Redaction suggestions and regular expressions

Zendesk highlights twelve kinds of personal data for a person to confirm, and you can add up to ten patterns of your own for the data only you would recognise.

Zendesk highlights twelve kinds of personal data for a person to confirm, and you can add up to ten patterns of your own for the data only you would recognise.

Set them up

1. Turn redaction suggestions on in Admin Center and choose which PII types to detect.
2. Decide when agents see the suggestions.
3. Add custom suggestions where you need them, using a RegEx pattern.
4. Use the Test detection field to paste example text: if Zendesk finds the value, the pattern is right, and if not, the pattern needs work.
5. Tell agents what a highlight means and that they may dismiss one.

What it recognises on its own

First name or surname, email address, address, credit or debit card number including last-four-digits-only forms, IBAN, password, Social Security Number, bank account number, date of birth, driver's licence, IP address and phone number. It works across languages, and context matters to it: a date is read differently depending on what surrounds it.

Findings are highlighted in orange in the comment. Admins and agents with the redact ticket content permission see them; light agents do not, so redacting cannot sit with somebody on a light seat.

PDF attachments are all-or-nothing

A PDF containing PII anywhere is flagged as a whole. You can redact the entire attachment, but not a single passage inside it.

Ten patterns of your own

The built-in types do not know your customer numbers, contract references or internal identifiers. For those, create a custom redaction suggestion with a RegEx pattern, up to ten in total. PII found through a custom suggestion is redacted by your redaction triggers as well, so the pattern you write here defines what automatic redaction can act on later.

Writing one

Be specific. A pattern for "any long number" matches order references, timestamps and phone numbers, and an agent who sees three false suggestions per ticket stops reading them. Start narrow and widen only when you see real cases being missed.

Test both directions

Real examples that should match, and near misses that should not. The second set is what tells you whether the pattern is usable, and the Test detection field is there for exactly this.

Suggest, do not remove

Automatic redaction is right for card numbers, where the pattern is unmistakable and the risk of keeping one is high. For everything else a suggestion lets a person judge, and judgement is what prevents the wrong text being destroyed.

Not alongside card redaction

The account-level setting that automatically redacts credit card numbers is incompatible with redaction suggestions, and Zendesk advises against running trigger-based redaction next to it. If card numbers matter to you, decide which of the two mechanisms handles them before you turn either on.

Say what to do with a suggestion

Confirm it if it is genuinely sensitive, dismiss it if not. Agents need to know they may dismiss, or they will redact everything flagged, including the order number somebody needed. After a month, look at whether suggestions are being accepted or dismissed: consistently dismissed means the pattern is wrong, and that is a fix rather than a training issue.

See also

• Redacting sensitive information with triggers
• Redacting identified personal data
• Automatically redacting credit card numbers