Short-term rental hosts in tourist markets deal with guests from everywhere. A week's worth of messages for a Vienna apartment might include German, English, Italian, French, Spanish, Polish, Portuguese, Hungarian, and the occasional Japanese. For years, the standard response was either to reply in English and hope for the best, or to maintain translated templates for the top 4-5 languages.
Both approaches are bad. English-only loses guest trust — nothing says "I don't really care about you" like replying to a German question in English. Translated templates cover the common cases but fall apart the moment a guest asks a specific question the template doesn't match.
The alternative, and the thing that makes multi-language guest support actually work today, is native multilingual AI. No templates. No language packs. No dropdown in the host settings. The model detects the guest's language from their first message and replies in it.
Why templates fail
Translated response templates have three failure modes that compound over time.
Mismatch between question and template. You have a template for "Wi-Fi password" in German. A guest writes "Hallo, welches Passwort ist für das Internet?" Your template keyword matches on "Passwort" — maybe. But if the template only triggers on the word "Wi-Fi," you've just missed it. Multiply by 40 templates, 40 languages, and countless phrasing variations, and the hit rate crashes.
Context lost in translation. Your German template says "Check-in is from 15:00." The guest arrived this morning at 10:30 and asks when they can get in. A template has no way to know "15:00" isn't the right answer anymore — the answer is "in 4 hours" or "at 15:00 today." Templates don't do context.
Tone mismatch. Templates are written by someone translating once. They're stiff. They don't match how you'd actually write. Guests feel it — the reply lands as "bot" even if the language is correct. Review scores drift downward because the hospitality cue is off.
How native multilingual AI changes the problem
Google's Gemini (and similar modern LLMs) are trained on enormous multilingual corpora. Unlike translation systems, they don't work by translating English to another language. They understand each language natively, including idiom, tone, and context. When a guest writes in Italian, the model reads Italian, understands what they're asking, looks up the answer in your apartment guide, and responds in Italian — with whatever tone feels natural.
We don't configure language. We don't translate your apartment guide into 40 versions. We just use the multilingual model end-to-end. The guest's first message tells us which language to respond in, and we stay in it for the rest of the stay.
The practical result: a host with an apartment in Vienna gets replies sent in whatever language the guest opened the conversation in. Italian guest arrives and asks about check-in in Italian → Italian reply. German guest asks about parking in German → German reply. English backpacker asks about the nearest supermarket in English → English reply. Same apartment, same guide, same host. Zero configuration.
What languages actually work well
In production, Gemini delivers high-quality replies in these languages for typical short-term rental conversations:
- Tier 1 (excellent): English, German, French, Italian, Spanish, Portuguese, Dutch, Polish
- Tier 2 (very good): Czech, Hungarian, Romanian, Swedish, Danish, Norwegian, Finnish, Greek, Turkish
- Tier 3 (good for basics): Russian, Japanese, Korean, Mandarin, Arabic, Hebrew, Ukrainian
We've tested responses in all of these. Quality varies slightly by language — Western European languages with large training corpora tend to produce the most natural output. Asian languages are good for practical answers but occasionally produce slightly formal phrasing that a native host might soften. For the short-form guest-message use case, all 40+ languages we've tested are acceptable or better.
Run your short-term rental in every guest's language.
Smoobu autopilot with native multilingual AI. From €22.90 per apartment per month.
When to override the AI's language choice
There's one edge case where automatic language detection can surprise you. A guest from Brazil is staying in Vienna and writes their first message in English — maybe because they weren't sure if the host would speak Portuguese. The AI, seeing English, replies in English. Mid-stay, the guest realizes you're flexible and switches to Portuguese. The AI should then switch too.
Our implementation tracks language per-message, not per-thread. If the guest writes in Portuguese, we reply in Portuguese, regardless of what came before. Hosts don't need to intervene. The thread just fluidly moves between languages as the guest does.
The alternative — forcing a language per thread — is what earlier chatbot generations did. It creates awkward moments when the guest's language preference shifts mid-stay.
What this means for marketing your apartment
Most hosts we talk to hadn't considered this, but it's the biggest downstream effect: if you know your AI co-host handles every European language natively, you can confidently market your apartment in all of them. The multilingual Airbnb chatbot isn't just a support convenience — it's a distribution channel. You can list in Polish Airbnb, Italian Booking.com, French direct-booking channels, with the same operational stack. No translation agency, no language skills needed.
Hosts in our early group who'd previously listed only in English and German added four more languages in their first month with us. Direct bookings from those markets were measurable within the first 90 days.
The role of your apartment guide
One important gotcha: the AI's multilingual capability reads the apartment guide's content in its original language and translates at reply time. If your guide says "Check-in is at 3 PM," that works perfectly whether the guest asks in English or in Japanese. The AI reads the underlying fact and renders it in the guest's language.
But if your guide is a mixed document — some fields in German, some in English, some in broken Italian — the AI might get tripped up on edge cases. Keep your guide in ONE language (your primary language works fine) and let the AI handle the translation at query time. Our smart guide import automatically normalizes fields so inconsistencies don't leak into guest replies.
Privacy and data residency
Gemini inference runs in Google's infrastructure. For EU-based hosts, we route through Google's EU endpoints — guest messages don't leave the EU for processing. For details on how we handle data, see our security page.
Getting started
Nothing to configure. When you enable the Smoobu autopilot on an apartment, multi-language support is already on. The first time a non-English guest sends a message, you'll see the reply go out in their language and know it's working. Most hosts discover it accidentally by scrolling through their Smoobu message log and seeing German, Italian, and French replies all flowing smoothly alongside each other — no template library, no configuration, no drama.
For hosts in multi-language tourist markets, it's typically the second-most-mentioned feature in reviews, behind only response speed.
Reply to every guest in their language.
Zero configuration. 40+ languages on day one.