PromptBro

ChatGPT Prompts for Customer Service

Customer service teams are under pressure to respond faster, stay consistent, and actually retain customers — not just close tickets. These prompts are built around real support scenarios: the angry refund request, the escalation you didn't see coming, and the knowledge base article nobody ever wrote. Each prompt gives the AI enough context to produce a draft you can actually send or adapt, not a generic template you'll rewrite from scratch.

Example Prompts

Refund response that retains the customer

You are a senior customer experience specialist who has handled thousands of refund requests for e-commerce brands. Your goal is always retention first — a well-handled refund often creates a loyal customer.

A customer named Priya purchased a $120 skincare set 22 days ago. She's requesting a refund because she says it caused a breakout. Our policy is 30-day returns for unopened items only, and hers is clearly opened and used. She is a repeat customer with 4 previous orders totaling $390.

Write a response email that:
- Acknowledges her experience with genuine empathy (not scripted sympathy)
- Explains our return policy clearly but without making it feel like a wall
- Offers a goodwill resolution: 50% store credit on her next order plus a free consultation with our skincare specialist (a real service we offer)
- Does not admit product fault (we cannot verify the reaction was caused by our product)
- Ends with a clear next step for her to take

Tone: warm, direct, and human. Under 220 words.

Escalation script for an angry caller

You are a customer service trainer who teaches de-escalation techniques at contact centers. You understand that an angry customer is usually expressing a need for control, acknowledgment, or urgency — and that matching their energy makes things worse.

A caller named Daniel has been on hold for 34 minutes trying to resolve a shipping error — his $340 order was delivered to the wrong address and he needs it for an event tomorrow. He's now speaking loudly, using phrases like "this is a complete joke" and "I want to speak to someone who actually cares." This is his second call about the same issue.

Write a call script for the agent that:
- Opens with a de-escalation phrase that acknowledges the wait and the problem without being defensive
- Gives the agent language to validate Daniel's frustration without accepting blame prematurely
- Provides a clear resolution path: emergency re-ship or same-day local pickup options
- Includes a branching note for what to say if he continues to escalate vs. if he calms down
- Ends with what to offer as goodwill (a $30 credit regardless of outcome)

Format as an annotated script with [Agent says:] and [If customer says X:] structure.

FAQ generation from a product description

You are a technical writer who specializes in creating customer-facing FAQ sections for software products. You write from the customer's perspective, not the product team's.

Here is the product description for our new feature, "Smart Scheduling" in our appointment booking app:

"Smart Scheduling uses AI to analyze your business's historical booking patterns, staff availability, and service duration data to automatically suggest optimal appointment slots to customers. It integrates with Google Calendar and Outlook. The feature is available on Pro and Business plans. Setup requires connecting at least 30 days of booking history."

Generate a FAQ section with 8 questions and answers that:
- Are written from real customer pain points, not marketing copy
- Cover: what it does, who it's for, setup requirements, plan availability, privacy concerns about AI analyzing data, what happens if you have less than 30 days of history, how it interacts with existing manual blocks, and how to turn it off
- Are written in plain English (no jargon, no internal product terminology)
- Each answer is 2–4 sentences — enough to be complete, short enough to scan

Format as Q: / A: pairs.

CSAT follow-up email after ticket resolution

You are a customer success manager who writes follow-up emails after support interactions. Your goal is to measure satisfaction without feeling transactional, and to catch dissatisfied customers before they churn silently.

Context: A customer named Sophie submitted a support ticket 4 days ago because she couldn't export her data in CSV format. The issue was a bug on our end — our engineering team fixed it yesterday and her export now works. Her ticket was open for 4 days, which is longer than our 24-hour SLA. She did not respond to our first reply asking for more details.

Write a follow-up email that:
- Confirms the issue is resolved and what specifically was fixed (the export bug)
- Acknowledges that the 4-day resolution time was longer than it should have been
- Includes a 1-question CSAT survey with a direct link placeholder: [SURVEY LINK]
- Leaves an open door if she has follow-up questions
- Is signed from a real person (Customer Success team, not a no-reply address)

Tone: professional but personal. Do not use phrases like "valued customer" or "we take pride in." Under 150 words.

Internal knowledge base article from a support ticket

You are a technical writer converting real customer support tickets into internal knowledge base articles for a support team. Your articles are used by agents who need to resolve similar issues quickly without escalating.

Here is a condensed summary of a resolved support ticket:

Issue: Customer could not connect their Shopify store to our analytics platform after updating to Shopify API version 2024-01. Error message shown: "OAuth token invalid — re-authentication required."
Root cause: Our OAuth integration was still using the deprecated 2023-07 API version. Shopify deprecated this version in January 2024. Any store that upgraded to the 2024-01 API automatically invalidates old OAuth tokens.
Resolution: Customer needs to disconnect and reconnect their Shopify integration in Settings > Integrations > Shopify. This generates a new OAuth token on the current API version. Estimated time: 3 minutes.
Affected users: Any customer who updated their Shopify store to API version 2024-01 or later.

Write a knowledge base article that includes:
- Title (clear and searchable)
- Affected versions / conditions
- Symptoms (what the customer sees)
- Root cause (one sentence, for agent context)
- Step-by-step resolution instructions (numbered)
- How to identify if a customer is affected (what to ask / what to check in admin)
- A note on whether a refund or goodwill gesture is warranted (it is not in this case)

Format for internal use — agents will read this during a live call.

Tips for Customer Service Prompts

Customer service prompts fail when they don't include the emotional context of the situation. AI models default to neutral, policy-compliant language — which can sound robotic when a customer is genuinely frustrated. Always describe the customer's emotional state (frustrated, confused, upset but polite) and the history of the interaction. A second-contact angry customer needs a different response than a first-time confused one.

Be explicit about what you cannot say. If you can't admit fault, can't guarantee a timeline, or can't exceed a certain compensation threshold, put that in the prompt. Without constraints, AI models often generate responses that sound great but commit you to things your policy doesn't allow. "Do not admit product fault" and "maximum goodwill offer is a $25 credit" are the kinds of guardrails that make a draft actually usable.

For tone calibration, giving the AI a word count target forces concision. Customer service emails that exceed 250 words rarely get read in full. Specify your target length and the AI will prioritize what matters — it won't pad with pleasantries if it knows it needs to cover the key points in 180 words.

Why use PromptBro for Customer Service?

PromptBro's voice-first flow lets you describe a customer scenario out loud — the situation, the customer's tone, what you can and can't offer — and builds a fully structured prompt in under 60 seconds. The 6-step process makes sure you've specified the expert role, tone, constraints, and output format before generating, so you don't end up with generic templates you have to rewrite anyway.

Try PromptBro free — build your first prompt in 60 seconds →

Related resources