Can I change the categories or add my own?

Yes. The category list lives in the prompt file and the routing table, both included in the package. Add a row to the table and a line to the prompt, re-import, and the next ticket uses it. No code changes are involved.

What happens if the model returns something unexpected?

Every reply is checked as JSON before anything is written. If the category is not in your table, the ticket gets a triage-unclassified tag and a private note, and nothing else changes. A Zendesk view filtered on that tag lists them for a human. Nothing is guessed onto the ticket.

Does the customer see any of this?

No. Tags, group and priority are internal fields, and the reason is written as a private comment. The flow never sends a public reply and never changes the ticket status.

Will this hit Zendesk or OpenAI rate limits?

Each ticket costs one Zendesk read, one update and one model call. Zendesk publishes a per-minute API limit for each plan, 200 requests a minute on Team, so a queue measured in hundreds of tickets a day stays well inside it. OpenAI limits scale with your usage tier and are not a concern at this volume.

Can it run on Make or Zapier instead of n8n?

Yes, and both versions ship in the package. On Zapier the urgency branch uses Paths, which needs a Professional plan or above; on Make it is a Router, available on every plan.

ZdAISMade to order

Customer support & tickets

Zendesk new ticket → AI triage: tags, group, priority + Slack for urgent

Every new Zendesk ticket is tagged, routed and prioritised before an agent opens it.

New tickets sit untouched until someone reads them and decides where they go. This flow reads each one the moment Zendesk creates it, classifies the request and its urgency with GPT, then writes the tag, group and priority back to the ticket with a private note explaining why. Urgent ones also land in a Slack channel with a link. Agents open a queue that is already sorted.

How it works

Zd
Trigger
Ticket created: a Zendesk trigger fires a Zendesk webhook (Admin Center → Webhooks) at the n8n Webhook node with the ticket ID
Zendesk
Zd
Step 1
Load the full ticket
Show Ticket returns subject, description, requester, channel and any tags the web form or an email rule already set.
AI
Step 2
Classify the request and its urgency
GPT reads subject and description and returns JSON with a category (billing, bug, how-to, account, cancellation, other), an urgency level and a one-line reason.
if
Step 3
Map the category to a tag, group and priority
A lookup table in the workflow turns each category into a Zendesk tag, a group ID and a priority; a category the table does not recognise gets the triage-unclassified tag and the note, nothing else.
Zd
Step 4
Write the triage back to the ticket
Update Ticket sets tags, group_id and priority, and adds the model’s reason as a private comment only agents can see.
if
Step 5
Stop unless the urgency is high or urgent
Normal and low tickets end here; only the two top urgency levels continue to the alert.
S
Step 6
Post the urgent ticket to Slack
A message in the channel you choose carries the ticket number, subject, requester email, category and a direct link into Zendesk.

What changes once it runs

Tags, group and priority are on the ticket within about a minute of it being created, before any agent has opened it.
Every classification comes with a private note giving the reason, so a wrong call is visible and can be corrected on the spot.
High and urgent tickets appear in Slack with a direct link; everything else stays out of the channel.
Routing rules live in one lookup table inside the workflow, so a new category is an edit, not a rebuild.
The model never replies to the customer; it only sets fields an agent can override.

What you get

n8n workflow export, plus Make and Zapier versions of the same flow
A setup guide covering the Zendesk webhook and trigger, API token, OpenAI key and Slack app, with screenshots
The classification prompt as an editable file, with the category list and the JSON shape it must return
A routing table template mapping category to tag, group ID and priority
A recorded walkthrough of the first ticket passing through end to end
30 days of email support for setup questions

Before you start

Zendesk Support or Zendesk Suite on any plan, with admin access to create a webhook and a trigger, plus an API token
An OpenAI API key with a payment method on file
A Slack workspace where you can install an app or bot with permission to post in the alert channel
Group IDs from Zendesk for each destination team (the setup guide shows where to find them)
An n8n webhook URL Zendesk can reach: n8n Cloud, or a self-hosted instance with a public address

Built for

Questions before you buy

Can I change the categories or add my own?: Yes. The category list lives in the prompt file and the routing table, both included in the package. Add a row to the table and a line to the prompt, re-import, and the next ticket uses it. No code changes are involved.
What happens if the model returns something unexpected?: Every reply is checked as JSON before anything is written. If the category is not in your table, the ticket gets a triage-unclassified tag and a private note, and nothing else changes. A Zendesk view filtered on that tag lists them for a human. Nothing is guessed onto the ticket.
Does the customer see any of this?: No. Tags, group and priority are internal fields, and the reason is written as a private comment. The flow never sends a public reply and never changes the ticket status.
Will this hit Zendesk or OpenAI rate limits?: Each ticket costs one Zendesk read, one update and one model call. Zendesk publishes a per-minute API limit for each plan, 200 requests a minute on Team, so a queue measured in hundreds of tickets a day stays well inside it. OpenAI limits scale with your usage tier and are not a concern at this volume.
Can it run on Make or Zapier instead of n8n?: Yes, and both versions ship in the package. On Zapier the urgency branch uses Paths, which needs a Professional plan or above; on Make it is a Router, available on every plan.

Nearby on the shelf

Whole catalog