Five hotel or guesthouse tasks where a model removes routine
- Replies to reviews in several languages. A guest wrote in German about cold water in the shower. The reply has to be in German or English, specific, and free of promises the front desk will not keep.
- Pre-arrival emails. How to get there, is there parking, what time is check-in, can we bring a dog. The model answers from your reference sheet, not from memory.
- A draft booking confirmation. Dates, room, price, prepayment and cancellation terms in an email the administrator checks and sends.
- House rules. The Ukrainian version is the hotel's document; the model produces clear translations and a short memo for messengers.
- Occupancy review. From a bookings export the model builds a table by week and calculates how many guests must be added for a discount to pay off. The owner decides on the price.
These tasks make no sense outside accommodation: here a review is written after a night in a room, and the contract follows rules a café or a shop does not have. Where the line between model and person lies in each task is covered below.
Reviews: what the platforms allow
A model writes a reply in seconds, but the platform publishes it, and each has its own rules. We compiled them from Booking.com partner help, Booking.com content guidelines and Google Business Profile help.
✓ — allowed or described directly; ≈ — there is a condition or delay; ✗ — not allowed. We did not analyse Google's content policies separately, so the rows on contacts and data point to them rather than give a verdict.
The most common trap is language. A review in Polish, a reply in Ukrainian: on Booking.com it will not appear. So the model's instruction needs one rule: 'reply in the review's language; if unsure of the language, in English'. The second trap is the phrase 'write to us directly and we'll give you a discount'. On Booking.com contacts that advertise services are banned by the content guidelines, so such a reply will not pass.
Booking.com also advises writing original replies that are specific and true. The model gives speed; the specifics come from the administrator: the request to the model should include not only the review text but a short note from a person ('boiler replaced on 12 October'). Without it you get a polite generic reply that tells the next guest nothing. How replies to Google Maps reviews work for a food business, we covered in the article on AI for a café or restaurant.
Bookings and guest emails: who confirms what
The Rules for the Use of Hotels and Similar Accommodation (Order No. 19 of 16.03.2004, status 'in force') cover hotels with six or more rooms and similar accommodation offering limited hotel services. If your guesthouse lets rooms without daily cleaning, check whether it falls under this definition; the logic below is useful either way.
Three provisions directly set the line for a model:
- Clause 3.1. A booking contract is concluded only upon the hotel's written acceptance of the request and agreement on all essential terms. So an email saying 'your booking is confirmed' is a legal act. The model drafts it; the administrator checks it against the room chart and sends it.
- Clause 3.2. If a booking is not confirmed and the guest has not arrived by 18:00 on the arrival day, the reservation is cancelled. The model can remind a guest of the time, but takes the booking status from your system rather than inventing it.
- Clause 4.4. Without the guest's consent the hotel may not provide paid extra services not included in the contract. A model email saying 'we've added breakfast for UAH 300' without the guest's reply is a breach; the wording must be an offer.
House rules: a translation, not a new document
Clause 1.12 of the Rules requires internal house rules that do not contradict the general Rules, and clause 2.3 requires them in every room in Ukrainian and, depending on the category, in foreign languages at the hotel's discretion. Clause 4.13 allows refusing accommodation or terminating the contract if a guest repeatedly breaks the house rules and this causes damage or inconvenience to others.
Hence the split. The owner writes and approves the Ukrainian version: that is the text you will rely on if a guest has to be evicted. The model does three things: translations into English, Polish, German (by your guests' languages), a short one-screen memo for the pre-arrival email, and a check that the memo does not contradict the full text. Have a native speaker proofread the translation at least once: a mistake in the rule on quiet hours or pets later becomes an argument in a dispute.
Seasonal prices: how many guests a discount must add
A hotel sets its own prices: clauses 3.4 and 3.5 of the Rules allow free prices and a discount system. A model is useful here as a calculator you can explain the task to in words, but the decision rests on one formula, and it is worth knowing.
A hypothetical guesthouse: 8 rooms, UAH 2,000 a night, 50% occupancy in November. Variable costs per occupied room-night (cleaning, linen, water, electricity, breakfast) are a hypothetical UAH 400. How many guests must be added for a discount not to reduce income after variable costs?
The most common mistake is to calculate on revenue: a 20% discount means occupancy must grow 1 ÷ 0.8 = 1.25 times, from 50% to 62.5%. But each new guest brings you not the price but the price minus variable costs. The correct formula: required occupancy = (price − costs) × current occupancy ÷ (new price − costs).
| Discount | New price | Required occupancy on revenue | Required occupancy including costs | Growth you need |
|---|---|---|---|---|
| 10% | UAH 1,800 | 55.6% | 57.1% | +14.3% |
| 20% | UAH 1,600 | 62.5% | 66.7% | +33.3% |
| 30% | UAH 1,400 | 71.4% | 80.0% | +60.0% |
Calculation: (2,000 − 400) × 50% ÷ (new price − 400). Price, costs and occupancy are hypothetical; plug in your own.
A 30% discount requires the guesthouse to be 80% full in November. If even your neighbours never reach that in November in your region, the discount will only cut income. The model is useful in two places here: it quickly turns a bookings export into a weekly table and calculates this formula on your numbers. But it does not know your variable costs per room; those come from accounting or the owner. The same logic for an online shop, with orders instead of rooms, is in the article on a Black Friday discount.
What the model costs per month
Volumes are hypothetical: 60 review replies (700 input tokens: review, administrator's note, instruction; 250 output), 300 guest emails (900 and 200) and 4 weekly occupancy reviews (3,000 and 600). Prices from the OpenAI API pricing page on 25 October 2026 (standard processing, short context, per million tokens): gpt-6-luna — $0.10 input and $0.50 output, gpt-6.1-sol — $2 and $10, gpt-6-astra — $10 and $50. The National Bank rate on 25 October 2026 is UAH 45.051 per dollar.
| Model | One review reply | One guest email | One weekly review | Month |
|---|---|---|---|---|
| gpt-6-luna | UAH 0.009 | UAH 0.009 | UAH 0.027 | $0.07 (UAH 3) |
| gpt-6.1-sol | UAH 0.18 | UAH 0.17 | UAH 0.54 | $1.42 (UAH 64) |
| gpt-6-astra | UAH 0.88 | UAH 0.86 | UAH 2.70 | $7.11 (UAH 320) |
The gap between models is about a hundredfold, yet even the most expensive costs less than one night in a room. So choose by how the model writes Polish or German and whether it mixes up dates. Ukrainian tokenises less efficiently than English, so recalculate on your own texts. The total excludes the app that connects email, messenger and model: take its cost from the plans of the platform you choose.
Passports and guest data
Under clause 4.1 of the Rules a room is provided on presentation of a passport or another identity document; for foreigners, a national passport and a visa where one is required. The administrator needs these documents at check-in; the model needs them for none of the five tasks. A name, phone number and dates of stay together are personal data too: the Law 'On Personal Data Protection' defines it as any information about a person who can be specifically identified.
In practice: the request to the model contains the review or email text and your note. Your software inserts the guest's name into the finished text. Passport scans, document numbers and children's data never go into the model. How to choose a plan and service settings so data does not go into model training is in the article on data for an AI assistant.
- Will the text go to Booking.com or Google as a reply to a review?Language, contacts and guest data have rules. Yes → the model writes in the review's language; the administrator adds the fact and checks
- Does the email confirm a booking, a price or an extra service?It is a written acceptance or a change to the contract. Yes → the model drafts; a person sends after checking the room chart
- Is it the house rules?The Ukrainian version is the hotel's document. Yes → the owner writes the original; the model translates and shortens
- Do you need to decide what price to set?The model does not know your variable costs. Yes → the model runs the formula on your numbers; the owner decides
- Will a passport, document number or a child's data get into the request?No task needs them. Yes → remove them from the request
A first step for a week
- Pick the last 20 reviews from Booking.com and Google, including all negative ones and all those not in Ukrainian.
- Add one factual sentence from the administrator to each: what was done or what will change.
- Give two models the same instruction: the review's language or English, no contacts or discounts, no names of other guests, up to 80 words.
- Count how many replies the administrator would publish without edits. Take the model with more of them; for another week a person reads every reply before publication.
Our position
A model saves a hotel the most time on reviews and emails in foreign languages: that is where a person spends long minutes choosing words, and a wrong language on Booking.com means the reply is not published. Booking confirmations, prices and extra services stay with the administrator, because under the Hotel Use Rules these are contract terms. And before discounting a low season, calculate the required occupancy including per-room costs: it is always higher than a revenue-only calculation suggests.
The position does not hold if you work through a property management system that generates replies and confirmations itself: then check that system's settings rather than a model. Nor does it hold for a guesthouse with a few rooms where the owner knows every guest and gets a handful of reviews a month: the time saved will be less than the time spent setting up. The discount calculation assumes unchanged fixed costs and hypothetical variable costs of UAH 400 per room.
Checked on 25 October 2026 against Booking.com partner help and content guidelines, Google Business Profile help, the OpenAI API pricing page and the National Bank of Ukraine; the legal rules against the texts of the Rules for the Use of Hotels and Similar Accommodation (Order No. 19) and the Law 'On Personal Data Protection' on zakon.rada.gov.ua. The arithmetic was recalculated in code. Room price, costs, occupancy and text volumes are hypothetical.