The best AI prompts for customer service, 23 to copy (2026)

A good AI prompt for customer service tells the tool who is answering, how the customer feels, what you want to achieve and what to avoid. That is what makes the reply sound human instead of robotic. This library collects 23 prompts to copy for ChatGPT, Claude, Copilot and Gemini. To train your team, see the AI program at Founderz.

Add Founderz as a preferred source

What you’ll take from this

This customer service prompt library helps you reply with empathy, save time in the support team and avoid sounding like a template.

• 23 ready-to-copy AI prompts for complaints, delays, technical support, retention, follow-ups and feedback.
• Prompts written for ChatGPT, Claude, Copilot and Gemini, based on what each tool does best.
• The method behind them: the six fields of an empathetic prompt and an example of a weak prompt turned into a strong one.
• How to build them into your support team workflow and where a person is still needed.

ChatGPT prompts for customer service

Many support teams try ChatGPT for an afternoon, get replies that sound like a corporate brochure and go back to writing everything by hand. The problem is almost never the tool, it is the prompt: a one-line instruction produces one-line quality. These prompts are sorted by use case. Copy them, replace the brackets with your details and adjust the tone to your brand. Reviewed in October 2026.

ChatGPT prompts for customer support: complaints and angry customers

These are the moments where a cold reply does the most damage: the key is to acknowledge the emotion before solving.

1. Reply to a complaint

Role: Support agent at [company], with a warm and honest tone.
Context: A customer complains about [problem, for example: the order arrived with a broken box]. They feel [frustrated and unheard].
Task: Write a reply that acknowledges the problem, apologizes without excuses and offers [solution, for example: a free replacement shipped within 48 hours].
Constraints: Do not promise deadlines we do not control. No corporate phrases like "we apologize for any inconvenience". Maximum 120 words.
Output format: Greeting with the name [Name], two short paragraphs and a closing line with the next step.

2. Very angry customer

Role: De-escalation specialist at [company].
Context: A customer writes in capital letters with an aggressive tone about [problem]. Their message: [paste the message].
Task: Acknowledge their anger genuinely before solving, then offer [solution] with clear steps.
Constraints: Do not get defensive or justify mistakes. Do not use "I understand" more than once. Do not state facts we do not have.
Output format: Reply of 4 to 6 sentences and, at the end, one line with the signal the agent should watch for in case the customer keeps escalating.

3. Repeated complaint

Role: Head of customer service at [company].
Context: The customer has already reported this problem [N times] and it is still unresolved. History: [summary of previous contacts].
Task: Write a reply that acknowledges the accumulated frustration, takes responsibility and escalates the case to [owner].
Constraints: Be specific about the next step and the contact date. Do not repeat the answer already sent.
Output format: Email with subject line included, maximum 150 words.

Useful prompts for delays, incidents and technical support: real use cases

4. Delay apology

Role: Support agent at an online store, human tone.
Context: [Customer]'s order is [X days] late because of [cause, for example: a carrier problem].
Task: Write an email that acknowledges the inconvenience, explains the cause briefly and honestly, and offers [compensation].
Constraints: No corporate language. Do not blame third parties by name. Do not promise a new date unless it is confirmed.
Output format: Subject line plus a body of three short paragraphs.

5. Step-by-step technical fix

Role: Technical support for [product] explaining to a non-technical person.
Context: The customer cannot [action, for example: reset their password]. They already tried [what they tried].
Task: Explain the solution in numbered steps in simple language.
Constraints: No jargon. One step per sentence. Close by offering more help without sounding like a template.
Output format: Numbered list of at most 6 steps and one closing sentence.

AI prompts for follow-up, retention and ticket analysis

6. Post-sale follow-up

Role: Customer success agent at [company].
Context: Three days have passed since [customer] received [product or service].
Task: Write a message that asks if everything is going well and offers help.
Constraints: Do not sound like a sales pitch. Do not ask for a review in this message. Friendly, natural tone.
Output format: Maximum 4 sentences, suitable for email or WhatsApp.

7. Retaining an at-risk customer

Role: Retention lead at [company].
Context: A customer showed signs of wanting to cancel [service]: [signal, for example: usage dropped 60% in two months].
Task: Write a message that acknowledges their situation, asks what we can improve and offers [alternative] without pressure.
Constraints: Understanding, never desperate. Do not open with a discount.
Output format: Short email with a single open question at the end.

8. Ticket analysis from a file

Role: Customer service quality analyst.
Context: I am attaching a CSV with last month's tickets: [columns, for example: date, reason, channel, reply, rating]. It contains no personal data.
Task: Group the tickets by reason, count which repeat most and flag the 3 with the lowest ratings.
Constraints: Base each conclusion on rows from the file and tell me how many. If data is missing, say so instead of guessing.
Output format: Table with reason, number of tickets and average rating, and below it 3 concrete actions.

Claude prompts for customer service

Claude stands out with long documents and careful drafting: it works for complex complaints, for polishing drafts that sound like templates and for setting your brand tone. Paste the full texts into the message or upload them as a file.

9. Complex complaint with policy and conversation

Role: Customer service lead, careful and clear.
Context: Below I paste the returns policy: [policy] and the full conversation with the customer: [email or chat thread].
Task: Summarize what the customer asks for and what they are entitled to under the policy. Then draft the reply.
Constraints: Quote the policy clause you apply. If something is not covered, say so and flag it for human review. Do not invent exceptions.
Output format: First a 3-line summary, then the reply ready to send and finally the open questions to review.

10. Rewrite a draft that sounds robotic

Role: Customer service copy editor.
Context: This draft sounds like a template: [paste the draft]. The customer is [emotion].
Task: Rewrite it so it sounds like a real person, keeping the solution and the facts.
Constraints: Remove empty phrases ("we appreciate your patience", "do not hesitate to contact us"). Do not change commitments or deadlines. Explain in two lines what you changed.
Output format: Rewritten text and below it the list of changes.

11. Brand tone guide from real replies

Role: Brand voice and content lead.
Context: I am pasting 8 support replies we liked, with no personal data: [replies].
Task: Extract how our brand speaks: 3 adjectives, phrases we repeat, phrases we would never use and how we greet and close.
Constraints: Base everything on the pasted texts. Include a before and after example of a cold sentence.
Output format: One-page guide that I can paste at the start of any future prompt.

12. Reply to a sensitive case, for human review

Role: Customer service writer preparing a draft for a manager to review.
Context: The customer reports [a delicate situation, for example: a bereavement and cancellation of a relative's service].
Task: Draft a sober, respectful reply that confirms the administrative steps.
Constraints: No stock phrases or emojis. Give no legal advice and promise no refunds without confirmation. Mark in [brackets] what a person must verify.
Output format: Draft of at most 120 words and a list of 3 checks before sending.

Copilot prompts for customer service

Copilot works inside Outlook, Teams, Word and Excel with your own files, emails and meetings, as long as your Microsoft 365 license includes it. That is why it suits support teams that already live in those apps. Check which features your plan has before you standardize the workflow.

13. Reply to an email thread in Outlook

Role: Support agent replying from Outlook.
Context: This email thread with customer [name], with [N] messages and an unresolved problem.
Task: Summarize the thread in 3 points and draft a reply that acknowledges what has already happened and offers [solution].
Constraints: Warm, professional tone. Do not repeat information the customer already gave. Maximum 130 words.
Output format: Bullet summary and below it the draft reply.

14. Summary of a call or meeting in Teams

Role: Customer service coordinator.
Context: The Teams meeting with customer [name] about [incident].
Task: Summarize what the customer asked for, what we agreed and what tasks remain, with owner and date.
Constraints: Use only what was said in the meeting. If no date was agreed, write "no date".
Output format: Table with task, owner and date, and a 4-sentence follow-up email for the customer.

15. Classify tickets in Excel

Role: Support analyst working in Excel.
Context: The table in this sheet has the tickets for [month] with the columns [date, channel, description, response time].
Task: Add a column with the ticket reason (at most 6 categories) and another with urgency. Then summarize average response time by reason in a pivot table.
Constraints: Always use the same categories. Flag doubtful tickets instead of guessing.
Output format: New columns in the original table and a pivot table on a separate sheet.

16. Create a canned replies document in Word

Role: Customer service quality lead.
Context: The document [returns policy.docx] and the tone guide [tone guide.docx].
Task: Create a document with 6 template replies for the most frequent questions about returns.
Constraints: Apply the tone from the guide. Each reply with gaps in [brackets] for name, order and date. Do not add conditions that are not in the policy.
Output format: One heading per reply, at most 90 words each.

Gemini prompts for customer service

Gemini is the natural option if your team works in Gmail, Docs and Sheets. Some features inside Workspace depend on your plan, so check what yours includes.

17. Reply in Gmail

Role: Support agent replying from Gmail.
Context: The customer email I have open, which asks for [what they ask] and is [emotion].
Task: Draft an empathetic reply that resolves their request and proposes [next step].
Constraints: Warm tone, no jargon. Do not include details that do not appear in the email. Maximum 100 words.
Output format: Reply ready to send, with greeting and closing.

18. Support playbook in Docs

Role: Training lead for a support team.
Context: My rough notes on how we handle [case type]: [your rough notes].
Task: Turn them into a one-page guide for new agents.
Constraints: Use only my notes. Include one example opening sentence and one closing sentence. Mark what is missing with [to complete].
Output format: Document with a title, 5 short sections and a final list of mistakes to avoid.

19. Detect emotion and urgency in Sheets

Role: Customer experience analyst.
Context: A sheet with customer comments in column [B], with no personal data.
Task: For each comment, give the main emotion (anger, doubt, gratitude or other) and whether it is urgent.
Constraints: Use only those emotion categories. If the comment is ambiguous, write "review".
Output format: Two new columns, emotion and urgency, next to each comment.

More AI prompt examples: use cases for closing, thanks, feedback and sales

These four work in any tool and help improve communication with customers. They cover the small messages that repeat the most and tend to sound the most automated.

20. Closing a ticket

Role: Support agent at [company].
Context: The ticket about [problem] is resolved with [solution].
Task: Write a closing message that confirms the fix, asks if anything is still open and thanks them for their patience in a specific way.
Constraints: Warm, professional tone. Do not use "we hope this was helpful".
Output format: Maximum 3 sentences.

21. Thank-you to a customer

Role: Customer relations lead at [company].
Context: [Customer] has just [renewed, recommended us or reached one year with us].
Task: Write a short thank-you email that sounds sincere, not automated.
Constraints: Mention one specific detail: [fact from their history]. Ask for nothing in return.
Output format: Maximum 4 sentences.

22. Feedback request

Role: Customer experience lead.
Context: We have just resolved an issue of [type] for the customer.
Task: Write a message that asks them to rate the service on a 1 to 5 scale plus one open question.
Constraints: It should read like a personal request, not a survey. No more than two questions.
Output format: Maximum 3 sentences with the questions on separate lines.

23. LinkedIn follow-up to a prospect

Role: Sales lead who nurtures the relationship before selling.
Context: A prospect downloaded [resource] from our site.
Task: Write a LinkedIn message that introduces you, adds a useful fact or resource and proposes a conversation.
Constraints: No aggressive sales tone or attachments. No "I am writing to introduce our services".
Output format: Maximum 5 sentences.

How I built these prompts and how to write a good prompt

They all follow the same structure, because empathy does not depend on the tool but on how much information you give it before asking it to write. A generic prompt like “Reply to this complaint” returns something correct but neutral. One that names the customer’s emotion and your brand tone returns a message the customer feels was written for them. Writing prompts like this is the foundation of prompt engineering applied to support. It lets you tailor each empathetic response, keep it concise and improve customer interactions in any channel, from a simple customer inquiry to a complex complaint.

How to structure a prompt: five tips for creating your own prompts

1. Start with the role. “Act as a support specialist in [industry]” sets where the AI answers from.
2. Name the emotion. Say explicitly how the customer feels, not only what they ask.
3. Define the tone in three adjectives and repeat them in every prompt, or paste a tone guide at the start.
4. Ban what you hate. A short list of empty phrases in the constraints removes half of the robotic effect.
5. Change one thing per test and save the prompts that work in a shared template.

Here is how a weak prompt gets transformed. This is the starting point:

Weak: “Reply to this customer who is angry because their order arrived late.”

It returns something correct but cold, with “we apologize for any inconvenience” and no solution. This is the strong version, with the six fields:

Role: Support agent at [online store], warm and honest tone.
Context: [Marta]'s order is 3 days late. She wrote: "[paste her message]".
Emotion: She is frustrated and feels ignored.
Task: Apologize, explain the cause in one sentence and offer [free express reshipment].
Constraints: No "we apologize for any inconvenience". Do not promise a date that is not confirmed. Maximum 110 words.
Format: Email with subject line and two short paragraphs.

The result names her frustration, gives a concrete step and sounds like a person. The difference is not length but the precision of each part.

AI tools: how the prompt changes in each one

• ChatGPT: the most widely used tool, with the most documented resources. It accepts files, such as a ticket CSV, so paste or upload your data in the same message. The prompts on this page work as written.
• Claude: stands out with long, nuanced replies. Paste policies, full threads or a tone guide and ask it to quote what it applies.
• Copilot: works with your Microsoft 365 emails, meetings and files. Instead of pasting the text, reference the thread, the meeting or the document.
• Gemini: integrates with Google Workspace. Use it to reply in Gmail or prepare guides in Docs, and check which features your plan includes.

Terms and plans reviewed in October 2026. Generative AI moves fast: check each platform for the free offer and the features of your plan before you standardize.

How to integrate AI into your support team workflow

A single prompt helps one person. A prompt built into the process helps the whole customer service team. The leap is moving from individual use to a shared, measurable system.

A concrete case: a three-person customer service team at an e-commerce company kept their best prompts in a loose document and everyone improvised. After creating a shared template with the base prompts and their tone variants, they cut the average time to write complex replies and gained consistency between agents. The efficiency did not come from AI alone but from structuring how they used it.

1. Create a shared template with the prompts validated by tone and use case.
2. Connect a chatbot for repetitive first-level queries.
3. Always review before sending. AI proposes, a person approves.
4. Measure efficiency: response time, customer satisfaction and tickets resolved on first contact.

This AI tool does not replace the team, it frees it from the repetitive part. To apply the same structure to other teams, see the AI prompts for marketing guide.

What an AI prompt for customer service is and who it is for

An AI prompt for customer service is an instruction that tells an artificial intelligence tool what to answer, in what tone and within what limits when handling a query. It turns a customer question or complaint into a clear, empathetic reply aligned with your brand, which directly affects the customer experience.

It is useful for several profiles. A support team uses it to write consistent replies under high ticket volume. A customer success manager applies it to prepare follow-ups and renewals. A sales team uses it to look after the customer relationship at every touchpoint. And small businesses, which rarely have a large customer service department, answer at the same level as much larger companies.

Writing good prompts is a skill, not a trick. Tools like ChatGPT are available to anyone, but the judgment to use them well is learned, which is why AI training has become a priority for support teams. If you are starting from zero, the AI literacy courses at Founderz give you the basics with no technical background needed.

Where AI helps and where human judgment is still needed

AI speeds up drafting, but there are situations where you should not delegate the reply, because a mistake has real consequences:

• Legal claims: any message with contractual or liability implications.
• Sensitive data: medical, financial or personal information that must not leave your systems.
• Customers in crisis: delicate emotional situations that need judgment and real empathy, not a template.

Frequently asked questions about AI prompts for customer service

These are the most common questions support, customer service and customer success teams ask when they start using AI.

What are the best AI prompts for customer service?

The ones that include role, context, customer emotion, goal and tone. They are not the longest but the most specific. This page has 23 ready to copy, from complaints and delays to retention and feedback.

How do I use ChatGPT for customer service?

Give it a role, the case context, the customer’s emotion, the goal and the tone, and ask for a concrete format, such as a 120-word email. Review the draft before sending. ChatGPT can also analyze files, for example a ticket CSV with no personal data.

How do I stop AI replies from sounding robotic?

Give it your brand tone in concrete adjectives, ask it to acknowledge the customer’s emotion and ban empty phrases in the constraints. Then edit whatever still sounds like a template. Generic output almost always comes from a generic prompt.

What is the best AI tool for empathetic customer service responses?

It depends on your case. ChatGPT has the most resources and analyzes files. Claude stands out with long documents and nuanced replies. Copilot fits if you work in Outlook, Teams and Excel, and Gemini if you work in Google Workspace. Try the free version of each with your real prompts and check current terms.

What platforms offer templated customer service prompts?

General assistants like ChatGPT, Claude, Copilot and Gemini all run templated prompts, and helpdesk tools such as Zendesk or Intercom have built-in reply suggestions. This page gives you a library you can paste into any of them, and the free generator builds a custom one.

Can I use Copilot for customer service?

Yes, if your Microsoft 365 license includes it. In Outlook it can summarize a thread and draft the reply, and in Teams it can summarize a meeting with the customer. Prompts 13 to 16 on this page are written for that.

How should an AI respond if a customer says “don’t pity me”?

Drop the sympathy language and move to action. Acknowledge once, briefly, then give the concrete next step and treat the customer as capable. Add “no pity or condescending phrases” to the constraints of your prompt.

Is it safe to use customer data in AI prompts?

Only if you know what the platform does with it. Avoid pasting identifiable data such as names, phone numbers or payment details, use bracketed markers and check the privacy policy of your company’s tool.

Can I use these prompts for marketing and sales as well as support?

Yes. The same structure works for digital marketing, email and sales teams: you change the role, context and goal. Prompt 23 is an example of a sales follow-up.

Your next steps with AI prompts for customer service

Pick one of the prompts, change what is in brackets and run your first test with a real case. Then change a single field, such as the tone or the constraints, and compare. That teaches you what each part controls much faster than reading theory. Knowing how to write a good support prompt, and creating your own for real use cases, is a matter of practice.

If you want your organization to learn to design and apply these workflows with judgment, the AI program at Founderz takes a practical approach, applied to the real work of your customer service team. The question is not whether AI will be part of your support, but whether your team will know how to direct it.