer a chat without annoying the customer?
Say the team will have the whole conversation and no repeating is needed — and make sure that is true. Making someone re-explain after a transfer reliably turns an inconvenience into a complaint.
What phrases should you avoid in live chat?
"I completely understand your frustration", "unfortunately that is our policy", "thank you for your patience" during a wait, and asking if there is anything else after not helping with the first thing.
How do you say no in a live chat?
Say the no first, then the alternative. Three lines of context before a refusal gets skimmed anyway, and the delay reads as evasion.
What do you say to an angry customer in chat?
Name what happened rather than the feeling. "Three days without a reply isn't acceptable" lands where "I understand how you feel" does not, because the second arrives instantly and reads as automated.
Should agents admit they do not know something?
Yes, and faster in chat than anywhere else. Customers act on chat answers immediately, so a confident wrong answer has usually been acted on before you can correct it.
How do you close a live chat?
State whether it is resolved. If it is not, say so and give a team and a time. "Anything else?" belongs only after you have actually fixed something.
Which chat scripts will AI take over?
Openings, lookups and routine refusals. What stays human is the annoyed customer, the second transfer, and every no where the reason is that you got it wrong.
Can you ask for card details in a live chat?
No. Transcripts are stored, searchable and visible to more people than anyone assumes. Send a secure link, and say so before the customer pastes anything.
Tools Mentioned
Full reviews, pricing tiers and where each one breaks.
Tidio Lyro
One of the few genuinely cheap autonomous agents, bundled with live chat and a basic help desk.
Front
Treats support as team email rather than tickets, with AI on top as drafts and summaries.
Help Scout
Chosen for simplicity rather than power. AI drafts and summaries, not an autonomous agent.
Re:amaze
GoDaddyHelpdesk and live chat for e-commerce brands, now part of GoDaddy's portfolio.
You Can Also Look Into
70 Customer Service Email Templates (Free, No Signup Required)
Seventy templates you can send after changing the bracketed parts. Grouped by share of your queue, with a note on which half an AI can handle.
40 Canned Responses You Should Delete From Your Help Desk
The phrases that read as scripted, the ones that quietly commit you to things, and the four that are actively costing you customers.
How to Respond to an Angry Customer (by What They Are Actually Angry About)
Four kinds of anger, each needing a different opening. Plus the six phrases that escalate, and what to do when the customer is wrong.
12 Customer Service Metrics That Matter (and 8 That Do Not)
Which numbers change a decision, which are vanity, and the four that get gamed the moment you put them on a wall.
WRITTEN BY AR · UPDATED 2026-08-04
I read the fine print. Vendor pricing pages, billing definitions, terms, funding filings and acquisition notices — then I do the arithmetic nobody publishes: what a platform actually costs at your volume, what its headline metric is really counting, and who owns it now. I do not run benchmarks, and no page here pretends otherwise.