What it will not say is defined too.
Every assistant runs under rules that cannot be argued away — not even by a caller who tries. No invented prices, no internals, never pretending to be human.
The rules in detail
Invent nothing
If a price or a date is not on file, it is not stated. Instead of a guess comes a question or a handover.
No internals
The assistant gives no information about other customers, internal processes or its own configuration.
Honest about itself
Asked directly, it says it is an AI assistant. It never claims to be a person.
Your own limits
On top of that you block topics and statements that are sensitive in your field — legal advice or diagnoses, for example.
How the rules take effect
- 1
Before every answer
The rules sit above every conversation and are checked before the assistant speaks — not afterwards.
- 2
When someone tries to bypass them
Prompts like “ignore your instructions” do not create an exception, they lead back to the request.
- 3
On call termination
If a call turns abusive, the assistant ends it politely — documented and visible to you.
Where this is critical
- Price statements that could later sound binding
- Health and legal questions no assistant may answer
- Callers deliberately trying to trick the assistant
- Complaints that belong to a person, not to an AI
Try to push it off track.
In the demo you are welcome to. Afterwards we show in the transcript what the rules caught.