Generate from Conversations lets the AI scan your inbox history and propose Knowledge Base entries automatically — so you don't have to write every entry from scratch. Generated items are always saved as a draft batch for you to review before anything goes live.
For the main Auto Pilot setup guide, see Hostify AI — Auto Pilot Mode. For best practices on writing KB entries that the AI agent reads at reply time, see Auto Pilot — Knowledge Base & Content Best Practices.
Generate from Conversations is currently in Beta.
What it does
The generator reads your guest-host inbox conversations for the period you choose, then asks the AI to extract three kinds of content:
General KB entries — broad topics that apply across all or most properties. Saved as Apply to all properties KB entries after you accept them. Examples: "Check-in coordination", "Urgent support & communication method", "Long-stay billing".
Listing-specific KB entries — operational details unique to one or a few listings. Scoped to those listings on save. Examples: "Pool & Outdoor — At Villa Margarita the pool opens mid-May…".
Suggested instructions — behavioral patterns the AI noticed your team applies in practice. Not auto-saved; surfaced with a Copy button so you can paste them into Auto Pilot's Custom Instructions.
The generator focuses on operational knowledge — how-to procedures, troubleshooting, local tips, nuanced policies. It deliberately skips topics that already live in standard listing data (descriptions, amenities, check-in/out times, pricing, written house rules, photos, Wi-Fi, tax IDs).
When to use it
First time setting up Auto Pilot. Run it once to seed your KB from your existing inbox patterns instead of writing every entry from scratch.
After acquiring new properties. Once the new listings have a few weeks of conversations, run a generation to capture their quirks.
Quarterly knowledge audit. Run it on the last 3 months to see what new questions guests are asking and update your KB.
After staff turnover. Surfaces patterns the team is currently answering inconsistently and forces a written-down standard.
Where to find it
Integrations → Hostify AI → Knowledge Base. The Generate from Conversations button sits next to Add at the top right of the KB list.
How it works (in plain English)
You click Generate and pick a time period (last month, last 3 months, last 6 months, or last year).
The system collects your conversations — up to 5,000 threads total, capped at 40 threads per listing so big customers don't have one listing hog the whole sample.
Pass 1 — General patterns. The AI looks at a sample across all your listings and extracts general KB and instructions.
Pass 2 — Per-listing batches. The AI then processes your listings in groups of 5 and extracts listing-specific entries for each one. For 50 listings that's 10 batches; for 100 it's 20.
You review the draft batch. The result lands on a review page split into the three sections above.
Save All publishes the accepted entries into your live KB.
Total time: usually 3–10 minutes depending on how many listings + conversations you have. The progress banner stays visible across page navigations — you can leave the page and come back.
Configuration options
Time Period
Pick the recency of conversations to analyze:
Last month — quick check-in on what's changed recently. Smaller sample → may miss seasonal questions.
Last 3 months (default) — best balance for most accounts.
Last 6 months — picks up seasonal patterns (summer vs. winter property quirks).
Last 12 months — fullest coverage. Use it once a year to catch year-round patterns.
Generate from scratch (toggle, on by default)
On (default) — existing KB is ignored. The AI generates everything from scratch as if your KB were empty. Best for the first time you run this on an account, or whenever you want to re-evaluate the whole KB. The review page may show some entries that already exist in your KB — reject the duplicates.
Off — your existing published KB is sent to the AI as context, and the AI is told not to duplicate topics you already have. New runs produce mostly net-new entries. Use this for incremental updates after the first full run.
The review page
After generation completes, you land on a three-section review page.
Section 1 — General Knowledge Base Entries
Topics that apply across all or most of your properties. Each card shows a topic and a comprehensive answer covering several related sub-questions. Per-card actions:
Accept — marks the item as keep. Stays as a draft until you click Save All.
Reject — marks the item as discard. Will be deleted on Save All.
Edit — inline edit of the topic + answer.
Show more — expands the answer preview if it's long.
Section 2 — Suggested Instructions & Policies
Grouped behavioral directives the AI noticed your team applies. Each card has a topic, a bullet list of directives, and a Copy button. There's also a Copy All Instructions button at the top of the section.
Suggested Instructions are copy-only. They never become KB entries. The intended workflow: copy them, then open Integrations → Hostify AI → Auto Pilot → Custom Instructions, paste, edit to taste, save.
When you click Save All, the suggested instructions are removed from the draft batch (they were never going to live in the KB).
Section 3 — Listing-Specific Items
KB entries that apply to one specific listing (or to a small group of listings that share identical content — e.g. all units in the same building with the same parking rules). Items are grouped by listing nickname with collapsible headers; click a header to fold/unfold its items. The same Accept / Reject / Edit / Show more actions apply per-card.
Footer action bar
Accept All KB — bulk-accepts every General KB and Listing-Specific item.
Discard All — deletes the entire draft batch.
Save All — publishes every accepted item, deletes every rejected item, and removes the Suggested Instructions drafts.
Accept vs. Save — the exact rules
There are two layers and they trip people up the first time.
Accept / Reject per-card — these mark each draft item, but nothing is finalized yet.
Save All in the footer — this is the commit step. It does:
Every accepted General KB and Listing-Specific item → flips to published status, becomes a live KB entry.
Every untouched item → also flips to published (anything you didn't explicitly reject is treated as accepted).
Every rejected item → deleted.
Every Suggested Instruction draft → deleted (instructions are copy-only).
If you walk away without clicking Save All, the draft batch sits in your account and you'll see it again as Pending Review the next time you open the KB page.
Pending Review banner
If you have a draft batch that hasn't been saved or discarded yet, the KB index page shows a yellow banner at the top:
Pending Review — N AI-generated items awaiting review [ Review ] [ Discard ]
You can't run a new generation while a draft batch exists. The Generate-from-Conversations modal will prompt you to Review Batch or Discard & Start Over until you handle it.
Best practices
Start with the default 3 months. It's the right balance of recency vs. sample size. Don't reach for "last 12 months" on the first run — review fatigue is real and 200+ items take time.
Run it after you have at least a few weeks of conversations. With less than ~50 threads total there isn't enough repetition for the AI to find patterns.
Always review item-by-item, at least the first time. The first pass is the AI's best guess at what your operational knowledge looks like. After the first proper review the system has a clearer picture and produces better incremental drafts later.
Reject aggressively. Better 30 great entries than 100 mediocre ones. The agent reads your KB on every reply — bloated KBs slow it down and dilute relevance.
Edit topic lines to phrase them as the guest's question. "How does the pool heater work?" beats "Pool & Outdoor". Topic phrasing is the agent's first signal that an entry is relevant.
Don't rerun back-to-back. Generations cost real money on the LLM side. After Save All, give the system a week of fresh conversations before another run.
Generate-from-scratch is on by default because the first run on any account benefits from a fresh look at all conversations. After your first proper review, switch it off for follow-up runs so the AI doesn't re-propose topics you already accepted last time.
For multi-unit setups, double-check the listing scope of items you accept. Items default to the parent listing — if it's genuinely unit-specific (different door codes, etc.), edit the entry post-publish to attach it to the correct child unit individually.
Copy the Suggested Instructions into Auto Pilot Custom Instructions. They're surfaced separately because they're behavioral patterns for the agent to follow, not facts to look up.
Permissions
View the Knowledge Base page, run Generate from Conversations, review and accept/reject items, save or discard a draft batch:
hostifyai/knowledgebasepermission. The Hostify AI integration must be active.Edit existing KB entries:
hostifyai/managepermission.
Troubleshooting
"A generation job is already running" — another generation is still in flight. Look for the progress banner on the KB index page; if there isn't one, wait a few minutes for the lock to expire.
"You have an unreviewed generation batch" — an older draft batch exists. Click Review on the Pending Review banner or use Discard to clear it before generating again.
Banner stuck at "AI is analyzing..." for >10 min — the LLM call is slow but not necessarily hung. The system will time out automatically if needed and report Generation failed.
Generation completed with very few items — the time window may be too short or have too little conversation volume. Try a longer time window. If you ran with Generate from scratch off, the AI may have skipped many topics that already exist in your KB — re-run with the toggle on.
Some items duplicate entries already in my KB — expected when Generate from scratch is on (the default), since the AI doesn't see your existing KB. Either reject the duplicates on the review page, or uncheck Generate from scratch to tell the AI about your existing KB so it skips known topics.
Suggested Instructions disappeared after Save All — this is intentional. Use the Copy buttons before clicking Save All.
What it does not do
It does not push generated content to channels (Airbnb, Booking.com, etc.). KB entries are read by the Auto Pilot agent at reply time; they are not channel listing descriptions.
It does not modify your Auto Pilot Custom Instructions automatically. Suggested Instructions are surfaced for you to copy manually.
It does not bypass the review step. Generated items are always drafts until you Save All — nothing goes live without your click.
It does not replace the Add / Edit flows. You can still manually create or edit any KB entry; generation is one source, manual is another, both end up in the same KB.
Related articles
Hostify AI — Auto Pilot Mode (main setup guide)
Auto Pilot — Knowledge Base & Content Best Practices (Custom Instructions, KB writing tips, Mr.GG)