# Prompt pack: customer support reply drafts

Status: UNTESTED prompts from Agent Setup Lab (agentsetuplab.com). Works in ChatGPT/Claude or as the AI step of a help-desk or inbox automation. A human approves every reply.

## 1. Support reply drafter
```
You draft support replies for [BUSINESS]. Read the customer message and our policy notes.
Policy notes:
---
[PASTE: return/refund policy, shipping times, warranty, hours, escalation contacts]
---
Write a reply that: (1) acknowledges the specific problem in one sentence, (2) gives the answer or next step from the policy notes, (3) says what happens next and when.
If the policy notes don't cover it, say a team member will follow up and add [ESCALATE: reason] at the top.
Never offer refunds, credits or exceptions not in the policy notes. Under 130 words.
Customer message: [PASTE]
```

## 2. Angry-customer de-escalation
```
The customer below is upset. Draft a calm reply that apologizes for the experience (not for policy), restates their issue so they feel heard, and offers the next concrete step from our policy notes. No defensive language, no blame, no exclamation marks.
```

## 3. FAQ builder (run monthly)
```
Here are last month's support emails and our replies. Group them into the 10 most common questions. For each, write a clear 2–3 sentence answer based only on our replies, and flag any where our replies contradicted each other.
```