An AI register for hotels: every AI use, its owner and its risk tier on one sheet

An AI register is one sheet listing every place artificial intelligence touches a hotel's work: the use, the tool, the data it touches, who checks the output, one named owner, a risk tier and a review date. It is an internal control document, not a certificate, and it exists so the hotel can review, restrict or stop a use it can see. It is the simplest AI governance framework a hotel can build.

A guest review gets a reply drafted by a chat tool and posted by marketing. A sales manager drafts proposals with a helper on a personal account. A feature inside the channel manager starts suggesting rates. None of it was decided in a meeting, and none of it appears anywhere on paper. After reading this guide you will have the sheet itself: what columns it needs, how to find every use in the building, how to tier each one, and how to review it quarterly.

Why does a hotel need an AI register before it adds another tool?#

Because you cannot review, restrict or stop what you cannot see. A register turns scattered habits into a list the hotel can actually decide about.

AI arrives sideways. A marketing assistant drafts review replies. A sales helper writes proposals. The front desk pastes guest messages into a translation tool. A supplier switches on a feature inside software the hotel already pays for, and the only announcement is a line in a release note nobody read. None of these went through a decision process, but all of them put hotel data and hotel reputation somewhere the general manager cannot point to.

Without a list, every question becomes an argument about memory. Who approved this? What data went in? Who reads the output before it goes out? With a list, the same questions take a minute each, because the answer is a row with a name in it.

The register is also the attachment point for the other decisions the hotel needs to make: the staff policy on what may be shared with AI chat tools, the business account checklist when choosing a tool, and the test that runs before staff or guests touch a new assistant. Each of those documents needs something to attach to, and rows are that something.

It is a control document, not a certificate. It records what the hotel decided and who is answerable, and it makes no claim about any external standard. It also survives staff turnover: when the person who set up a tool leaves, the row still says what the tool does and who owns it now.

Keep it to one sheet. A register nobody reads is worse than no register, because it suggests control that does not exist.

How do you find every AI use in a hotel?#

Ask each department the same four questions, then read the software you already pay for. The list is almost always longer than the general manager expects and shorter than the staff fear.

Do the sweep by role, not by tool. Front office, housekeeping, reservations, sales and marketing, food and beverage, finance and the back office each answer separately, because each sees different habits. The receptionist who pastes a guest message into a translation tool at midnight will never volunteer that to a manager asking about "AI tools".

The four questions to ask each department#

Ask them in writing, so people can answer without an audience:

1. What do you paste into a tool?
2. What drafts text for you?
3. What does our software now do automatically that used to be manual?
4. What do you check before it goes out?

The fourth question matters as much as the first three. It tells you which uses already have a checker and which ones go straight from a machine to a guest or a record.

Where AI hides inside software you already pay for#

Read the release notes and the settings screens of the systems already in place: the property management system, the channel manager, the review platform, the customer database. Suppliers can add AI features quietly, so read the release notes and settings screens, and treat a feature that drafts, suggests or decides as an AI use even if nobody in the hotel chose it.

Check browser add-ons and extensions on work machines, and ask directly about personal accounts used for work. A personal account is still a hotel use if hotel data goes into it. List the uses you decide to ban as well as the ones you keep, with the date and the reason; a ban nobody wrote down gets undone by the next new hire.

Do not judge while collecting. The first pass is an inventory; tiering and approval come after the sheet is complete. For a picture of what safe uses look like in each department, see safe everyday AI uses, department by department.

What goes on the sheet? Seven columns#

Seven columns, one row per use: the use in plain words, the tool and account type, the data touched, who checks the output, the owner, the risk tier and the review date.

Use. One sentence a new receptionist would understand, for example "drafts replies to guest reviews". Not the product name, not the feature name. The product name tells the next reader nothing about what actually happens.

Tool and account type. The product, and whether it runs on a business account or a personal one. The account type decides whether the hotel has any control at all over what happens to the data.

Data touched. Guest names, contact details, identity or passport data, payment data, contracted rates, staff data, or none. Write "none" when it is true; it is the most useful word in the column.

Who checks. A role, not a team. "The marketing lead reads every reply before it is posted" is checkable; "the team reviews it" is not.

Owner. One named person who answers for the row at the review, not the person who happened to set the tool up.

Tier and review date. The tier from the next section, and the date the row is next looked at. Add a status column for approved, tolerated or banned, and keep retired rows with their date instead of deleting them.

Illustrative rows only, not a recommendation of any product.

What are the three risk tiers for a hotel?#

Tier by consequence, not by how clever the tool is. Ask what the output can do if it is wrong: stay inside the hotel, reach a guest, or move a booking, money or personal data.

Tier 1: internal drafts#

Brainstorming, summarising a document, drafting an internal memo, translating a staff notice. Nothing leaves the hotel and a person rewrites it anyway. The control is that no guest or payment data goes in.

Tier 2: words a guest reads#

Review replies, website copy, offer emails, FAQ answers, an assistant answering guest questions. The control is a named person reading every output before it is published or sent. A guest who reads a wrong sentence may not complain, but they remember.

Tier 3: bookings, money and identity#

Changing a rate or availability, confirming or cancelling a booking, issuing or sending an invoice, preparing a guest identity notification, refunds, anything touching identity documents. The control is a named person who confirms each action and a record of who confirmed what. If you want a fuller treatment of where the human line belongs, see which hotel AI actions need a person to approve them.

Two rules prevent most arguments: when in doubt, move the row up a tier; and a row moves down only after a review period with no incidents, recorded with its date. The tier is the hotel's own policy decision. It is not a standard, not a certification and not a claim of compliance with anything.

A useful side effect: the tier tells you which rows need a full test before use, and which can be tried by one person on a Monday morning.

How does the quarterly review work?#

One short meeting, the same sheet, three questions per row: did it change, did it go wrong, do we keep it. If the sheet is current, the meeting takes under an hour.

Attendees are the general manager, the KVKK contact person (the person responsible for personal data protection under Turkish data protection law), and the owner of each row. Owners answer for their rows; nobody else speaks for them.

"Did it change" means the tool, the account type, the data going in, or the supplier's own release notes since the last review. A feature that quietly widened its scope is a change even if nobody in the hotel touched a setting. "Did it go wrong" includes near misses: a wrong answer that reached a guest, a paste that should not have happened, an output nobody could explain, a message sent to the wrong person.

Record the decision per row: keep, tighten, retest or stop. Stopping is a normal outcome and should be written down like any other. Retire rows, do not delete them; a stopped use with a date and a reason is the most useful evidence the hotel has that the register is real. Close by setting the next review date on every row, and by adding any new use found since the last meeting. Rows in tier 2 and 3 that changed should get the pre-use test again; here is how to test a hotel AI assistant before staff or guests use it.

Who signs off on the register?#

The general manager owns the sheet, the KVKK contact person owns the data column, and each department head owns their rows. Three signatures, one document.

The general manager signs the tier assignments. Tiering is a management decision about risk appetite, not a technical one, and it should be signed by the person who answers to the owner. The KVKK contact person reviews every row that touches personal data and decides what the hotel needs before that use continues; confirm the specifics of your own case with a legal adviser. The register records the decision, it does not replace the advice.

Department heads are accountable for the checker named in their rows. If the checker leaves, the row is not compliant with the hotel's own policy until a new name is written in.

The register is evidence of a decision process, not a certificate. It supports no claim of certification, compliance or approval by anyone outside the hotel. And keep it where staff can read it: a register locked in a manager's folder cannot guide the person deciding whether to paste a guest message into a tool at 23:00.

What to do next#

Start the sheet this week with the rows you already know, put a name and a tier on each one, and book the first quarterly review before the sheet is finished. The staff policy, the business account checklist and the pre-use test all attach to those rows; without them, each of those documents floats free.

The same idea can be applied inside software. In AI Otel's PMS assistant, every action the assistant can take sits in one of three safety tiers: some run directly, some need a person's confirmation, and the most sensitive need two (built, June 2026). That is the vendor's own tiering for its own product; a hotel's register is the hotel's policy, and the two are separate documents. If you want to see how a tiered assistant behaves in practice, you can request a briefing.