Writing prompts for hotel work: a simple method with examples

A prompt is the written brief you give an AI assistant: who it is writing as, what the task is, which facts it may use, what format the answer takes and what it must not do. Hotel staff get better drafts by putting the facts in the prompt and checking the result.

A receptionist asks an AI chat tool to write a polite reply to a late check-in request, and gets a warm paragraph that promises a held room, a late check-out and a rate the hotel never agreed to. The tool was not wrong. The brief was empty. This guide to AI prompts for hotels gives your team one method, five short blocks, and shows it on three everyday tasks: a reply to a late-arrival message, a German pool notice and a shift log summary. Each example shows a weak prompt, a better one, and what a person still checks.

Why does the same question get a better answer when it carries the facts?#

Because the assistant has no access to your hotel: it answers from what the prompt contains, and where the prompt is silent it fills the gap with something plausible.

A language model produces text by continuing a pattern. It holds no record of your rates, your rooms, your opening hours or your policy. An unstated fact becomes a guessed fact. The usual failure is not a rude answer but a confident, wrong one: a promised late check-out, an invented fee, a policy the hotel does not have.

Compare two briefs. "Write a reply about late check-in" gives a generic paragraph. A brief that states the arrival time, the last check-in hour, whether the room is held and what the guest must do gives a reply you can almost send. The difference is not cleverness; it is the facts.

Prompts are not magic words. The same five parts work in Turkish and in English, and no special phrase has to be typed to make the assistant "switch on". The practical test of any prompt: could a new colleague answer correctly from it without asking you a question? If not, the prompt is missing facts, not politeness.

Once a prompt works, keep it as a template. The value sits in the facts and the limits, not in the wording. For more on where AI tools fit into daily hotel work, see our guide to safe everyday uses of AI tools, department by department.

What are the five parts of a hotel prompt?#

Role, task, facts, format and limits: five short blocks that turn a vague request into a brief a colleague could follow.

Role sets who the assistant writes as: front office, reservations or sales. It shapes tone and vocabulary; it does not give the assistant knowledge it does not have. Task is one verb and one output: draft a reply, translate a notice, summarize a log. Three jobs in one prompt produce three half-finished answers.

Facts are only what the assistant may use, pasted in as a short list, with placeholders standing where anything personal would go. Format fixes length, language, structure and layout: how many words, bullet points or prose, whether the line breaks of the original notice must be kept.

Limits say what the assistant must not do: no promise of availability, no price it was not given, no new facts, no opinion on a legal duty. Order matters less than completeness, but the limits block is the one people drop first, and it is the one that prevents a promise the hotel cannot keep.

How do you write a prompt for a late check-in request?#

State the arrival time, the last check-in hour, whether the room is held and what the guest must do, then ask for a short reply in the guest's language.

The weak version, "write a reply to a guest asking for a late check-in", produces a warm paragraph that may promise a desk staffed all night. The better version names the role, the task with a word limit, the facts as a list, the format, and the limits: promise nothing outside the facts, mention no rate.

Keep placeholders in the template. [Guest name], [arrival time] and [reservation reference] are filled in the hotel's own system, not typed into a chat window. For example, a guest writes that the flight lands at 23:10. The facts block says the desk closes at 01:00, the room is held, and no charge applies. The reply then confirms three things and promises nothing else.

What a person checks: that the last check-in hour is right for that date, that the room really is held, that no fee or rate appeared from nowhere, and that the name is spelled as in the reservation. The same template covers early check-in, late check-out and airport transfer timing; only the facts block changes.

When the message turns into a complaint or a change of booking, it stops being a drafting task and goes to a person. Our article on drafting reservation emails without wrong promises covers that line in more detail.

How do you use a prompt to translate a pool notice?#

Give the assistant the audience, the tone, the fixed terms and the layout, then have a person read the result before it goes on the wall.

"Translate this into German" returns a correct but flat text that may not fit the sign and may quietly soften a safety rule. The better prompt names the audience (hotel guests, including families), the tone (plain and formal), the fixed terms (pool, depth, opening hours, children under supervision), the layout (same line breaks, same order, no longer than the original) and the limits (add no rule, remove no warning).

Ask for uncertain terms to be listed separately at the end, so a person decides them instead of the assistant choosing silently. What a person checks, in this order: safety wording, then numbers such as depth and hours, then the names of facilities, then whether the text still fits the sign.

A native speaker or a professional reviewer signs off. A machine translation of a safety notice is a draft, never a finished sign. Keep a short glossary of the hotel's fixed terms so that every notice, in every language, uses the same words for the same thing. If your property serves several language groups, a language plan for your guest mix tells you which languages earn that effort.

How do you summarize a shift log for the next team?#

Remove names, room numbers and anything personal first, then ask for a fixed structure: what happened, what is still open, who owns it.

The shift log is the most sensitive text anyone will paste into a prompt. It carries guest names, room numbers, complaints and sometimes health details. Anonymize before pasting: "a guest in a sea-view room" instead of a name and number, no booking references, no card data, no diagnosis, no staff surnames.

The better prompt states the task (summarize for the incoming shift), the facts (the anonymized log and nothing else), the format (one line per item, grouped as open issues, follow-ups and closed items, original times kept) and the limits (add nothing that is not in the log, mark anything unclear as "check").

What a person checks: that nothing was dropped, that each owner is right, that a complaint is not softened into a "feedback note", and that no name came back into the summary. The summary is a handover aid. The original log stays the record, and the summary is filed beside it, not instead of it.

If the log cannot be anonymized without losing its meaning, do not paste it. Write the summary by hand and keep the prompt for the next, cleaner task. For the wider rules, see what staff may share with AI chat tools.

Where should the team keep its prompt templates?#

In one shared document the team can find, with a name, an owner and a review date for each template.

Name templates by the task, not by the tool: late check-in reply, pool notice translation, shift log summary. Every template carries its limits block; that is the part people delete first when they copy a prompt into a chat window. Give each template one owner and a review date. A template that promises something the hotel stopped offering is worse than no template at all.

Placeholders stay in the template. The filled-in version lives in the hotel's own systems, never in the shared document. New staff learn the method from the templates, not from a training session they forget by the second week.

Version the document lightly: a date and one line on what changed is enough for a team of ten or a team of a hundred.

What must a person check before anything is sent?#

Facts, promises and personal data: the three things an assistant cannot verify for you.

Facts: rates, hours, availability, names and dates. The assistant repeats what it was given and invents what it was not. Promises: anything that commits the hotel, such as a held room, a refund, a late check-out or a discount, needs a person who is allowed to authorize it. Personal data: confirm that no guest identity, card detail or health information has traveled into the draft or into the tool's history.

Language and tone matter too. Read it as the guest will read it, in the guest's language, and check that a complaint is answered as a complaint. The rule of thumb: if you would not send it after reading it twice, do not send it after one read because a machine wrote it. Log the sent message where the hotel keeps its guest correspondence, so the next person sees what was promised and by whom. When a message needs judgment rather than drafting, hand it over to staff.

What to do next#

Pick one task your team repeats this week, write the five-part brief for it, run it once, correct what came back, and save the corrected version as the team's first template. The method holds together because each part does one job: the facts stop invention, the format stops rewriting, the limits stop promises, and the person at the end stops everything else.