Guide
How to triage support tickets with AI
ChatGPT or Claude can do the first pass on a batch of support tickets, if it works from your written rules rather than the customer's tone. Paste the tickets with your severity rubric and routing, and ask for each one: a single category, an S1 to S4 severity with the rubric line that decided it, an owner, and a first reply of one to three sentences that says what you understood, what happens next and when, and asks for the one fact you're missing, without promising a fix. Ask it to raise three or more reports of the same failure within an hour to S1 as a possible incident, and to sort the table worst first. An angry message about a typo stays S4; a polite one about lost data is S1. A person checks the S1s before acting.
Short answer
- Severity from your rubric, not the tone.
- One category, one owner per ticket.
- First reply: understood, next step, one question.
- Worst first; a person checks the S1s.
The brief to copy
Ticket triage brief
"Triage these tickets.
Severity rubric: S1 many customers blocked, data loss, security exposure or money taken wrongly (reply within 1 hour); S2 one customer blocked on a core task with no workaround (4 business hours); S3 broken with a workaround, or a how-to (1 business day); S4 feature request or cosmetic (2 business days). If unsure between two, pick the higher. Three or more tickets about the same failure within an hour: all S1, flag a possible incident.
For each ticket: one category, the severity with the rubric line that decided it, the owner from our routing list (or [route: unclear]), and a first reply of 1 to 3 sentences: what we understood, what happens next and by when, the one fact we need. No fix promises.
Sort S1 first, then S2, oldest first within a level. Put duplicates on one row.
Routing and tickets: [paste]"
A worked example
Illustration: a made-up business. A booking app's support queue on a Monday morning, in part:
| Ticket | Severity and why | Owner | First reply |
|---|---|---|---|
| #2211, #2214, #2219: "bookings vanished overnight" | S1: data loss, three reports in an hour, possible incident | Engineering on-call | Thanks for telling us right away. Our engineers are looking now and we'll update you within the hour. Which studio and which day's bookings are missing? |
| #2216: "charged twice for October" | S1: money taken wrongly | Billing owner | Sorry about the double charge. Billing is checking it now and we'll update you within the hour. |
| #2215: "how do I add a second location?" | S3: how-to | Support | Happy to help. We'll send the steps today. Are both locations under the same account? |
| #2218: "TERRIBLE, the logo is blurry!!" | S4: cosmetic | Product | Thanks for flagging the blurry logo. We've passed it to the product team and will reply within two business days. Could you send a screenshot? |
The support lead confirms the three vanished-bookings tickets are one incident and opens it before the replies go out.
What stays with the team
- Write the rubric and routing down; the AI can only apply what it's given.
- A person confirms every S1 and anything about security or data.
- Remove card numbers and passwords from tickets before pasting; use the company's account.
- Read each first reply before it goes; it speaks for the company.
- Save the brief for the next batch.
Next step: replies in one tone
Once tickets are sorted, the replies themselves should sound like one team. Making your support team's AI replies sound like you covers the tone guide every agent's AI uses.
Skills for this
- Customer successSupport ticket triageTriages incoming tickets into a table with category, S1 to S4 severity, routing, and a first reply for each, sorted by what to answer first.
- Customer successCustomer reply in our toneDrafts a support reply in your team's tone: answers the question first, owns mistakes, drops jargon, and ends with one clear next step.
- Customer successFAQ from support ticketsTurns a batch of tickets or chats into an FAQ: questions in customer words, ranked by count, short answers, and flags for answers to confirm.
- Customer successBug report from a support ticketTurns a customer complaint into an engineering-ready bug report: steps to reproduce, expected vs actual, environment, impact, evidence, and questions to ask.
Questions
Can AI triage support tickets?
It can do the first pass on a batch: give each ticket one category, a severity from your written rubric with the line that decided it, an owner, and a short first reply, then sort the table so the worst comes first. A person checks the S1s and anything security-related before acting.
How do you set ticket severity?
From a written rubric, not the customer's tone. A common shape is four levels: S1 for many customers blocked, data loss, security exposure or money taken wrongly; S2 for one customer blocked on a core task; S3 for something with a workaround or a how-to; S4 for requests and cosmetic issues. When unsure between two, pick the higher.
What should the first reply say?
One to three sentences: what you understood, what happens next and when, and the one missing fact you need, such as a screenshot or the account email. No promise of a fix you can't yet give.
AI skill
Have your AI triage incoming support tickets
Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. With Support ticket triage, each person's AI triages a batch of tickets: one category each, an S1 to S4 severity from your rubric with the line that decided it, an owner, a first reply that promises no fix, and the table sorted by what to answer first.
14 days free · No card needed
Sources
- PagerDuty: Severity levels — a written scale; if unsure, pick the higher severity (read 2026-11-04)
- Atlassian: Incident severity levels (read 2026-11-04)
- Zendesk: Metrics that matter in customer support — first reply time (read 2026-11-04)