By the Onsites AI team · Last updated · 6-minute read
Ask a support person where the day goes and the honest answer is the same everywhere: forty percent of their typing is text they have typed before. The greeting, the "one more detail, please", the refund explained eleven ways, the delay apology, the "just checking if this solved it". Canned responses — saved templates you insert and lightly edit — give that forty percent back. This guide lists the twenty worth saving from week one, with the one rule that keeps them from reading like robots.
A template is a starting point, not a stamp. The mechanics that keep them human: every template has one blank line you always rewrite — the reference to what the customer actually said. "Thanks for the order details" is a template. "Thanks — checked with the warehouse about the 20ft reefer" is a person. Save the bones, hand-write the one sentence that proves a human read the thread. Teams that skip the blank line get the "dear valued customer" tone-deafness that customers screenshot; teams that keep it get twice the speed with none of the damage.
The second rule: templates are descriptive, not prescriptive. They encode your best current answer to a recurring question — when policy changes, update the template the same day, or the desk keeps apologizing for a price you no longer charge. Review the library quarterly and delete the ones nobody uses; a short library people trust beats a long one people ignore.
1 · Greeting / arrived. "Hi {name} — thanks for writing in, this is {agent}. I see {one-line summary of their issue}. Let me get you an answer." Sets ownership instantly and proves a human read it.
2 · Need one detail. "Happy to dig in — could you send {the one thing missing}? With that I can give you a definite answer rather than a guess." Asking for exactly one thing gets replies; asking for three gets silence.
3 · Looking into it. "Thanks for your patience — I'm checking this with {the team/source} now and will come back to you by {time}." The time is the important part; vague "soon" reads as never.
4 · Confirming resolution. "That should do it — could you confirm it looks right on your side? If anything's off in the next day, reply here and it comes straight back to me."
5 · Out of scope, with a door open. "That one sits outside what we support, but here's the path that does cover it: {link or short path}. If you get stuck anywhere along it, write back here and we'll take a look."
6 · Repeat customer greeting. For threads that reopen after a close: "Hi {name} — good to hear from you again. Picking up where we left the {topic} conversation..." The tag history from your queue tells you this is a repeat.
7 · Refund approved. "Done — the refund of {amount} is on its way to {payment method}. Depending on your bank it lands in {2–5 business days}. The original charge of {amount} was {one-line cause}." Precision reads as competence; vagueness reads as hiding.
8 · Refund declined, with the reason and the alternative. "Here's what our policy says for {case}: {one honest sentence}. What I can do instead: {credit, replacement, or the specific alternative}. Want me to set that up?" The refund playbook has the full wording for each path.
9 · Quote ready. "Hi {name} — the quote is ready: {items, total, validity date}. Any line you'd like adjusted, say the word and I'll re-cut it." The chat-to-quote guide covers turning the conversation into the document.
10 · Quote follow-up. "Checking the quote I sent on {date} — any questions on the {item} line? If the numbers work, I can move it straight to an invoice." The follow-up cadence covers the whole sequence.
11 · Invoice overdue. The friendly version: "Small nudge — invoice #{number} from {date} shows unpaid on our side. If it's paid, ignore me; if the payment is stuck anywhere, tell me and I'll chase it with you." Firm and final versions live in the overdue invoice guide.
12 · Price update notice. "A quick heads-up before it hits you: from {date}, {what changes}. Your current rate holds until {date}. Here's why we're doing it: {two honest sentences}." The price increase guide has the templates.
13 · Shipping delay, proactive. "You should hear it from us before you spot it: {shipment} is running late — {carrier} shows {new ETA}. Sorry about the wait; here's what we're doing: {action}. And here's {compensation/alternative} to make it right." Announced delays land softer than discovered ones — the delay playbook covers every carrier scenario.
14 · Out of stock, with an alternative. "{Product} is out of stock until {date}. The two options: hold your order and we ship the day it lands, or {alternative product} ships today." The backorder guide has the wording.
15 · Handing off internally. "Bringing {colleague} in on this — they own {area} and can answer properly. {Name}, {one-line summary of where the thread stands}." The customer sees a baton pass, not a ping-pong.
16 · Escalation promise. "This deserves a better answer than I can give — I've escalated it to {owner} with everything you've told us. You'll hear from {them/us} by {time}." The escalation matrix defines who owns what at 2 a.m.
17 · Thank you + small ask. After a closed issue: "Great — glad it's sorted. If you have thirty seconds, a rating on this one helps us more than you'd think (it's how we find the slow spots)." The CSAT-without-a-tool guide turns this into honest measurement.
18 · Reopening a closed thread. "Hi again — you wrote back on {original topic}, so I've reopened the conversation. Give me the {detail} and I'll pick it up from where we left it." New messages on closed threads should reopen automatically; this reply sets the tone for round two.
19 · Feedback received. "Logged as a feature request with the team — honestly useful, thank you. No promise on timing, but I'll tag you here if it ships." Feature requests without this template rot in the queue; with it, they become a public record of listening.
20 · The Copilot-assisted draft. When a reply needs your desk's real context — policy, history, a past quote — let AI draft from your handbook and edit from there. Onsites' Copilot drafts cost 3 credits (about three cents), suggested replies 6, and the draft only sees what you give it, so treat templates 1–19 as its starting vocabulary. The Copilot trust guide covers the review habit that keeps drafts on-brand; the AI cost math guide keeps the meter honest. Start with these twenty saved by hand — the AI then drafts in the style your templates already set, and every reply leaves the desk faster than it arrived.
Three maintenance rules. Owner the library: one person owns the canned responses; everyone else proposes edits. Version the changes: when a template moves from "we usually say" to "our policy", the edit lands in the same week and the desk is told. Prune quarterly: anything unopened for a quarter gets deleted or rewritten, because a stale template is worse than no template — it answers a question nobody is asking with a policy nobody has anymore. Save these twenty in Settings on day one of your free ticketing setup; they cost nothing, and they are the difference between a desk that answers and a desk that replies.
What are canned responses?
Saved reply templates an agent inserts with one click and lightly edits — greetings, clarifications, refund explanations, delay notices, quote follow-ups. They cut forty percent of the typing out of a support day while keeping replies consistent.
How do you keep templates from sounding robotic?
Keep one hand-written line in every reply that references what the customer actually said, update templates the same day policy changes, and prune anything unused for a quarter. The template carries the structure; the human carries the specifics.
How many canned responses should a new team start with?
Around twenty covers most of the recurring volume: greeting, need-a-detail, refund approved and declined, quote ready and follow-up, invoice overdue, shipping delay, out of stock, escalation, thank-you and re-open messages.
Should saved replies be used for pricing and refunds?
Yes — those benefit most, because the wording must be exact and legally consistent. Link each template to the underlying policy internally so agents explain decisions rather than recite them.
Do canned responses work with AI drafted replies?
Yes — AI drafts inherit the tone and structure of the saved templates. Written templates define the desk's voice; Copilot drafts within it at 3 credits per draft, and a human approves before anything is sent.