Haven learns from your replies

You answer the same guest questions again and again: whether early check-in is possible, where to park, whether dogs are welcome. Haven reads the replies you write and turns your answers into AI knowledge, so its AI assistants give guests the answer you would give, in words that sound like yours. Everything it learns is listed under Messages → AI knowledge → Learned, with the replies it came from. You can keep, edit or remove each item.

When to use

There is nothing to set up. Learning is on for every account. Open the Learned tab when you want to:

check what the AI now tells guests because of your replies,
decide which answer stands when your recent replies disagree with an entry you wrote,
OK an answer that mentions a price,
turn learning off.

What Haven reads

Every thread Haven can see for your account: your Haven inbox, guest email, text and WhatsApp threads, and the booking threads your PMS shares with Haven (Airbnb, Vrbo, Booking.com and direct guests, through Guesty, Hostaway, Hospitable, Hostfully, Lodgify, OwnerRez, Track and the other connected PMSs).
Replies written by people. That means you and team members who can edit. Replies typed while someone from Haven support or a partner is working in your account never count.
Not automations. PMS scheduled messages, Haven's own automatic messages, and AI replies you sent without changing them are skipped. Some PMSs do not mark which messages were scheduled. For those, a message whose text appears in three or more of your threads is treated as a template and skipped.
AI drafts you edited count. When you correct an AI draft before sending it, Haven learns the answer you gave.

Haven reads a thread once it has been quiet for two hours. It learns only from messages sent since Haven began recording who wrote each one, which happened when this feature launched. In older messages it can't tell your own typing from a scheduled message, so it never learns from them.

What it learns

Answers. Each one becomes an AI knowledge entry marked Learned:

Where it applies. It covers the listing the thread was about. An answer you gave the same way for two or more listings applies to all your chats.
Who can see it. Public means any guest may know it (a pet policy). Private means only guests whose arrival details are unlocked get it (how to find the key box).
Why it was learned. Seen in N threads shows the reason and the replies it came from, with Open thread for each.
Kept up to date. Until you keep it, Haven updates a learned answer when your later replies say something new.

How you write. Your replies also teach your voice (tone, length, how you sign off). Those notes appear under How you write on the same tab, and in Settings → Account & Profile → AI assistant → Your voice. See Teach Haven your voice.

What it never keeps

Door, gate and lock codes, Wi-Fi passwords, phone numbers, email addresses and links.
Guest names, and anything about one guest or one stay ("your late checkout tomorrow is fine").
One-off deals and quotes.
The guest's own messages. The learner sees a short summary of what the guest asked, never their words.

How to review what it learned

Open Messages → AI knowledge and choose Learned. An amber number on the tab counts decisions waiting for you.
Under Which answer stands?, compare Your knowledge says with Your recent replies say:
- Use new puts the new answer into your entry.
- Keep current leaves your entry as it is. The same disagreement is not raised again unless you change that entry.
Under Needs your OK, check each answer that states a price (a pet fee, an early check-in charge). Guests do not get these until you press Keep.
Under Answers, look over everything the AI now uses:
- Keep makes an answer yours, and Haven stops updating it.
- Edit changes the wording, where it applies or who sees it. Saving an edit also keeps the answer.
- Remove takes it away, and Haven will not learn that topic again for that listing.
Open Recent reads to see each time Haven read your threads: what it added, updated or skipped, and why.

Learned answers also appear in the Sites view, under the listing they belong to, with a Learned pill. A disagreement shows only on the Learned tab. Until you decide, guests get your own entry.

Turn it off

Switch off Learn from my replies at the top of the Learned tab. Haven stops reading your threads. Everything it already learned stays, and you can still keep, edit or remove it. If you switch learning back on, Haven reads only replies you write from then on. Replies written while it was off are never read.

Key details

Learned answers reach every Haven AI assistant that uses AI knowledge: AI replies to OTA guests, inquiries included, and the chat on your sites and guest portal.
Your own entries always take priority. Haven never changes an entry you wrote or kept. When your replies disagree with one, it asks you.
Each listing holds up to 40 learned answers.
Learning from your replies does not use your AI credits.