What should an AI receptionist do when it doesn’t know the answer?

I’m building Reception AI, and one problem I’ve found much more interesting than simply making an AI answer the phone is deciding what it should do when it doesn’t know something. Imagine a caller asks: “Do you offer emergency appointments on Sundays?” If that information isn’t in the approved knowledge, I don’t think the AI should guess. A confident wrong answer can be much worse than admitting uncertainty. So we’re exploring a Knowledge Gap approach: Caller question → AI detects uncertainty → safe fallback → knowledge gap gets flagged → business provides the correct answer For higher-risk information like pricing, cancellation policies, availability, or promised callback times, the system could use stricter fallback rules. I’m curious how other founders and AI builders would approach this: When an AI receptionist doesn’t know something, should it immediately transfer to a human, collect the caller’s details for follow-up, or use a predefined fallback and continue the conversation? I’m especially interested in how you’d balance customer experience vs. preventing confident misinformation.