Home / Guides / Escalations and support operations

AI prompts for support case work: from first read to closing note

By Draft First. Updated October 2026

Most of a support case isn't the fix. It's reading a long thread, working out what's missing, writing the update, handing it over at the end of a shift and closing it cleanly. AI is good at that reading and drafting, as long as it sticks to what's in the case and a person sends the result.

These six prompts follow a case from the first read to the closing note. Swap the [brackets] for your own details and paste the thread after the prompt.

The rules to put above every case prompt

• Use only the case. Facts come from the thread, the ticket and the notes you paste. Nothing else.
• Quote promises exactly. Any date or commitment made to the customer is quoted, with who said it and when.
• Causes are hypotheses until someone in the case confirms them.
• Draft only. Nothing goes to the customer until you've read it.

1. First read of a case

Read this case and give me: the problem in the customer's words, the
product area and version if stated, what has been tried and what happened,
current status, who is waiting on whom, and any promise we made, quoted
with who made it and when. Then list what you couldn't tell from the case.
[paste the ticket and thread]

2. Ask the customer for what's missing

From this case, list the details we still need to move forward, like
steps to reproduce, when it started, how many users are affected or an
exact error message. Then draft a short, friendly reply that asks for
them in a numbered list. Don't guess at the cause and don't promise a
fix or a date. [paste the case]

3. Case handoff note

Write a handoff note for [next shift or next tier] from this case: one
line on the problem, what's been tried, current status, the next action
and who owns it, any deadline, and any promise to the customer, quoted.
Put anything due in the next [4] hours at the top. Keep it under 150
words. [paste the case]

For a whole queue at once, the shift handoff prompt in AI prompts to summarize support tickets works across all your open tickets.

4. Customer update

Using only the facts in this case, draft an update to the customer: what
we know, what we're doing now and when they'll hear from us next
([next update time]). Acknowledge the impact in their words if the case
has them. Leave out internal names and any theory about the cause.
Never promise a fix, a date or a credit. Under 150 words.
[paste the case]

5. Escalation brief for leadership

Summarize this case for a leader who has two minutes: customer impact,
timeline with dates, current status, owner, next update and what we need
from leadership. Label any cause as a hypothesis and say who raised it.
Under 250 words. [paste the case]

There's a full walkthrough with a worked example in how to write an escalation brief.

6. Closing note

Draft a closing note for this case: the problem, what fixed it or why
we're closing it, any steps the customer took, and a dated timeline of
the key events. List anything that should become a help article or a bug
report. Mark anything you can't confirm from the case as [CONFIRM].
[paste the case]

Check the draft before you trust it

Run this after any of the prompts above, in the same chat:

Compare what you just wrote with the case. List every statement that
isn't in the case, any promise that isn't quoted word for word, and
anything stated as a cause that nobody in the case confirmed. Don't
rewrite it. Just list the problems, or say "No problems found".

Want help routing cases in the first place? See the support escalation matrix, and for upset customers, how to respond to an angry customer email.

Questions

Can AI work support cases on its own?

No. Let it read and draft. A person checks the facts, decides and sends anything a customer will see.

What should a case handoff include?

The problem in the customer's words, what's been tried, current status, who is waiting on whom, the next action and any promise with its date.

Is it safe to paste case details into an AI tool?

Only into tools your company approves for customer data. If you're not sure, take names and contact details out first.