Plan requirement
| Subscription | Suite Professional or higher, Support Professional or higher |
| Also required | Copilot |
| Access | Admin |
Your topic list is your own, so start by checking whether it holds a value that covers junk. Suppressing tickets automatically deserves care whichever value you use.
First, find the value
There is no universal spam topic. Topics come from the list on your own account, and you read it in Admin Center > AI > Intelligent triage > Topic. Values are hierarchical, written with two colons, as in Account::Account activation. Look for one that covers the junk you actually receive; if there is none, this rule has nothing to run on and the suspended tickets queue is the route instead.
Set it up
- Create a view with a condition on the Topic field set to that value. On accounts that took Copilot before 11 June 2026 the field is still called Intent.
- Read it for two weeks without acting on it.
- If it is accurate, add a trigger that moves those tickets out of the main queue.
- Keep checking that view.
Move, do not delete
Send them to a group or a status where they are out of the way and still recoverable. Deleting on an automatic label means a misclassified customer message is gone with no trace.
The failure mode is silent
A wrongly suppressed ticket produces no complaint, because the customer thinks they contacted you and you think nobody wrote. It surfaces weeks later as a customer who says they never got a reply.
That asymmetry is why this rule deserves more caution than its benefit suggests.
Read the suppressed pile weekly
Five minutes scanning subjects. Genuine messages stand out immediately, and the check is what makes automatic suppression safe to run at all.
Check which languages get classified
Topic and sentiment are assigned for 30 languages, and Dutch is one of them. Language detection covers a far longer list, so the two are easy to confuse: a ticket whose language Zendesk recognises is not therefore a ticket that gets a topic. What a ticket outside the 30 comes back with is not documented, so you cannot assume it arrives unclassified.
That makes such a rule one whose behaviour you have not seen for part of your customer base. If any of your customers write in a language off that list, watch what those tickets get labelled before you automate, or exclude them from the rule first.
The tickets this rule cannot see
With Ignore agent-initiated tickets switched on, a recovered suspended ticket gets no predictions. A rule on topic therefore never sees the tickets you rescued from the suspended queue, which is exactly the pile a spam workflow touches most. What intelligent triage does lists the other cases in which a ticket arrives without labels.
Watch the borderline cases
Short messages and unusual addresses are the other ones most likely to be misjudged. So are first messages from new customers, which are exactly the ones you cannot afford to lose.
Combine with something else
The topic plus no prior tickets from that requester is far safer than the topic alone. A returning customer is very rarely spam whatever the label says.
Comments
0 comments
Article is closed for comments.