Communications

FAQ Knowledge Base

The FAQ knowledge base is the operator-curated source of truth the AI Suggest layer draws from when drafting guest replies. Every Q+A you add becomes a fact the AI can cite.

Anatomy of an FAQ entry

Each entry has:

Question — the canonical phrasing of what guests ask ("Do you have a parking spot?").
Answer — the response you want the AI to give (or a starting point).
Property scope — applies to all properties, or a specific one.
Tenant scope — applies tenant-wide.
Tags — informal labels (parking, wifi, pets, checkin) used to prioritise relevance.
Language(s) — entries are usually written in the operator's language; the AI translates as needed at suggestion time.

Where to edit

Operations → FAQ (or via the Add FAQ entry task that fires on common-pattern detection). The editor uses the same placeholder syntax as task templates — you can write "Parking is at {{property.parkingInstructions}}" and the answer is per-property personalised.

How the AI uses it

When AI Suggest builds a reply:

Loads the FAQ scoped to (this tenant, this property).
Loads the inbound message.
Asks the LLM to use the FAQ as primary context and the message thread as secondary context.
The LLM is instructed to only invent facts not in the FAQ when the user's question doesn't match any entry — at which point it tends to be conservative ("Let me check and get back to you").

The more complete your FAQ, the more accurate and on-brand the AI's replies. A common pattern: when you find yourself manually replying to the same question twice, file it as an FAQ entry.

Adding via "Add FAQ entry" task

A task template can fire when a guest asks a question the AI couldn't answer with high confidence. The task lands in your queue with the inbound message attached; you draft an FAQ entry inline and save. Future similar questions get covered.

⚠ Before #185, the Add FAQ entry task action would fail with "Property not found" due to a {{property.id}} placeholder schema regression. Fixed in commit referenced in the issue.

Best practices

Keep answers concise — one or two sentences. Long answers feel canned.
Per-property facts go in the property's config, not in FAQ — WiFi password, parking spot number, key location.
Tenant-wide facts go in FAQ — refund policy, late check-in policy, smoking rules.
Tag aggressively — wifi, parking, pets tags help the AI prioritise.
Review periodically — out-of-date FAQ is worse than no FAQ. Review monthly.

#185 — Add FAQ entry task action fix (placeholder schema regression).
#78 — LLM prompt injection via FAQ + knowledge base (defence-in-depth, no operator-visible change).