If you host short-term rentals on Airbnb, Booking.com, or through Smoobu, you already know the rhythm: the same half-dozen questions, over and over, from every guest. "What's the Wi-Fi password?" "How do I get into the apartment?" "Is there parking?" "Can we check in early?" Each one takes you maybe 60 seconds to answer. But you're not answering them in peace — you're answering at 11:47 PM, during dinner, while driving, in line at the supermarket. They interrupt your life all day, every day, and the emotional tax is bigger than the time one.
This is the problem that AI message autopilot solves. Not by replacing you. By handling the 80% of guest messages that are completely predictable, freeing you to handle the 20% that genuinely need your judgment.
The surprising shape of guest messages
We analyzed a week of anonymized guest messages across our early user base. The distribution was striking: across more than 4,000 messages, 81% fell into five predictable categories. Check-in logistics (how do I find the key box, what's the code, can I check in early) made up 27%. In-unit utilities (Wi-Fi, heat, hot water, dishwasher) accounted for 21%. Local information (best restaurant nearby, nearest metro station, airport transfers) was 16%. Checkout logistics (where do I leave the keys, do I need to strip the beds, what time do I have to be out) made up 11%. And generic acknowledgments and pleasantries (thanks, we arrived safely, see you next time) came in at 6%.
That leaves about 19% of messages that need real thought. Complaints. Special requests. Anything ambiguous or emotional. In a typical week of 50 guest messages, that's 10 messages that need you, and 40 that don't.
AI message autopilot handles those 40. You handle the 10.
Why templates fail and AI wins
The traditional solution to repeat guest questions has been canned responses or saved templates. Every major PMS platform offers them. Every host who's used them knows they don't really work.
Templates fail for three reasons. First, guests don't ask questions the way you expect. "Is there WiFi?" and "how do I connect to internet" are the same question but miss a keyword match. Second, answers are always contextual — the guest arriving today needs different information from the guest arriving in three days. A template can't know that. Third, guests write in their native language; your templates are in yours.
A properly-tuned AI — we use Google's Gemini model, which has strong multilingual capability — reads the guest's actual message, looks up the answer in your apartment's guide, and replies in the guest's language with the specific information relevant to their booking. That last part is the real unlock: an answer that says "your check-in is at 3 PM tomorrow, Thursday the 24th — here are the key-box instructions" is qualitatively different from a generic "check-in is 3-8 PM".
The intervention shield: why this isn't a chatbot
Here's where most automation stories go wrong. They hand over the entire conversation to the AI, and suddenly the host is out of the loop. Guest has a problem, AI gives a polite but useless non-answer, guest leaves a 3-star review, host finds out three days later.
Our approach is different, and it's the thing we care most about. The AI knows when to step back. If you reply to a guest — even once, from any device — we flag that conversation as "host intervention active" and the AI retreats for an hour. If the guest sends anything emotional, urgent, or outside the guide's knowledge, the AI escalates to you via Telegram with a full context card: guest name, reservation, message, suggested response, and four one-tap buttons (Pause AI, Mark Solved, Reply, View Thread).
The result: you're always in the loop when it matters, and never when it doesn't. It's a co-pilot, not an auto-pilot.
Let the AI answer the next Wi-Fi question for you.
Message autopilot answers guest questions in seconds, in their language. You stay in the loop, never in the thread.
What 10 hours a week actually looks like
The math is simple. If you have five apartments and each gets an average of 10 guest messages per week, that's 50 messages. Assume each one takes 90 seconds to read, think about, and respond to — that's 75 minutes of active time. Plus the context-switching tax, which researchers estimate at 2-3 minutes per interruption. For 50 interruptions, that's an additional 100-150 minutes of lost focus. Total: around 3 hours a week of real time, 2 hours of context-switching overhead.
Now scale that to 10 apartments: you're at 10 hours per week of attention drain. And that's not counting the invisible cost — the way you never fully relax knowing the next ping could come at any moment.
With AI autopilot, 80% of that goes away. The 10-apartment host still gets pinged for the 10 messages that matter. But those 10 messages are flagged as important in Telegram, with context, and you can handle them when it works for you — not the moment they arrive.
The honest downsides
No automation is perfect. Three honest caveats.
First, onboarding matters. AI quality is a direct function of how good your apartment guide is. If your "guide" is three sentences about check-in, the AI has nothing to work with. We've built a PDF import tool that extracts typed fields automatically from your existing house manual, so onboarding usually takes 10 minutes per apartment, not hours.
Second, the AI will occasionally escalate too cautiously. A guest asks "is the place quiet at night?" and the AI escalates because it can't answer definitively from the guide. That's by design — we prefer false-positive escalations to the AI inventing answers. But you'll sometimes get a Telegram ping for something you'd have answered in a word.
Third, it's not zero-attention. We call it a co-pilot for a reason. You still read escalation cards. You still verify upsell approvals. The 80/20 split is a dramatic improvement, but it's not "set and forget."
Getting started
If you're already on Smoobu, connecting Virtual Host AI takes one click. Paste your API key, we import your apartments, guides, reservations, and active message threads. The AI doesn't reply to anything until you turn on Autopilot per apartment — so there's no surprise first message sent on your behalf. You can watch it for a day, see what it suggests, then enable it when you're comfortable.
Ten hours a week doesn't sound like much until you get it back. Try it for a month and tell us how your Sunday mornings change.
Ready to stop answering the Wi-Fi question?
Connect Smoobu in one click. First guest reply in under 10 minutes.