For Customer Service Representative (In-House)s ·
What you'll accomplish
Refund request, shipping delay, account issue, escalation: each case type gets a tested, effective prompt. Copy it, fill in the specifics, get a quality draft immediately. This guide builds that 10-15 prompt library and keeps it one tab away.
What you'll need
Open Notion (or any notes app) and create a new page titled "My CS Prompt Library."
Structure it with sections for each case type you handle regularly. Starter sections:
Think about the case type you handle most often. For most CSRs, this is "general customer inquiry email response." Start here because you'll get the fastest payback.
Open ChatGPT or Claude in another tab.
For each case type, try 2-3 variations of a prompt. You're looking for a prompt that:
Example test for email response:
Try: "Draft a professional customer service email response. Issue: [describe]. Resolution: [describe]."
Check: Does the output sound right? Too formal? Too casual? Adjust the prompt: "Draft a warm, professional customer service email, not too formal. Issue: [describe]. Resolution: [describe]."
Keep testing until you find the phrasing that reliably produces good output.
Once you find a prompt that works, paste it into your Notion page under the right section.
Format each entry like this:
## Refund Confirmation Email
Use when: Customer's refund was processed, you're confirming it
PROMPT:
Draft a professional email confirming a customer's refund. Details: Amount: [X]. Processing date: [date]. Timeline for bank to show it: [3-5 business days]. Tone: reassuring, brief.
NOTES:
- Add customer name before sending
- If amount is over $100, add "If you have any questions about this refund, please don't hesitate to call us directly"
Cover these core cases:
The key to using your library is making it frictionless. Keep it open in a browser tab all day. When you need a prompt, switch to the tab, find the case type, copy the prompt, paste it into your AI tool, fill in the brackets, and get your draft.
For each one: test in ChatGPT/Claude, refine, save the winning version.