What rules should govern a customer AI assistant?

A customer AI assistant should be governed as a brand behaviour because every response shapes the customer experience, not just the wording of an answer. Voice and tone, content strategy and visual style can make an interaction recognisably on-brand, but they do not tell an assistant how to make decisions under pressure or resolve inconsistency. Rules should translate brand values into observable behaviours: how the assistant explains complexity, treats uncertainty, handles frustration, protects customer dignity and prioritises useful next actions. Those behavioural rules need to be defined before chatbot design, prompt writing or AI content production. A customer-facing assistant is part of a B2B customer experience, where accuracy, ease of doing business and supplier confidence matter alongside brand consistency.

Clear response boundaries make responsible AI practical by defining what the assistant may answer, which trusted content it may use, what it must disclose and when human handoff is required. An assistant should distinguish supported information from uncertainty rather than inventing a confident answer. It should also be clear from the first interaction when a customer is interacting with AI where that is not already obvious. Human escalation rules should reflect the stakes of the moment. Routine, factual tasks may suit a direct and transactional interaction, while sensitive, exceptional or emotionally difficult situations may require a person without relying on a complex decision tree. Structured rules, trusted content sources and human oversight help personalised answers remain accurate, relevant and valuable rather than generic automation presented as intimacy.

We believe a customer AI assistant should express a company's brand through its decisions as well as its language. Tone-of-voice guidance matters, but it is too narrow to govern an assistant handling thousands of varied interactions, especially when a customer is frustrated or needs an exception. We start with the behaviours a brand intends to demonstrate, then turn them into practical rules for response, uncertainty, escalation and use of knowledge. We have applied that approach with UST in developing NavigatorAI, a generative AI assessment experience that used structured rules, trusted content sources and human oversight to provide tailored recommendations. The work combined prompt design with supporting content and ongoing refinement, rather than treating AI as a layer added after brand strategy.