Suggested Replies for Customer Service: 8 Templates That Sound Like You, Not a Bot

Gmail, Outlook and most helpdesks now offer suggested replies for customer service: three grey chips under the email saying "Thanks for reaching out!", "I'll look into it", "Sounds good". They are fast, and they are the fastest way to sound like every other business your customer emailed today. No customer chose you for the sentence "Thanks for reaching out!"

The fix is to build your own set. Eight replies, written once in your voice with your real policies, saved where you can reach them in two clicks, and served up by the ChatGPT or Claude you already pay for whenever a message does not fit one cleanly. This post gives you the eight, the setup, and the monthly check that keeps them honest.

Why built-in suggested replies for customer service fall flat

The chips are generated from the email in front of them, with no idea what you sell, what you promised, or how you talk. So they come out polite and empty. A customer asking whether their order will make it before Friday does not need "I'll check on that"; they need the honest date and what happens if it slips.

Your version has two things the chips will never have: your policies (the return window, the deposit rule, the rain-day rule) and your voice. Both fit in a single document. That document is what you are building.

The eight replies worth writing first

Open your sent folder and count. Most small businesses find that eight situations cover well over half of what they send. These are the eight we see most, each as a template you can paste and edit. Fill the brackets, cut anything that is not true for you, and keep each under 100 words; longer than that stops being a suggested reply and becomes a letter.

1. Order or booking status

Hi [NAME], quick update on [ORDER / BOOKING]: it is currently [STATUS]
and on track for [DATE]. If anything changes on our end I will email
you before you have to ask. Anything else you need before then, reply
here.
[SIGN-OFF]

Send this the moment someone asks, even when the honest answer is "still in production". Silence bothers people far more than a delay does.

2. The delay, before they ask

Hi [NAME], I want to flag this before you have to chase it: [ORDER /
JOB] is running behind. The reason is [ONE HONEST SENTENCE]. New date:
[DATE]. To make up for it we are [WHAT, IF ANYTHING]. Sorry for the
mess; you will hear from me again on [DATE] either way.
[SIGN-OFF]

Sent before the customer notices, these words earn forgiveness; sent after a "where is my order?" email, the same words read as an excuse. If delays are a weekly event, the wording is not your problem, and handling complaints with AI covers what to do when better sentences stop helping.

3. Return or refund instructions

Hi [NAME], happy to sort this. Because [ORDER] is within our [WINDOW],
here is how it works: [STEP 1], [STEP 2], and the refund lands within
[N] business days of us receiving it. If you would rather swap it for
[ALTERNATIVE], say so and I will set that up instead.
[SIGN-OFF]

Every number in this one comes from your written policy, not from memory. If you catch yourself guessing at the window, stop and write the policy down first; a template with a wrong number in it is a refund you did not plan to give.

4. The pre-purchase question

Hi [NAME], good question. [DIRECT ANSWER IN ONE OR TWO SENTENCES]. The
one thing people often do not realise is [THE CAVEAT OR TIP THAT SAVES
THEM A SURPRISE]. If you want, reply with [THE DETAIL YOU NEED] and I
will confirm it fits before you order.
[SIGN-OFF]

This is the reply that quietly sells. The caveat line is the trust move: naming the surprise before they hit it is what a good salesperson across a counter would do.

5. Something arrived wrong or broken

Hi [NAME], that is on us and I am sorry. Could you reply with a photo
of [THE ISSUE] so I can [RESHIP / CREDIT / SEND SOMEONE OUT]? As soon
as it lands I will [ACTION] the same day. You do not need to send
anything back unless I ask.
[SIGN-OFF]

Ask for exactly one thing (the photo) and promise exactly one action. Two asks in a damaged-goods reply doubles the back-and-forth with a customer who is already annoyed.

6. "Can you do it cheaper?"

Hi [NAME], I understand, and I would rather be straight with you than
haggle: [PRICE] is the price for [WHAT IS INCLUDED], and here is why:
[ONE CONCRETE REASON]. If budget is the issue, the option that fits is
[SMALLER PACKAGE / PAYMENT PLAN / DIFFERENT DATE]. Happy to do either.
[SIGN-OFF]

The template holds the price and offers a door. Decide the door once, while you are calm, so you are not inventing discounts at 9 pm.

7. The polite no

Hi [NAME], thanks for asking. That is not something we do, because
[HONEST REASON IN ONE LINE]. The people I would point you to are
[ALTERNATIVE, IF YOU HAVE ONE]. If [THE THING YOU DO OFFER] would help
instead, I am glad to.
[SIGN-OFF]

A no with a reason and a referral gets you remembered. A bare no gets you a one-star review from someone you never even served.

8. The check-in after a problem

Hi [NAME], it has been [N] days since we sorted out [THE ISSUE] and I
wanted to make sure it is holding up. A one-word answer is fine. If
anything is off, tell me and I will fix it.
[SIGN-OFF]

Few businesses send this one, which is why it works. A three-line check-in a week after a problem turns a complaint into a story the customer tells in your favour.

If you want a bigger set, the Support Macro Library prompt generates 25 in one pass, organised by category, with a [CONFIRM POLICY] flag wherever it would otherwise be guessing.

Make ChatGPT serve them in your voice

This is the trick that makes the eight worth more than a doc of canned text. Paste all eight into a Claude Project or a ChatGPT custom GPT, together with a short policy section and two or three replies you were proud of, and add one instruction:

When I paste a customer message, pick the closest template above, fill
it from the message and OUR POLICIES, and give me the reply ready to
send. If no template fits, or the message is angry, involves a legal or
safety issue, or asks for something outside our policies, do not draft.
Say which case it is and what decision you need from me.

Now the suggested reply is yours: your window, your tone, your caveat line, filled in from the message in front of it, in about ten seconds. The refusal clause matters as much as the templates, because the messages that need a human are exactly the ones a generic chip would have answered with "Thanks for reaching out!". The Customer Support Autopilot is the full version of this setup, with the triage rules and internal notes already written, and the ten drafting prompts for customer service cover the situations that never fit a template.

The monthly check that keeps the set honest

Templates rot. A price changes, the return window moves, and the reply set keeps sending last spring's numbers with total confidence. Two habits stop that. When a policy changes, sweep the set the same day; the Knowledge Base Copilot does this for help articles and will hold your reply set too. Then, once a month, paste ten replies you sent (edited templates included) into the Support Reply Auditor and let it grade them against your policies and your voice. The drift it finds is usually small and always a little embarrassing: "sorry for the inconvenience" creeping back into four replies, or a template that quietly stopped matching how you talk.

Where this breaks

A saved reply cannot see your order system, so every date, amount and status still gets typed in by a person who checked. Anything with a legal threat, a chargeback dispute, or a safety issue gets written from scratch by a human, with no template and no assistant draft. And templates handle questions; the customer who has been quietly unhappy for six months needs a phone call, and there is no chip for that.

Start with the three situations you answered most often last week, write those three properly, and use them for a fortnight before adding the rest. When support is more than a Tuesday-night job, the small-team support playbook has the system around these replies.