Zendesk Intelligent Triage: How to Set It Up and Use It

Intelligent triage is the part of Zendesk that reads a ticket before a person does and writes down what it thinks the ticket is about. It is a genuinely good feature and it is also the one people most often turn on and then forget, because switching it on produces no visible change to anything. The value arrives at the second step.

This guide covers what it detects, how to switch it on, how to turn a classification into a routing rule, and the terminology change that will confuse you if nobody warns you about it first. That last one is worth reading even if you already have triage running.

What does Zendesk intelligent triage actually detect?

Four things, according to Zendesk's documentation on automatic classification. Each one lands on the ticket as a field with an associated confidence score, and agents can change a value if the prediction is wrong.

• Topic: what the ticket is about, drawn from prebuilt industry values or from values you define yourself.
• Sentiment: the customer's emotion, on a five point scale from Very Positive through Neutral to Very Negative.
• Language: the language of the request, across roughly 150 languages per their documentation.
• Entities: specific pieces of information inside the ticket, such as a product name.

Channel coverage is broader than most people assume, and it includes closed tickets, which is useful when you want to classify history rather than only new arrivals.

On availability, their documentation lists intelligent triage on Suite and Support Professional plans and above, and states that using the classifications in workflows requires the Copilot add-on. It also notes that the Copilot add-on turns on all classification types by default, so there is nothing to enable type by type once you have it.

Why does my account say Intent when the docs say Topic?

Because Zendesk renamed it and the rename is still working its way through the product. Their announcement, published on 26 May 2026, describes the term Intent becoming Topic across intelligent triage in the Copilot add-on, with the rollout beginning on 11 June 2026 and completing on 18 June 2026.

The announcement is emphatic that this is a naming change only, with no impact on workflows, ticket data or functionality, and no setup or configuration needed. The custom ticket fields named Intent and Intent confidence keep their original names for now, and Zendesk says you may see both terms for a while in the Agent Workspace and in Admin Center, with those field names scheduled to change later in 2026.

So if a colleague says intent and a help page says topic, they are talking about the same thing. Tell your agents before they file a bug report about it.

How do you switch intelligent triage on?

In Admin Center, under AI, then Intelligent triage, then Topic, using Manage settings. The main thing you are doing here is telling Zendesk which channels to classify. Their documentation notes that some channels are already added by default, so check the list rather than assuming it is empty.

1. Open the triage settings — In Admin Center, click AI in the sidebar, then select Intelligent triage, then Topic. Click Manage settings.
2. Add the channels you care about — Add the channels your customers actually use. Some are already added by default, so review what is there before adding more.
3. Save — Click Save. At this point tickets start being classified, but nothing else in your workflow changes yet.
4. Wait for a few days of real tickets — Predictions on your own data are the only way to judge whether the topics fit your business. A day is not enough to tell.
5. Review the topics against reality — Look at what got classified and what did not. Missing coverage here is what you fix in the recommendations step later.

Whichever route you take, the thinking is the same and it is worth doing on paper first. Write out the three categories of ticket that most need to reach a different person, and what should happen to each. Everything after that is mechanical.

How do you turn a classification into a routing rule?

With a ticket trigger. Zendesk's setup documentation walks through building one that reads the topic field and acts on it, and this is the step that turns triage from an interesting label into something that saves your team time.

1. Open the trigger builder — In Admin Center, go to Objects and rules, then Business rules, then Triggers. Click the Ticket tab and select Create trigger.
2. Name it so future you understands it — Fill in the Name, Description and Category. A description that says why the rule exists saves an argument six months from now.
3. Add the topic condition — In Conditions, add the topic field and the value you want to match. Zendesk's own example matches a topic under an account category, such as account activation.
4. Add the action — In Actions, set what should happen. Their example assigns the ticket to a specific group, which is the most common use of triage.
5. Create the trigger and watch it — Click Create trigger, then check the next few matching tickets by hand. A rule that fires on the wrong topic is worse than no rule at all.

Start with one trigger, not ten. The failure mode with rule based routing is not that any single rule is wrong, it is that a dozen rules interact in ways nobody modelled, and then a ticket lands somewhere nobody is looking. One rule that clearly works beats a full set that nobody can reason about.

Where do agents see the classification?

In two places, and you control both. Zendesk's documentation describes topic and sentiment appearing in the ticket header, controlled from Admin Center under AI, then Agent copilot, then Suggestions, where there are toggles for showing topic in the header and showing sentiment in the header.

The second place is the ticket form itself, where topic, sentiment and language appear as fields an agent can review and change. Field visibility is managed under Objects and rules, then Tickets, then Forms. Their documentation notes a behaviour difference worth knowing: with a single ticket form the fields are visible automatically and you remove them with the x icon, while with multiple forms they are not automatic and you drag them onto each form you want them on.

That difference explains a common complaint, which is that triage seems to work for one team and not another. Usually the second team is on a different form that nobody added the fields to.

How do topic recommendations work?

Zendesk generates suggestions from your own ticket data rather than making you guess at a taxonomy. Their documentation describes weekly, data driven recommendations for new topics tailored to your account, plus a second kind that flags duplicate or overlapping topics when you change your structure.

You find them on the Topic page under AI, then Intelligent triage, then Topic, in a Recommendations tab, with a banner above the topics list when new ones exist. Accepting one uses Add topic and it starts classifying incoming requests immediately. Declining opens a dialog where you say why. Their documentation says recommendations begin roughly a week after you activate triage, and that manual approval is always required.

This is the part of the feature we like most, and it is the same instinct behind how a Sistava AI Employee works: watch real work, propose, wait for a human yes. A system that quietly changes its own categories is a system nobody trusts by month three.

Sources

• Zendesk help, Automatically classifying tickets with intelligent triage
• Zendesk help, Setting up intelligent triage topics
• Zendesk help, Configuring intelligent triage for the agent experience
• Zendesk help, Reviewing personalized intelligent triage topic suggestions
• Zendesk help, Announcing the Intent to Topic terminology update in the Copilot add-on
• internalnote, Deep-dive into Zendesk Intent Detection