Skip to main content
Brand Voice teaches AI Assistant recurring phrasing your support team actually uses in public Zendesk replies. Open AI → Brand Voice. Brand Voice is different from manually writing “sound friendly” into a prompt. RAUM mines repeated language from real agent replies, shows the evidence behind each suggested phrase, and requires an administrator to approve what becomes active.

​
What RAUM analyzes

A Brand Voice mining pass reads approximately the last six months of public agent replies available through the connected Zendesk environment. It looks for repeated patterns such as:
  • openings;
  • closings; and
  • common connective phrasing.
The review UI shows how often a phrase appeared across tickets and distinct customers. RAUM also applies recurrence thresholds before suggesting a phrase. The current thresholds are shown in the Brand Voice interface rather than hardcoded here.

​
Drafts do not affect replies

A newly mined profile starts as DRAFT.
Nothing in a draft Brand Voice profile reaches a customer reply until an administrator approves and activates it.
This makes the mining process observational first. RAUM can discover a phrase without silently turning it into policy.

​
Review a profile

Each profile shows:
  • language;
  • status;
  • number of analyzed tickets;
  • mined date;
  • opening phrases;
  • closing phrases; and
  • common phrasing, including ticket/customer support counts.
During review you can remove individual phrases before activation.

​
Profile actions

Administrators can:
  • Save edits after removing unwanted phrases;
  • Approve & activate a draft;
  • Disable an active profile;
  • Re-activate a disabled profile; or
  • discard a profile entirely.

​
Refresh from newer tickets

A later mining pass can produce a newer version of an existing profile. When that happens, the currently approved phrases remain live until you explicitly pull in the newer mined version for review. This prevents a background refresh from changing customer-facing language without human review.

​
House voice and agent evidence

The Brand Voice system can represent organization-level voice material and mined agent-level evidence. Administrators own the approval process rather than individual agents silently activating their own profile.

​
Brand Voice versus Guidelines

Use Brand Voice for language patterns that genuinely recur in your team’s replies. Use AI Guidelines for intentional rules such as:
  • required terminology;
  • forbidden phrases;
  • response structure;
  • legal/compliance wording; or
  • language-specific Translate rules.
A phrase being common does not automatically make it correct policy. If the team has repeatedly used an outdated phrase, remove it from the voice profile and fix the underlying guideline or procedure.

​
When to rerun mining

Consider a new mining pass after:
  • a major support-style change;
  • a new product or brand rollout;
  • substantial team turnover;
  • revised terminology; or
  • enough new ticket volume that recent replies better represent current practice.

AI Guidelines

Define intentional response and translation rules.

AI Coaching

Use poor outcomes to improve procedures, guidance, and AI memory.