The Knowledge Base holds facts: what is true about a property. Workflows hold procedures: what to do when a situation occurs.

That split is the single most important thing to get right here. "Checkout is 11 AM" is a fact and belongs in a host note. "If they ask for a late checkout, check the turnover and then offer or decline" is a procedure and belongs in a Workflow.

Putting procedure into the Knowledge Base produces confidently wrong answers, because everything here is treated as unconditional fact. Besty will not infer the "if" part. It will simply state the thing.

Go to Knowledge Base, then Host Notes. Each note is a card holding the note body, editable in place, an Apply to scope dropdown with a listing selector, and Save and Delete.

Scope is per note and is not global by default. Every note has to be pointed at a listing, a listing set, or all listings. A note that is not scoped where you expect is the usual reason Besty does not know something it appears to have been told.

The toolbar has a search box, a blue plus to add a note, and scope filter chips for All notes, All listings, Listing set and Individual, with the total count and pagination on the right.

Because scope is set per note, those filters are how you find notes that were pointed at the wrong property.

Beneath each note, Besty splits the content into two labelled panels: VALID INFORMATION with green ticks, and INVALID INFORMATION with red crosses. A single note can produce entries in both.

The pattern is that concrete, universal statements tend to validate, while compound, narrative or conditional ones tend to get flagged. A note that says three things at once will often have one part accepted and another rejected.

Treat these panels as guidance rather than a gate. The classifier is not deterministic, so rewording a flagged note and resubmitting it is not a reliable fix and can absorb a lot of time for nothing. If a note is flagged, the more productive move is usually to split it into separate, simpler notes, one fact each.

Keep notes general to the property, not specific to one guest or one moment. Anything you write is applied to every guest at that listing, so a note about one guest's circumstances gets applied to all of them. A qualifying preamble does not reliably prevent that.

There is also no need to copy in content your PMS already holds. Listing descriptions and automated messages are already read by Besty, so duplicating them here creates two places to maintain and two things to disagree with each other.

Under Knowledge Base, then Suggested Host Notes, Besty proposes facts it extracted from real conversations. When a team member tells a guest something new, it turns up here as a suggestion rather than silently entering the Knowledge Base.

Each suggestion carries the proposed text, editable before you accept it, an Apply to scope dropdown, Accept and Reject, and a metadata row showing a Pending badge, the listing, and a link back to the source conversation it was learned from. That link is the important part, because it lets you check the context before accepting something out of it.

Scope still has to be set when you accept. Suggestions are not global.

Pending suggestions are not reliably used in answers until they are approved, so a large backlog shows up as Besty being inconsistent: it knows something in one conversation and not in the next.

Review the queue at least weekly, ideally every day or two, and aim to keep it at zero. It is a short job when done often and a long one when left.

Reach out via chat or at support@getbesty.ai.