10 Chatbot Apology Scripts for a Service Mistake

Where the business sent the wrong thing, the customer should not be asked to do anything difficult to put it right.

4. Wrong item sent

We sent the wrong item, which is our mistake entirely. The correct [item] is going out today and there's a prepaid label in your email for the other one. You don't need to wait for us to receive it.

Not making the replacement conditional on the return arriving is what separates a good recovery from a grudging one.

5. Item missing from an order

The [item] should have been in that delivery and wasn't. I've arranged for it to be sent today, arriving [date]. Nothing needed from you.

Short, specific, and closes the loop without asking the customer to prove anything.

6. Damaged on arrival

That shouldn't have arrived in that condition. A replacement is on its way for [date]. Please don't worry about returning the damaged one unless we ask.

Removing the return burden on a damage claim is standard in most policies and worth stating explicitly.

Money errors need the correction stated with a timeframe, and anything beyond a simple correction should reach a person.

7. Charged twice

You've been charged twice for [what]. I've refunded the duplicate, which takes [timeframe] to show depending on your bank. Reference [number].

Only appropriate where the automation can genuinely issue the refund. Where it cannot, escalate rather than promising.

8. Wrong amount charged

The charge should have been [correct amount] and was [charged amount]. I've flagged this for correction and someone from [team] will confirm the adjustment by [time].

Where the correction requires judgement, committing to a person and a time is better than an automated promise.

9. Service charged after cancellation

You cancelled on [date] and shouldn't have been charged after that. I'm passing this to [team] to refund and check nothing else has gone out. They'll confirm by [time].

This category frequently indicates a systems problem affecting others, which is a reason to escalate rather than resolve quietly.

During a widespread issue, the apology should be honest about what is known and what is not.

10. Service unavailable

[Service] has been down since [time] and we know it's affecting you. The team is working on it and the next update will be at [time]. I'll message here as soon as it's back.

Committing to a next update time is the element that stops customers contacting repeatedly, and it must be honoured.

Not overclaiming during an incident

Promising a resolution time during an active incident produces a second failure when it passes, which is worse than saying the timing is not yet known.

Committing to an update rather than a fix is the honest version and is accepted readily.

Following up afterwards

Messaging affected customers once the issue is resolved closes the loop and prevents trailing contacts from people who never heard.

This is also where a goodwill gesture, if one is offered, lands best.

A handful of phrasings reliably make an apology worse, and all of them are common in automated messages.

The conditional apology

Apologising that the customer feels a certain way, or for any inconvenience caused, avoids acknowledging what actually happened.

Customers recognise this construction immediately and it converts a recoverable situation into a complaint.

Repeated apology

Three apologies in one message communicates discomfort rather than accountability, and it lengthens a message that should be brief.

Once is sufficient, followed immediately by what is being done.

Deflecting to a third party

Blaming a carrier, a supplier or a system reads as avoiding responsibility, since the customer bought from you rather than from them.

Explaining the cause is fine, provided the accountability stays where the customer placed it.

Admitting liability automatically

An automated message accepting fault for something with legal or financial consequence can create exposure the business did not intend.

Acknowledging the customer's experience while routing the liability question to a person is the safe construction.

Several categories should never receive an automated apology, regardless of how well written.

Financial loss beyond a refund

Where a customer has incurred costs beyond the transaction, the remedy requires judgement and an automated apology will understate it.

Acknowledging briefly and routing to a person is the correct handling.

Legal language or formal complaints

Any reference to consumer rights, legal action or a formal complaint should reach a person immediately without an automated apology preceding it.

Continuing automatically at that point reliably escalates the situation further.

Safety, health and vulnerability

Anything involving injury, health, bereavement or a vulnerable customer requires a person, and an automated apology in these situations reads badly whatever it says.

Recognition rules should be conservative, since the cost of an unnecessary escalation here is trivial.

Chatboq shifts register when frustration or a complaint is detected, and escalates the categories that should not be handled automatically.

Register shift on detection

Responses become shorter and plainer when a complaint or frustration is detected, which prevents the cheerful default appearing at the worst moment.

Apology wording is drawn from the same tone configuration as everything else, so it does not read as a separate voice.

Actions attached to apologies

Where the assistant can issue a replacement, refund or rebooking, the apology includes the action rather than promising one, which is the difference customers notice.

Where it cannot, the message commits to a named team and a time rather than an open-ended follow-up.

Escalation configured explicitly

Financial loss, legal language, safety and vulnerability route to a person before any apology is generated, using the same rules the playbook gives agents.

Keeping one list for both means an automated response cannot handle something the playbook says requires a person.

Measuring recovery

Satisfaction and repeat contact after an apology show whether the recovery worked, which is more useful than the apology wording in isolation.

Conversations that received an apology and then escalated anyway are the clearest signal that the script or the remedy was insufficient.