The customer support prompts below help agents and team leads with the daily work: replying to tickets from your own docs, calming upset customers, saying no to a refund within policy, escalating bugs to engineering, writing macros and help articles, and spotting trends across hundreds of tickets.
How to use these prompts
Every prompt works in ChatGPT, Claude or Gemini. Replace the blanks in double braces, or keep the default after the colon: {{product:a project management app}} means "a project management app" unless you change it.
Support replies need facts from your company, not from the model's general knowledge. Paste the relevant help article, policy or ticket history into the prompt and tell the model to use only that. Anthropic's guide to reducing hallucinations suggests explicitly allowing the model to say "I don't know" and having it find a supporting quote for each claim. Both habits are built into the reply prompts here.
OpenAI's prompt guide for ChatGPT at work uses a support persona as its example: "You are a customer support lead drafting a calm, helpful reply." A short role line like that, plus your tone rules, keeps replies consistent across agents.
Remove personal data you don't need before you paste a ticket. Replace names, emails and order numbers with placeholders like [CUSTOMER] and [ORDER].
Which prompts help reply to tickets?
1. Reply from your help docs
The everyday prompt. It answers only from what you paste and flags what it can't answer.
You are a support agent for {{product:a project management app}}.
Tone: {{tone:friendly, plain and brief}}.
Help docs and policies:
{{docs}}
Customer ticket:
{{ticket}}
Write a reply that answers the customer's actual question in the first
two sentences, using only the docs above. Include steps as a numbered
list if there are more than two. If the docs don't cover something, say
so in a note to me (not in the reply) and suggest what to check. Under
{{length:150 words}}.
2. Answer a ticket with several questions
For long tickets that mix a bug, a billing question and a feature request.
This ticket contains several questions or issues. First list each one
separately. Then write a single reply that answers each in order, with a
short heading or number per issue, using only the docs below. For any
issue you can't resolve from the docs, say what happens next and when.
Docs:
{{docs}}
Ticket:
{{ticket}}
3. Clarify before you guess
For vague tickets like "it's not working". One good question beats three rounds of back and forth.
This ticket doesn't have enough detail to solve:
{{ticket}}
Write a short reply that thanks them, says what we'll do once we know
more, and asks for the {{max:3}} most useful pieces of information, in
the order that would help most (for example: device and browser, the
exact error, steps to reproduce, account email). Make each question easy
to answer. Under 90 words.
Which prompts handle upset customers?
4. De-escalate an angry customer
Paste the message and the facts. The model removes defensive phrasing you might not notice under pressure.
A customer sent this angry message:
{{message}}
What actually happened: {{facts}}
What I can offer: {{offer:a credit for one month}}
What I can't do: {{limits}}
Write a reply that acknowledges the specific problem in their words,
takes responsibility for what was our fault, explains briefly what we're
doing, and ends with a concrete next step and time. No blame, no
"we apologize for any inconvenience", no promises outside what I can
offer. Then tell me if anything in their message suggests I should
escalate to a manager.
5. Say no within policy
For refunds, exceptions or features you can't provide. A clear, kind no with an alternative keeps more customers than a vague maybe.
A customer asked for {{request:a refund after the 30-day window}}.
Our policy says: {{policy}}
Write a reply that says no clearly in the first two sentences, explains
the reason in plain language without quoting legal text, and offers
{{alternative:the best alternative we can give}}. Warm, not robotic.
Under 120 words. Don't hint that an exception might be possible unless
I said so.
6. Apologize for an outage or incident
For the reply to customers affected by downtime, a data issue or a wrong charge.
Write a message to customers affected by {{incident:a 3-hour outage on Tuesday}}.
What happened (plain facts): {{facts}}
Impact on customers: {{impact}}
What we fixed: {{fix}}
What we're changing: {{prevention}}
Any credit or remedy: {{remedy:none}}
Under 180 words. Apologize once, plainly. Say what happened without
jargon, what they need to do (if anything), and where to get help.
Don't speculate on causes beyond the facts above.
Which prompts help escalate to engineering?
7. Ticket to bug report
For handing a customer issue to developers with everything they need to reproduce it.
Turn this customer conversation into a bug report for the engineering
team.
{{conversation}}
Format: title (what breaks, where), steps to reproduce, expected
result, actual result, environment (device, browser, plan, version if
known), how many customers reported it if mentioned, and the customer's
impact in one line. Mark anything unknown as "unknown". Remove the
customer's personal details.
8. Handoff summary for another agent
For shift changes, escalations or tickets moving between tiers. Nobody should have to read 30 messages.
Summarize this ticket for the next agent picking it up.
{{ticket_history}}
Give: the customer's goal in one sentence, what has been tried and the
result of each, what the customer has been promised and by when, their
current mood, and the single next step. Under 120 words. Quote any
promise made to the customer word for word.
Which prompts build your support knowledge base?
9. Write a help article from solved tickets
For questions you answer every week. Paste three to five solved tickets on the same topic.
Below are several solved tickets about {{topic:connecting a calendar}}.
{{tickets}}
Write a help center article: a title phrased as the customer's question,
a two-sentence answer, numbered steps, common problems with fixes, and
when to contact support. Use the words customers used, not internal
names. Mark any step you inferred rather than saw in a ticket with
"Check:".
10. Create macros from common replies
For saved replies in your help desk. Each macro has placeholders and a note on when to use it.
Below are replies our agents have sent for {{topic:password reset problems}}.
{{replies}}
Write {{count:3}} reusable macros that cover the common cases. For each
give: a name, when to use it (one line), and the reply text with
placeholders in square brackets such as [CUSTOMER_NAME]. Keep each
under 120 words and in a {{tone:friendly, plain}} tone. Note any case
that still needs a custom reply.
11. Draft a system prompt for a support assistant
For teams setting up an AI assistant. Anthropic's customer support agent guide recommends adding at least four to five sample "good" interactions and a list of guardrails drawn from your support policies.
Draft a system prompt for an AI assistant that helps customers of
{{product:a project management app}}.
Scope: {{scope:account, billing and how-to questions}}
Never handles: {{out_of_scope:legal questions, refunds over the standard policy}}
Tone: {{tone:friendly, plain and brief}}
When to hand off to a human: {{escalation}}
Include: the role, tone rules, what to do with off-topic questions, how
to answer when the docs don't cover something, the handoff rule, and
four example conversations (two normal, one off-topic, one handoff).
Which prompts help support leads?
12. Tag and triage a batch of tickets
For a backlog or a weekly sample. It turns a pile into categories you can act on.
Here are {{count:50}} recent support tickets, one per line.
{{tickets}}
Tag each with one category from this list: {{categories:bug, billing, how-to, feature request, account access, other}}.
Mark urgency as high, normal or low with a short reason for every high.
Then give a table of counts per category and list the three tickets
that most need attention today.
13. Weekly ticket trends report
For the report to product and leadership. It connects ticket volume to things the company can fix.
Below is a summary of this week's tickets by category, and last week's
for comparison.
This week: {{this_week}}
Last week: {{last_week}}
Sample ticket text: {{samples}}
Write a short report for {{audience:the product team}}: the top three
issues by volume, what changed versus last week and a likely reason if
the samples show one, two customer quotes that show the impact, and one
suggested fix per issue. Under 250 words. Don't invent numbers.
14. Review a reply before it goes out
For coaching new agents or checking replies on sensitive tickets.
Review this support reply against our standards before it is sent.
Standards: {{standards:answers the question first, correct per docs, friendly, no jargon, clear next step}}
Docs: {{docs}}
Customer ticket: {{ticket}}
Draft reply: {{reply}}
Score each standard as pass or fix, quote the line behind each fix, and
show a corrected reply. Point out any statement not supported by the
docs.
How do you get better support replies from AI?
Paste the policy, every time. The model doesn't know your refund window, plan limits or shipping times, and it will guess politely if you don't tell it.
Write the tone rules down. Three or four lines, such as "answer first, short sentences, no exclamation marks, never blame the customer", do more than "be friendly". Put them in every reply prompt or in a saved instruction.
Keep a human on the send button for anything involving money, account access, security or legal terms. Use AI to draft and to check, and let the agent decide.
Feed the loop back. When a ticket shows a gap in your help docs, run prompt 9 and publish the article. Fewer repeat questions beat faster replies.
Make your favorite one a skill
If your team runs the same reply or triage prompt all day, turn it into a skill with your tone rules and escalation policy built in, so Claude, Claude Code or Codex applies them every time. Paste the prompt into the free SKILL.md generator. For a support assistant's instructions, see more system prompt examples; for help center writing, the documentation prompts go deeper; and for any non-support email, use the email writing prompts.
Kipt stores prompts like these as Markdown in a private GitHub repo you own. Save each reply prompt with its blanks, tag it by ticket type, and press ⌘K to find the right one mid-ticket.