How it stays on-voice, and who can change it
A fair question to ask before a tool like this goes wider: can someone pull it off-brand? The short answer is no, not by using it. The agent is locked to the approved voice by design, it does not learn from anyone’s brief, and the only way to change what it knows is a reviewed change made by an approved trainer. Here is exactly how that works.
Why it cannot drift
It writes from a fixed, approved voice
Every draft is grounded in a set of approved knowledge files: the voice codebook, the customer persona, the brand foundation, the product reference, the on and off-voice examples, and the approval checklist. That set is the agent. It does not write from anywhere else.
It does not learn from briefs
Typing a brief never changes what the agent knows. Two people who type the same brief get drafts grounded in the same approved voice. Nobody trains it by using it, so no one can quietly pull it off-brand.
Changing it is a reviewed, tracked change
The only way to change what the agent knows is to change a knowledge file. That is a deliberate, version-controlled edit by an approved trainer, with a record of who changed what and when. The agent picks it up on the next publish. There is no silent drift.
Who can change it
Three roles, with a clear line between using the agent and changing it. Who holds each role is yours to decide.
| Role | Can | Cannot |
|---|---|---|
| Approved brand trainer | Owns the voice codebook, persona, and brand foundation. Approves any change to how the agent sounds or who it speaks to. | N/A. This is the role that decides voice. |
| WiscAI engineering | Makes the reviewed knowledge-file change on the approved trainer’s say-so, and keeps the version history. | Never changes the voice or the brand position on its own. It only carries out an approved change. |
| Everyone else (the pod and the wider team) | Uses the agent, types briefs, gets drafts, edits the output like any other draft. | Cannot change what the agent knows. Their briefs never become training. |
How a change happens
- 1An approved brand trainer asks for a change. For example: stop using a phrase, add an on-voice example, update a product or a price.
- 2The change is made to the specific knowledge file it belongs to, and nowhere else.
- 3It is version-controlled: the date, the person, and the exact change are recorded.
- 4The agent picks up the change on the next publish. Until then, it keeps writing the approved way.
What is locked
The agent knows the brand through these pieces. Each is owned by a role and changes only through the process above.
| What it is | Owned by |
|---|---|
| How the brand talks: the words it uses and avoids | Approved brand trainer |
| Who the customer is, and the segments | Approved brand trainer |
| What the brand stands for | Approved brand trainer |
| The real products and prices it may name | Brand and merchandising |
| Examples to write like | Approved brand trainer |
| Examples to avoid | Approved brand trainer |
| The rules a campaign is checked against before it ships | Brand and compliance |
| The limits of each channel (lengths, sizes) | WiscAI and brand |
Making changes on your own
Today, changes route through us: an approved trainer says what to change, and we make the tracked edit. In a later phase, approved trainers can make those changes themselves, directly in the tool, without waiting on us. The guardrails do not change. It is still only approved trainers, every change is still tracked, and the agent still never learns from a brief. What changes is that the controls move into your hands.