AI prompts for customer support are ready-to-copy instructions that help support teams respond faster, stay consistent, and handle common ticket types without starting from scratch each time. This page covers a curated pack of 20-plus prompts across five categories: first replies, escalation, refunds and policy, internal documentation, and follow-up. It is built for support team leads and managers who want their whole team working from the same playbook, not just the one person who knows how to prompt.
What is in this pack
This pack contains prompts for the full support workflow, from the first reply on a new ticket through to closing a conversation and updating internal docs.
The prompts are grouped into five categories: first-contact replies, escalation and complaint handling, refunds and policy explanations, internal knowledge base writing, and post-resolution follow-up. Each one is written to work across ChatGPT, Claude, Copilot, and Gemini without modification.
One honest caveat up front: these prompts handle language and structure well, but they cannot pull live data from your CRM or ticketing system. Any prompt that references order numbers, account history, or SLA timelines will need that information pasted in manually. If your team needs real-time system integration, a dedicated support platform with native AI features is likely a better fit than a prompt library.
The pack also includes a category breakdown table, five prompts with real captured outputs, and a short section on adapting everything for your team's tone of voice.
The prompts, by category
These 20 prompts cover the situations that come up every day in a support queue. They are grouped by job, not by AI tool, and they work across ChatGPT, Claude, Copilot, and Gemini without modification.
One honest caveat before you start: AI-drafted responses need a human check before they go to a customer. These prompts will save your team significant time on drafting and research, but they will occasionally get a policy detail wrong or miss emotional nuance. Build review into your process, especially for billing disputes and complaints.
Category overview
| Category | Prompts | Best for |
|---|---|---|
| Drafting responses | 1-5 | First-contact replies, tone matching |
| Escalation and complaints | 6-9 | Angry customers, policy exceptions |
| Internal documentation | 10-13 | SOPs, handoff notes, FAQs |
| Self-service content | 14-16 | Help articles, chatbot scripts |
| Quality and coaching | 17-18 | CSAT review, agent feedback |
| Onboarding and training | 19-20 | New agent ramp-up |
Draft a first-contact reply to a common question
When to use it: A customer asks something your team answers ten times a day. What to change first: The tone instruction, to match your brand voice.
Variables: [PRODUCT NAME], [QUESTION SUMMARY], [TONE: friendly/formal/concise]
You are a customer support agent for [PRODUCT NAME].
A customer has contacted us with the following question:
"[QUESTION SUMMARY]"
Write a first-contact reply that:
- Answers the question directly in the first sentence
- Uses a [TONE] tone throughout
- Stays under 120 words
- Ends with one clear next step for the customer
Do not use filler phrases like "Great question!" or "I hope this finds you well."Acknowledge a complaint without conceding fault
When to use it: A frustrated customer is waiting for an investigation to complete. What to change first: The acknowledgment language, if your legal team has specific requirements.
Variables: [PRODUCT/SERVICE NAME], [ISSUE DESCRIPTION], [TIMELINE FOR RESOLUTION]
You are drafting a customer support email for [PRODUCT/SERVICE NAME].
A customer has complained about: [ISSUE DESCRIPTION]
We are still investigating and cannot confirm fault or offer compensation yet.
Write an email that:
- Acknowledges the customer's frustration genuinely, without admitting liability
- Confirms we are looking into it and will respond by [TIMELINE FOR RESOLUTION]
- Uses a calm, professional tone
- Is under 150 words
Avoid vague phrases like "we take this seriously." Be specific about the next step.[NEEDS REAL OUTPUT]
Match the tone of an angry customer
When to use it: A customer message is emotionally charged and a neutral reply will read as dismissive. What to change first: The escalation threshold, if your policy routes certain complaint types to a senior agent.
Variables: [CUSTOMER MESSAGE], [AGENT NAME], [COMPANY POLICY ON ISSUE]
Read the following customer message carefully:
"[CUSTOMER MESSAGE]"
Draft a reply from [AGENT NAME] that:
- Mirrors the urgency and emotion in the customer's message without becoming defensive
- Explains what we can do under [COMPANY POLICY ON ISSUE]
- Offers one concrete action we will take today
- Does not over-apologize or use the word "unfortunately"Write a polite refusal
When to use it: A request falls outside policy and needs a clear no that preserves the relationship. What to change first: The alternative you offer, which should be real and available.
Variables: [REQUEST DESCRIPTION], [REASON FOR REFUSAL], [ALTERNATIVE OPTION]
A customer has requested: [REQUEST DESCRIPTION]
We cannot fulfill this because: [REASON FOR REFUSAL]
Write a reply that:
- States the refusal clearly in the first sentence
- Explains the reason briefly (one sentence)
- Offers [ALTERNATIVE OPTION] as a genuine alternative
- Closes warmly without being effusive
- Is under 100 wordsSummarize a long thread for a handoff note
When to use it: Passing a ticket to another agent or escalating to a specialist. What to change first: The output format, if your helpdesk uses a structured template.
Variables: [PASTE FULL THREAD]
Read the following customer support thread:
[PASTE FULL THREAD]
Write a handoff note that includes:
1. The core issue in one sentence
2. What has already been tried or offered
3. The customer's current emotional state (calm / frustrated / very frustrated)
4. The single most important thing the next agent needs to know
5. Suggested next action
Keep the note under 100 words.[NEEDS REAL OUTPUT]
Draft an internal FAQ entry from a recurring ticket
When to use it: The same question has appeared three or more times this month. What to change first: The audience line, so tone matches whether agents or customers will read it.
Variables: [QUESTION], [CORRECT ANSWER], [AUDIENCE: agents / customers], [PRODUCT NAME]
We keep receiving the following question about [PRODUCT NAME]:
"[QUESTION]"
The correct answer is: [CORRECT ANSWER]
Write an FAQ entry for [AUDIENCE] that:
- States the question exactly as a customer would ask it
- Answers it in the first sentence
- Adds any necessary detail in two to three sentences
- Flags any exceptions or edge casesWrite a help center article from a support ticket
When to use it: A resolved ticket reveals a gap in your self-service documentation. What to change first: The reading level instruction, depending on your customer base.
Variables: [TICKET SUMMARY], [PRODUCT NAME], [READING LEVEL: simple / intermediate / technical]
A customer support ticket revealed the following gap in our documentation:
[TICKET SUMMARY]
Write a help center article for [PRODUCT NAME] at a [READING LEVEL] reading level.
Structure:
- H1: A question the customer would actually search
- One-sentence answer
- Step-by-step instructions (numbered)
- One "if this doesn't work" troubleshooting note
- No marketing language
Keep it under 300 words.[NEEDS REAL OUTPUT]
Generate chatbot response variants
When to use it: Building or updating a scripted flow and need options to A/B test. What to change first: The character limit, to match your chatbot platform's constraints.
Variables: [INTENT NAME], [KEY INFORMATION], [TONE: friendly/formal], [CHARACTER LIMIT]
Generate three different chatbot response variants for the following intent:
Intent: [INTENT NAME]
Information to convey: [KEY INFORMATION]
Each variant must:
- Be under [CHARACTER LIMIT] characters
- Use a [TONE] tone
- Avoid questions that could confuse a customer who expected a direct answer
- Differ meaningfully from the other two (not just word substitutions)
Label each variant A, B, and C.Score a support reply for CSAT risk
When to use it: QA review, or before sending a response to a high-value account. What to change first: The scoring criteria, if your team uses a custom rubric.
Variables: [DRAFT REPLY]
Read the following customer support reply:
[DRAFT REPLY]
Score it on each of the following criteria, 1 to 5:
1. Clarity: Does it directly answer the customer's question?
2. Tone: Is it warm without being obsequious?
3. Accuracy risk: Are there any claims that could be factually wrong?
4. Length: Is it longer than it needs to be?
5. Next step: Does the customer know exactly what happens now?
For each score below 4, give one specific suggestion to improve it.[NEEDS REAL OUTPUT]
Create a coaching note for an agent from a real ticket
When to use it: Weekly 1-on-1s, or after a ticket that ended in a poor CSAT score. What to change first: The framing tone, from developmental to corrective, depending on the situation.
Variables: [TICKET TRANSCRIPT], [AGENT NAME], [AREA OF FOCUS: tone / accuracy / resolution speed]
You are a customer support team lead preparing feedback for [AGENT NAME].
Review the following ticket transcript:
[TICKET TRANSCRIPT]
Write a coaching note focused on [AREA OF FOCUS] that:
- Opens with one thing the agent did well (specific, not generic)
- Identifies one specific moment where a different approach would have helped
- Suggests a concrete alternative phrasing or action for that moment
- Closes with an actionable goal for the next five tickets
Keep the tone developmental, not critical.Write onboarding scenarios for a new support agent
When to use it: First week of training, before a new hire handles live tickets. What to change first: The difficulty level, starting with straightforward and adding edge cases as the agent progresses.
Variables: [PRODUCT NAME], [COMMON ISSUE TYPE], [DIFFICULTY: beginner / intermediate / advanced]
Create three realistic customer support scenarios for a new agent at [PRODUCT NAME].
Each scenario should involve [COMMON ISSUE TYPE] at a [DIFFICULTY] difficulty level.
For each scenario include:
1. The customer's opening message (written as if real)
2. One complicating detail that appears if the agent asks a follow-up question
3. The ideal resolution path (two to three steps)
4. One thing an inexperienced agent commonly gets wrong here
Format each scenario under a numbered heading.[NEEDS REAL OUTPUT]
Rewrite a canned response to remove stale language
When to use it: Auditing your macro library for replies that feel outdated or robotic. What to change first: The examples of phrases to remove, which should come from your actual macros.
Variables: [EXISTING CANNED RESPONSE], [BRAND VOICE DESCRIPTION]
Here is an existing canned response from our support team:
[EXISTING CANNED RESPONSE]
Our brand voice is: [BRAND VOICE DESCRIPTION]
Rewrite this response so that it:
- Removes any phrases that sound automated or generic (e.g., "per my last email," "as per our policy," "please be advised")
- Reads like a person wrote it
- Keeps every piece of factual information from the original
- Is no longer than the original
Show the original and the rewrite side by side.Translate a support reply without losing tone
When to use it: Responding to a customer in a language your team does not have a native speaker for. What to change first: The instruction to flag cultural considerations, which matters most for formal/informal register in languages like French or Japanese.
Variables: [ORIGINAL REPLY IN ENGLISH], [TARGET LANGUAGE], [TONE: formal / informal]
Translate the following customer support reply into [TARGET LANGUAGE].
Original reply:
[ORIGINAL REPLY IN ENGLISH]
Instructions:
- Use a [TONE] register appropriate to customer service in [TARGET LANGUAGE]
- Preserve the meaning of every sentence exactly
- Flag any phrase where a direct translation would sound unnatural or rude
- Do not add information that is not in the originalIdentify the root cause category from a batch of tickets
When to use it: Monthly reporting, or before a product review meeting where support data should inform the agenda. What to change first: The categories list, to match your existing tagging taxonomy.
Variables: [PASTE UP TO 20 TICKET SUMMARIES]
Read the following ticket summaries:
[PASTE UP TO 20 TICKET SUMMARIES]
Group them into root cause categories. For each category:
- Give it a short descriptive name
- List the ticket numbers that belong to it
- Write one sentence describing the underlying issue
- Note whether the root cause is a product problem, a documentation gap, or a process failure
Output a table with columns: Category | Ticket Numbers | Root Cause Type | One-line DescriptionDraft a proactive outreach message for a known issue
When to use it: Before your support queue fills up because a bug or outage is already affecting users. What to change first: The estimated resolution time, which must come from your engineering team, not from the prompt.
Variables: [ISSUE DESCRIPTION], [AFFECTED USERS: all users / specific segment], [ESTIMATED RESOLUTION TIME], [WORKAROUND IF ANY]
Draft a proactive customer communication about the following known issue:
Issue: [ISSUE DESCRIPTION]
Affected: [AFFECTED USERS]
Estimated resolution: [ESTIMATED RESOLUTION TIME]
Workaround: [WORKAROUND IF ANY]
Write this as an email subject line and body that:
- States the problem in the subject line (not "Important update")
- Opens with the impact on the customer, not on us
- Explains what we are doing about it in one sentence
- Gives the workaround clearly if one exists
- Commits to a follow-up communication at a specific timeWrite a post-resolution follow-up email
When to use it: After closing a ticket that involved significant effort or a frustrated customer. What to change first: The CSAT ask, which you may want to remove if your platform sends a separate survey automatically.
Variables: [CUSTOMER NAME], [ISSUE RESOLVED], [AGENT NAME], [PRODUCT NAME]
Write a follow-up email from [AGENT NAME] to [CUSTOMER NAME] after resolving:
[ISSUE RESOLVED]
The email should:
- Confirm the issue is resolved in the first sentence
- Briefly explain what was done (one sentence, non-technical)
- Invite the customer to reply if the problem recurs
- Include a low-pressure CSAT ask in the final sentence
- Be under 100 words
Do not use "I hope this email finds you well" or any variant of it.Extract a process improvement from a negative review
When to use it: Monthly review of one-star feedback, or when preparing a support retrospective. What to change first: The output format, if you feed results directly into a project management tool.
Variables: [NEGATIVE REVIEW TEXT], [PRODUCT OR SERVICE NAME]
Read the following customer review of [PRODUCT OR SERVICE NAME]:
[NEGATIVE REVIEW TEXT]
Identify:
1. The specific support failure the customer experienced (not just "bad service")
2. The point in the support journey where it went wrong
3. One process change that could prevent this in future
4. Whether this appears to be a one-off or a systemic issue (and your reasoning)
Output as a structured list, not a paragraph.Prioritize a mixed inbox by urgency
When to use it: Start of shift, when an agent inherits a queue with no clear ordering. What to change first: The urgency criteria, to match your SLA tiers.
Variables: [PASTE SUBJECT LINES OR TICKET SUMMARIES]
Here are the tickets currently in the queue:
[PASTE SUBJECT LINES OR TICKET SUMMARIES]
Rank them from highest to lowest urgency using these criteria (in order):
1. Service outage or data loss affecting the customer's business
2. Billing error or unauthorized charge
3. Legal or compliance mention
4. Repeat contact (same customer, same issue)
5. General question or feature request
For each ticket, give: Rank | Ticket | Urgency Category | Reason (one sentence)
If two tickets share a category, rank the older one higher.Suggest self-service alternatives before escalating
When to use it: When a ticket could be resolved without agent involvement if the right resource existed. What to change first: The list of available resources, which must reflect what you actually have published.
Variables: [TICKET SUMMARY], [LIST OF AVAILABLE SELF-SERVICE RESOURCES]
A customer submitted the following ticket:
[TICKET SUMMARY]
Our available self-service resources include:
[LIST OF AVAILABLE SELF-SERVICE RESOURCES]
Assess whether this ticket could have been resolved through self-service.
Output:
1. Yes / Partially / No
2. If yes or partially: which resource, and the exact section or step that answers it
3. If no: what would need to exist for future customers with this issue to self-serve
4. A suggested reply that points the customer to the right resource (if applicable), in a tone that does not feel like a brush-offReal example outputs
Five prompts in this pack have captured model outputs attached. They are marked [NEEDS REAL OUTPUT] as placeholders until the Convergence team replaces them with screenshots or verbatim text from a live session. The table below shows which prompts are queued for that treatment and what reviewers should look for when the outputs are added.
| Prompt | Category | What to notice |
|---|---|---|
| Triage a vague complaint | Triage | Whether the model asks one clarifying question or several |
| Write a refund-denial response | Difficult situations | Tone: firm but not cold |
| Draft an escalation summary | Escalation | How much it compresses without losing key facts |
| Suggest a knowledge base article | Proactive support | Whether the title and structure are immediately usable |
| Respond to a public review | Social and reviews | Whether the reply stays professional without sounding scripted |
Until real outputs are published here, the prompts by role page has a growing set of annotated examples across other functions.
How to adapt these for your team
The prompts above work as written, but they'll work better once you've replaced the generic variables with your actual product names, policies, and tone.
Start with [COMPANY TONE]. If your brand voice guide lives in a doc somewhere, paste the relevant paragraph directly into the prompt. "Friendly but concise" means something different to every team until you show the model what it looks like in your words.
Next, audit the two or three ticket types that consume the most time each week. Those are the categories worth customizing first. Swap out [PRODUCT/SERVICE NAME] and [POLICY DETAIL] for specifics, run a few test replies, and check whether the output matches what a senior agent would actually send.
One honest note: these prompts don't replace human judgment on sensitive tickets. Refunds involving unusual circumstances, complaints that hint at legal action, anything emotionally charged. Use the prompt to draft, then review before sending.
Once you have versions that consistently hit the mark, store them in a shared folder your whole team can access so no one is rebuilding from scratch.
Save this folder to your workspace
These prompts work as one-off copies, but the real gain comes from keeping them in a shared place where your whole team can find, use, and improve them together. When everyone pulls from the same set of templates, replies stay consistent across agents, shifts, and channels.
A few practical steps before you share:
- Replace the generic [BRACKETS] with defaults that fit your product. The less blank-filling an agent has to do, the faster they move.
- Add a short note to each prompt explaining when not to use it. That context matters as much as the template itself.
- Set a review reminder. Prompts drift out of date when products change, and stale templates cause the same inconsistency you were trying to fix.
For a broader set of starting points, the full prompts-by-role library covers other team functions alongside support. If you manage more than one team, the AI prompts for business page groups prompts by operational context rather than department.
Frequently asked questions
Can I use these prompts across ChatGPT, Claude, and Gemini?
Yes. Every prompt in this pack is written in plain language, so it runs on any major AI model without modification. You may notice slight differences in tone or length between models, but the structure and output quality hold up across all of them.
Do these prompts work for a small team without a dedicated support function?
They work especially well in that situation. A solo founder or a two-person team handling support alongside other work will get the most value from prompts that produce a full draft in one pass.
What should I change before using these prompts?
Replace every bracketed variable with your actual product name, policy, and customer context. Generic output is almost always the result of skipping this step.
Are these prompts suitable for live chat or only email?
Most prompts in this pack target written, asynchronous support. For live chat, shorten the output instruction and ask for a single paragraph rather than a full reply.
Where can I store these prompts so my whole team uses them?
Shared folders in Convergence let your team access, edit, and reuse a single set of prompts across ChatGPT, Claude, Copilot, and Gemini.