Onboarding

Smart guide import: turning your house manual into AI knowledge

Upload your existing PDF house manual and our AI extracts the structured fields — check-in times, Wi-Fi, parking, house rules — into your property guide automatically.

6 min read Published April 22, 2026 Category: Onboarding

Summarize this article with ChatGPT

Short on time? Open ChatGPT with a prefilled prompt that summarizes this post for you in 5 bullets.

Summarize →

Every short-term rental host already has a house manual. It might be a Word document, a PDF handed to cleaners, a Notion page, or a series of images saved to a phone. It contains the check-in instructions, the Wi-Fi password, the parking details, the building quirks, the closing-time of the bakery next door. Every apartment has one because every host has answered the same questions enough times to know what to write down.

When you sign up for an AI host platform, the system needs to know what's in that manual. If it doesn't, it can't answer guest questions. Traditionally this means a data-entry marathon: open the AI platform, find the "apartment guide" page, paste your manual into a textarea, save. Repeat for every apartment.

For a host with five apartments and different guides per apartment, that's two hours of typing and copy-pasting. Which is enough friction that many hosts try the AI platform once and never get past onboarding.

So we built a shortcut.

What smart guide import does

Upload the PDF of your existing house manual. We extract the text, identify typed fields automatically, and populate your property guide with them. For a standard 3-page house manual, the extraction takes about 8 seconds. The result is a structured guide — not a giant block of text — with distinct fields for check-in time, Wi-Fi SSID, Wi-Fi password, door code, parking instructions, and house rules.

The value of the structure isn't obvious until you use it. When the AI answers "what's the Wi-Fi password?", it doesn't have to fuzzy-match the question against a sea of text. It looks up the wifi_password field. Fast, accurate, never wrong.

Why "typed fields" matter

There's a real question of whether structured fields are worth the effort, or whether you could just shovel the whole manual text into the AI and let it figure things out. Modern language models are good enough that the lazy approach mostly works.

But mostly isn't always. A few concrete failure modes we hit before moving to typed fields:

Ambiguity: "The password is WELCOME2023" could be the Wi-Fi password, the door code, the smart-lock password, or the welcome gift code. Without a typed field, the AI has to guess from context — and sometimes guesses wrong.

Contradiction: A manual might say "check-in is 3 PM" in one paragraph and "check-in is at 14:00" in another (due to a format inconsistency that the host didn't notice). The AI picks one and runs with it. Typed fields force a single source of truth.

Missing info: If a manual doesn't mention parking, the AI might invent something — "there's usually parking on the street nearby" — because it's trying to be helpful. Typed fields with an explicit "no parking information" state let the AI say "I'll check with the host" instead of hallucinating.

We settled on roughly twelve typed fields covering 90%+ of what guests actually ask. The rest gets grouped under a free-text "additional notes" field that the AI can draw from for edge cases.

How the extraction actually works

Under the hood, the smart guide import is a two-stage pipeline.

Stage one is PDF text extraction. For typical house manuals — which are either exported from Word or scanned into PDFs — we use pdf-parse to pull the text content. It handles modern PDFs in milliseconds. For older scanned PDFs we fall back to OCR via Tesseract; it's slower (30-60 seconds) but it works.

Stage two is field extraction via Gemini. We pass the extracted text to Google's Gemini model with a prompt that asks for structured output. The prompt is something like: "Extract the following fields from this house manual text. If a field isn't present, return null. Respond in JSON." Gemini is good at this kind of extraction and gives us clean JSON back. We validate the shape, populate the typed fields in the apartment's guide record, and display the result for the host to review.

The host sees every field the AI extracted before anything is saved. They can correct what's wrong (rare — Gemini is accurate) and confirm. The confirmed fields become the source of truth for guest-message handling from that moment on.

Upload your house manual, skip the data entry.

Virtual Host AI reads your PDF and fills in the guide fields for you. Review once, done.

Try smart import

The review pass is the unlock

One thing we learned from early users: the value isn't just speed of data entry. It's that reviewing a structured extraction is faster than reading your own manual.

Most hosts wrote their manual once, a year ago, then added a few things. They don't remember what's in there. Looking at twelve typed fields — here's my check-in time, here's my Wi-Fi, here's my parking — they immediately spot the gaps. "Oh, I never wrote down that the dishwasher needs dishwasher salt specifically, not just any salt. Let me add that." Or: "I wrote the door code wrong — it's 4828 not 4827."

The structured view is the forcing function. It turns a monolithic document you never re-read into a checklist you actually verify.

What to put in the manual before uploading

For maximum extraction quality, your manual should contain these items in natural language:

That's about 90% of what guests actually ask. If your manual covers these nine topics, the AI will handle the overwhelming majority of guest questions correctly on day one.

Multiple apartments: the bulk-apply flow

If you have five apartments with similar or identical manuals (different addresses, same rules), we support a "bulk apply" flow. Upload once, edit the fields that differ per apartment, apply the common fields to all of them. Setup time for five apartments with identical structure: about 15 minutes.

If your apartments are genuinely different — a one-bedroom city flat and a countryside villa with a pool — you'll want distinct manuals per apartment. We recommend keeping them as separate PDFs, imported separately, so each guide stays tight to its apartment.

Why we're still making this better

Smart guide import is not done. We're working on: image upload (guests with photos of the manual instead of PDFs), automatic update (re-import every 30 days to pull in changes the host makes in the manual), and multi-language extraction (a single manual that has English and German sections gets split into both).

The north star is: you shouldn't have to type anything the AI could have read for you.

Upload your house manual. We do the rest.

15 minutes from Smoobu connect to AI answering its first guest message.

Get started
← Back to all posts