AI Prompts for Customer Service: Personalized Replies That Don't Sound Robotic

Why generic AI replies backfire

Customers can tell when a reply is a template. Generic AI text apologises vaguely, repeats the question and promises nothing specific, which often makes an annoyed customer angrier.

The fix is to give the model context: who you are, how you talk, what happened and what you can actually offer.

Set up your brand voice once

Save a short voice brief and paste it at the start of every session, or store it as custom instructions.

• Prompt: You are a customer service agent for [business], which sells [product]. Our tone is friendly, direct and human. Use short sentences, the customer's name, and no corporate phrases like 'we apologise for any inconvenience'. Never promise refunds, discounts or dates unless I give them to you.

Core reply prompts

Use these with the customer's message pasted in, with personal data removed where you can.

• Personalized reply: Here is a customer message: [paste]. Facts: [order status, policy]. Write a reply under 120 words that answers their exact question and ends with one clear next step.
• Angry customer: This customer is frustrated: [paste]. Acknowledge the specific problem in the first sentence, explain what we will do, and avoid defensive language.
• Refund within policy: Draft a reply confirming a refund of [amount] for [reason], processed in [timeframe].
• Saying no kindly: Our policy does not allow [request]. Write a warm reply that explains why and offers [alternative].
• Follow-up: Write a short check-in for a customer whose issue was resolved [X] days ago.

Turn tickets into FAQ and macros

Paste ten recent questions (anonymised) and ask: 'Group these into themes and draft a clear FAQ answer for each.' Then turn the best replies into saved macros in your help desk so the whole team benefits.

Review AI-drafted FAQs against your real policies before publishing them.

Rules for using AI safely in support

AI drafts, people decide. Check every fact, amount and date, and never let the model invent policies. Avoid pasting payment details, passwords or sensitive personal data into general chatbots; use tools your business has approved for customer data.