Your team needs guidance it can use while doing the work. Start with a short agreement covering tools, information, permissions, review and responsibility.
Name the approved tools and who can authorise a new one. Set boundaries for customer details, confidential documents and unpublished business information, based on your obligations and the service settings you have checked.
Define what AI may prepare and what it may publish, send or change. Require approval where a mistake could affect trust, money or an individual. Assign an owner for exceptions and incidents.
For marketing outputs, review factual claims, representation, permissions and potentially misleading imagery. Agree when people should be told that content or an interaction uses AI.
Test the guidance with an everyday scenario: a colleague wants to upload customer interview notes and generate testimonials. Can they tell what is allowed, what requires permission and who to ask?
Keep a simple record of decisions and review the guidance when tools or workflows change. Guardrails become useful when people understand how to apply them.