AI for hotel HR: job posts, interview guides and onboarding packs for the season

AI for HR in a hotel means using a language model to draft documents: job posts, interview guides, onboarding packs, built from facts you supply. It does not mean letting software screen, rank or reject candidates. Drafting is fast when a person checks the wording; deciding stays with your managers, and CVs are personal data.

Every spring a seasonal hotel rewrites the same documents it wrote last spring, in three languages, for a team that will be gone by October. The drafting is repetitive and slow, and the temptation is to hand the whole cycle, CVs included, to a chat tool. This guide takes the other route. You will come away with a written split between what AI drafts and what people decide, a wording check for job posts, a structure for interview guides, and a rule for candidate data.

The seasonal hiring workload: what AI drafts and what it must not decide#

AI can draft the documents around a hire: the job post, the interview guide, the onboarding pack, the internal announcement. It should not decide who is shortlisted, rejected or hired.

The paperwork is repetitive and deadline-bound. Posts go up months before opening, in several languages, for roles that repeat every season with small changes. Drafting is a text task: given role facts, a model produces a usable first version in minutes. The value is in that first version, never in the final one, because the final version is written by your department head.

Deciding is a judgment task with legal consequence. Ranking CVs, filtering applications, auto-rejecting, scoring recorded video answers: keep all of it out. A model has no access to the candidate's context and cannot explain a rejection defensibly to the candidate, to a manager or to a lawyer.

The dividing line is easy to state and easy to check: if the output changes what happens to a person, a person makes it. Seasonal hotels have a specific pattern on top of this: high volume in a short window, and a list of returning staff. Whether a returning applicant is rehired is a human decision with a history behind it.

Write the split down once, on one page, so department heads and the HR team apply the same rule in February and in June. If you are also training team members on where AI fits their role more broadly, this page belongs in that in-house AI training program by role.

How do you write a hotel job post with AI without importing bias?#

Write the role facts first as a plain list, let the model turn them into prose, then run a wording check before anything is published.

Role facts before tone#

List the facts before drafting: department, reporting line, season start and end, shift pattern, languages and the level needed, physical demands such as standing or night shifts, systems used, and what the person is accountable for on a bad day.

Feed facts, not adjectives. "Hardworking, dynamic, young team" carries assumptions; "carries a 30-cover breakfast service alone between 07:00 and 10:00" is checkable in an interview or a trial shift. Every requirement you state should be testable in one of those two settings.

A language requirement needs a level and a job reason. For example: "Russian at check-in level, because a large share of summer guests arrive from that market" rather than "Russian speaker preferred". The reason ties the requirement to the job, which is what makes it defensible.

For multi-language posts, draft in one language, translate, then have a speaker of each language read the final text. A literal translation of a job title can describe a different job. Pay, benefits and contract terms are inserted by the hotel, never generated; a model will invent plausible figures if asked.

The wording check#

Before publishing, read the draft against these categories: age words (young, energetic, digital native), gendered job titles, nationality or native-speaker requirements, marital or family status, appearance, and anything that assumes a disability. Ask your legal adviser which wording counts as discriminatory in your own case; this guide gives categories to check, not legal rulings.

What do structured interview questions look like for a hotel role?#

A structured guide asks every candidate the same questions in the same order, with a written scale for what a weak, adequate and strong answer contains. AI can draft the questions and the scale; people run the interview and give the scores.

Structure beats intuition: same questions, same order, notes taken during the interview, scores written immediately after while the answers are fresh. Four question types cover most hotel roles: past behaviour ("tell me about a check-in you had to fix"), situational ("a guest arrives at 02:00 with no reservation"), technical ("walk me through closing the night audit"), and a practical trial task for housekeeping or kitchen.

The scale is the hard part. Define what a 1, a 3 and a 5 answer contains before the first interview, so two managers score the same answer the same way. AI's role here is narrow and useful: generate eight to ten candidate questions per role from the role facts, then the department head cuts to five and rewrites the scale in the hotel's own words.

Remove questions that invite protected information before use: plans to have children, age, where someone is originally from, health, religion. Ask in the language the job actually needs, and say in the post that the interview will be in that language, so no candidate is surprised at the door. Keep the written scores; they are the record that shows the decision rested on the job's requirements and not on an impression.

How do you build an onboarding pack that survives translation?#

Build the pack from the first week's actual tasks, one page per task, then translate it and have a speaker of each language read the final page before day one.

The pack is a set of short documents, not a handbook: a day-one checklist (uniform, locker, contract signing, system access), a week-one task list per department, a safety and emergency page, a guest-data rules page, and a who-to-ask-for-what page.

Draft from your own steps. A model given only "what a housekeeping attendant does in the first three days" produces something plausible and often wrong; give it the steps your supervisor actually teaches. The department-level content itself is a separate job; this guide covers the drafting and the checking, and a first-week training checklist by department covers the content.

Translation is where packs fail. A safety instruction that loses a step in translation is a hazard, not a style problem. Version control matters too: one owner, a date on every page, and old versions removed from the shared folder so nobody trains from last season's rules. Keep each page short enough to be read on a phone during a shift; a long document is not read at all in the first week.

Can you put CVs and candidate notes into an AI chat tool?#

No, not into a public tool. A CV is personal data about an identifiable person, and it belongs in an approved system with a defined retention period and a named owner.

A CV typically carries name, contact details, birth date, photo, nationality, work history and sometimes health or family information. Under KVKK (kişisel verilerin korunması kanunu, Türkiye's personal data protection law), all of that is personal data. In a public chat tool, what you paste may be stored or reviewed depending on settings you do not control. Treat anything pasted as having left your control.

The practical rule: draft with role facts and invented examples, never with a real CV, a candidate's name or interview notes. If a tool is used for candidate data at all, it must be an approved one, under a contract, with a stated purpose, access limited to the people hiring, and a retention period after which the data is deleted. Tell candidates what happens to their data and for how long, and have the wording checked by your legal adviser rather than copied from a template found online.

Deletion is a task with a date. Put it in the calendar at the end of the season, with a named person responsible. The same rule covers interview notes, reference checks and any spreadsheet of applicants kept on a personal drive. A wider hotel AI policy on what staff may share with chat tools should repeat this rule for every department.

Who makes the shortlist, the rejection and the offer?#

People do, using the written scores against the job's requirements, and they write down the reason at the time.

Shortlist by the scored interview against the stated requirements, not by a model's summary of a CV or a keyword match. Rejection is a short, factual message; the reason recorded internally is the one tied to a requirement, not to an impression. The offer, including pay, contract type and start date, comes from the hotel's own terms and is never generated.

Keep the post, the questions, the scores and the decision together in one file. That file is what makes the process explainable six months later. Consistency across departments matters in the same way: the same questions and the same scale for the same role, whoever runs the interview, so two candidates are never compared on two different conversations. If roles and decision rights are unclear in your property, an organisation chart of who decides what settles them before the season does.

AI still helps after the decision: the offer letter template, the internal announcement, the welcome message and the first-week schedule are all drafting tasks. Agree who signs off each step before the season starts, so a department head is not inventing the process under pressure in July.

What to do next#

Write the split on one page before the season starts: what AI drafts, who checks each document, where candidate data lives, and who decides. Then take the document you rewrite every year, the job post, and run the wording check on it before it goes up.