When a guest texts, Suzy first scores every FAQ in your bank against the guest's message using the Keywords field on each entry - this narrows things down to the FAQs that are actually relevant.
She then uses AI to write a natural-sounding reply, but only using the information in those matched FAQs - she doesn't invent policies or details that aren't in your dashboard.
If she isn't confident she has the right answer (below a set confidence threshold), she tells the guest she's checking with the front desk instead of guessing. That's a hard rule, not a suggestion - Suzy is built to say "let me check" rather than make something up.