How a Small Inn Owner Cleared Multilingual Booking Inquiries in 30 Minutes with AI

It was 10 p.m. I had just seen off the last guest, opened my inbox, and found three booking inquiries in English waiting for me. “We’d like to stay on December 31, two kids, can you do a vegetarian dinner?” Run it through a translation app, pull up the rate sheet, check availability, type a reply in English. Twenty minutes per email. By the time I looked up, the date had changed, and I still hadn’t prepped the next morning’s breakfast service.

That’s a story I heard from an innkeeper friend of mine, about one autumn night. A small hot-spring inn with eight guest rooms. The staff is three family members. Overseas guests had become a welcome part of the revenue, but every time the English, Chinese, and Korean inquiries piled up, the night got shorter. “I don’t want to turn people away. But my body can’t keep up.” That’s what she told me.

What I did was not build some magic auto-reply system. I just changed the shape of the work: let the AI draft the reply first, and the innkeeper only checks the content and sends it. Twenty minutes per email dropped to under five, checking included. Today I’ll walk through how it works, using as little jargon as I can.

Key takeaways

• For an inn handling multilingual inquiries, “writing every reply from scratch” is the worst part. Let the AI build a draft and have a human focus on checking it, and the whole thing gets dramatically lighter.
• Hand off “translation,” “first-draft wording,” and “organizing common questions.” The final call on rates, availability, and allergy handling always stays with a human.
• I’ve included a copy-paste prompt template and a check script that gathers scattered inquiries into one tidy list.
• Strip personal data (guest names, contact details) before anything reaches the AI. This is the one rule I never compromise on.
• Even a tiny inn frees up 30 minutes to an hour of admin time a day. Saving 20 hours a month is not unusual.

Where front-desk work at an inn actually jams up

Let me be clear about who this is for. I’m picturing a small-to-mid inn of maybe 20 rooms or fewer, where the owner or front-desk person juggles reservations, inquiry replies, and guest info all at once. There’s no dedicated web person, no IT person. It’s OTAs (online travel agencies like Booking.com or Expedia), plus email, phone, and the occasional social-media DM. Inquiries coming in through scattered channels is its own classic problem at this size.

Lay out the workflow and it looks like this:

1. An inquiry arrives (email, OTA, phone, DM).
2. Read it. If it’s a foreign language, translate it.
3. Check availability and rates in the reservation ledger and channel manager.
4. Check allergies or special needs (pickup, accessibility).
5. Write the reply. If it’s a foreign language, translate it and send.
6. Once the booking is confirmed, log it in the ledger and fold it into that day’s prep.

Steps 2 and 5 - the reading, writing, and translating - are where the time gets swallowed. Foreign-language replies especially: the harder you try to write them warmly, the longer they take. And when you’re slow, the guest books somewhere else. “Reply speed is directly tied to your booking rate” is the scary part of inbound tourism.

Common rework and headaches

• The translation app’s literal output is stiff, and the warmth that makes a hot-spring inn feel like one disappears.
• You answered a similar inquiry before, yet you’re rewriting the wording from scratch every time.
• You meant to write clearly in English, but the way you phrased the rates (tax, bathing tax, child pricing) was vague, so a follow-up question comes back.
• Things pile up late at night or in peak season, replies slip to the next day, and you lose the booking.
• Reply quality varies by who’s on shift, and a new hire takes even longer.

What to hand to the AI, and what a human must decide

This is the most important part, so I’ll put it in a table. Try to hand off everything and you’ll have an accident. Draw the line first.

Task	Hand to AI	Human must decide
Reading and translating foreign languages	Yes - build the draft	Final check on subtle nuance
Drafting the reply	Yes - mimic past good replies	Pressing send
Answering rates and availability	No - never let it write these	Yes - human enters the figures checked in the ledger
Allergy and medical handling	Partly - organize the question only	Yes - the promise of “we can do it” is a human call
Turning common questions into an FAQ	Yes - collect and organize	A human approves the content for accuracy

The way to remember it is simple. Anything that, if wrong, hurts a guest; anything that involves money; anything that becomes a promise - a human holds. Everything upstream of that, the prep work, you hand to the AI. That’s it.

Never tell the AI “just answer the rates and availability nicely.” The AI hasn’t seen your actual ledger, so it will write plausible-sounding fake numbers. For an inn, that’s fatal. So in the draft, always make it leave the availability and rate parts blank or marked as placeholders, and run it so a human fills in the figures they’ve verified.

Three concrete use cases

Use case 1: Drafting multilingual inquiry replies

This is the one that pays off most. Paste in the foreign-language inquiry you received, and have it produce both a plain-language summary and a reply draft in the guest’s language, all at once. The trick is to make it leave a marker for rates and availability saying “a human fills this in here.”

Before: 20 minutes for a single English reply. Stacks up late at night. After: The summary and draft appear in 30 seconds. The owner verifies the figures, polishes the wording, and sends in under five minutes.

Use case 2: Use past good replies as “models” to keep the voice consistent

Every inn has its own “way we say things.” Reproducing that from scratch every time is hard. So hand the AI three to five replies that got good feedback in the past as models. Then even when a new hire handles it, a draft comes out warm, the way the owner would have written it. The “guest-service voice” that used to live in one person’s head becomes something the whole team can share.

Use case 3: Organizing an FAQ and translating it into multiple languages

“Is there Wi-Fi?” “Do you pick up from the nearest station?” “Do you have child-size robes?” The same questions come in over and over. Have the AI organize these into an FAQ table in English, Chinese, Korean, and your own language. From then on, you just copy the relevant part. Use the checklist below to spot holes in your own inn’s FAQ.

• Check-in and check-out times, and whether early arrival or late stay is allowed
• Rate breakdown (room rate, bathing tax, child pricing, tax-inclusive labeling)
• Meals (whether allergy, vegetarian, and halal handling is possible)
• Pickup and access from the nearest station
• Cancellation policy (how many days out charges begin)
• Facilities (Wi-Fi, private bath, accessibility, parking)
• Payment methods (on-site, prepaid, accepted credit cards)

A copy-paste prompt template

This is the prompt for building an inquiry-reply draft. Rewrite the [ ] parts with your own inn’s info before using it. The key point: don’t paste guest personal data - leave only the substance of the request.

You are the front-desk person at a long-established hot-spring inn.
For the overseas inquiry below, produce two things:

1. A plain-language summary (bullet points: what is being asked)
2. A reply draft in the guest's language

Rules:
- Rates, availability, and pickup are NOT yet confirmed.
  Always leave the marker "[TO VERIFY: human enters the figure]".
  Do not invent numbers.
- Keep the tone polite but not stiff - the warm tone of a hot-spring inn.
- At the end, add a natural sentence in the guest's language that
  conveys "please feel free to reach out with any questions."

Past good replies to use as models:
[Paste 1-3 past replies that received good feedback here]

The guest's inquiry text:
[Paste the inquiry here. Remove the name and email address first]

The heart of this prompt is that it explicitly says “do not write rates on your own, leave a marker.” If you want to sharpen your prompt design further, read Advanced prompt engineering for Claude Code. If Claude Code itself is new to you, starting from Getting started with Claude Code is the fastest path.

Check script: gather piled-up inquiries into one list

If you save the inquiries that arrived scattered across email and DMs into a single text file, you can mechanically pull out the gist and turn it into a list. It’s a small script that runs anywhere Node.js is installed. It also does a light “strip personal data” pre-pass before anything goes to the AI.

import { readFile, writeFile } from "node:fs/promises";

// inquiries.txt holds inquiries separated by "---"
const raw = await readFile("./inquiries.txt", "utf8");
const blocks = raw.split("---").map((b) => b.trim()).filter(Boolean);

// Mask anything that looks like an email address (minimal PII pre-pass)
const maskEmail = (t) => t.replace(/[\w.+-]+@[\w.-]+\.\w+/g, "[email redacted]");

// Roughly classify the request by keyword
const tagOf = (t) => {
  if (/vegetarian|halal|allergy/i.test(t)) return "Meals";
  if (/cancel|refund/i.test(t)) return "Cancellation";
  if (/pickup|station|access/i.test(t)) return "Pickup / Access";
  if (/price|rate|cost/i.test(t)) return "Rates";
  return "Other";
};

const rows = blocks.map((b, i) => {
  const clean = maskEmail(b);
  const head = clean.replace(/\s+/g, " ").slice(0, 40);
  return `| ${i + 1} | ${tagOf(clean)} | ${head}... |`;
});

const table = ["| No | Request | Opening |", "| --- | --- | --- |", ...rows].join("\n");
await writeFile("./inquiries-summary.md", table, "utf8");
console.log(`Classified ${blocks.length} inquiries into inquiries-summary.md`);

You run it with just something like node summarize.mjs. Now you can see at a glance: “Today there are three meal questions and two pickup questions.” Because you can prioritize your replies, you drop fewer of them during peak season.

Before, after, and a rough ROI

This is back-of-the-envelope, but here’s a target. Say you get five inbound inquiries a day, and each reply takes 15 minutes.

• Before: 5 inquiries x 15 min = 75 min/day
• After (draft + check): 5 inquiries x 5 min = 25 min/day
• Saved: about 50 min/day. At 25 working days a month, roughly 20 hours a month

At an hourly value of $20, that’s about $400 worth of time freed up each month. Bigger than the money is getting the owner’s nights back. “I can finally handle the next morning’s prep” is the part that lands hardest on the ground.

If you want to roll this out across multiple staff as a business or facility, and set up the operating rules and review process at the same time, asking for help beats figuring it out alone. For training and rollout consulting, see the training and consulting page. If you just want to try it solo first, start from the learning materials page, which has a free PDF and paid materials.

Personal data and security notes

This part is non-negotiable for an inn. Guest names, email addresses, phone numbers, and credit card details should be stripped before anything reaches the AI, as a rule. Leave only the “request” itself and you can still build a perfectly good reply draft. Masking email addresses in the script above is the first step of that.

Writing your internal rules down keeps the workflow steady even when a new hire arrives. With Claude Code, there’s a way to write your inn’s rules into a file and have it read them in. For how to do that, CLAUDE.md best practices is a useful reference. At a minimum, settle these three:

• Don’t paste personal data (names, contact details, card numbers, booking numbers).
• A human always reads the AI-written reply before it goes out (no auto-send).
• Only real figures for rates, availability, and what you can accommodate, verified in the ledger, go in.

It helps to know the official public guidance too. In the US, the FTC’s guidance on protecting personal information lays out the basics of handling customer data. Lodging businesses keep guest records, so it’s worth a read at least once.

FAQ

Q. Can I use this even if I’m bad with computers? A. For the initial setup only, it’s safer to borrow a hand from someone who knows their way around. Once setup is done, you just ask in plain language. You don’t need to write code. A getting-started path for non-engineers is laid out in Claude Code for non-engineers.

Q. Are Chinese and Korean fine too? A. For the major languages, you get a perfectly usable draft. That said, ideally have someone who understands the language do the final check. If that’s hard, replying to guests with English alongside reduces misunderstandings.

Q. Won’t the replies come out sounding robotic? A. Whether you hand it past good replies as “models” changes this a lot. Without models, it tends toward a stiff literal translation. Just handing it three replies makes it noticeably more like your own inn.

Q. Wouldn’t it be easier to let it answer the rates automatically too? A. Don’t. The AI hasn’t seen your actual availability or rates, so it will write plausible lies. Rates and availability alone, a human looks at the ledger and enters them. That part is not up for negotiation.

Q. How long until I see results? A. Drafting replies works from day one. For the FAQ, let it accumulate for one to two weeks and then organize it, so it fits your own inn. I’d start with one language and one type of request.

What happened when I actually tried it

I tried this for two weeks at the inn run by the owner from the opening. I wanted to confirm three things: whether drafts really cut the time, whether the wording came out sounding like the inn, and whether any rate accidents happened.

Time per English reply dropped from an average of 18 minutes to an average of 6. After I handed it three model replies, the owner herself laughed and said “it’s like I wrote it,” so voice reproduction passed. And the rate accident I worried about most? Zero, thanks to writing “don’t invent numbers, leave a marker” into the prompt. The AI leaves the blank, the owner fills in the ledger figure. That one extra step is what makes it work.

The unexpected win was that, because the evening emails stopped piling up, the owner could give herself room for the next morning’s breakfast prep. What got cut wasn’t just time, it was the late-night anxiety. Don’t chase perfect automation; settle deliberately for drafts plus a check. The smaller the inn, the more that trade-off pays off - that’s my honest takeaway from this one.