Hotel SOPs Written With AI: Check-In, Housekeeping, Night Audit

To write hotel SOPs with AI, let it interview whoever does each job best: a receptionist for check-in, the head housekeeper for room turnover (adding photographs), and the night auditor for the audit (adding your system's end-of-day help text). Turn the answers into numbered steps, test each draft on a newcomer, then sign it off with a version number.

Templates from hotel-industry sites describe somebody else's property: a different booking system, different key cards, a different payment terminal. What makes an SOP worth having is the detail only your team knows, such as which report the night auditor opens first or what happens when a card pre-authorisation fails, and a chat assistant cannot know that until someone tells it. The same goes for anything the law or your insurer dictates, such as fire evacuation, allergens at breakfast or guest registration. Copy those from the official source and have your adviser approve them.

Follow me on Instagram@sagnikteaches

Choose one owner per procedure before opening a chat window

A standard operating procedure (SOP) is the written version of a job that repeats, so that two people do it the same way and a new starter can follow it without asking. In a hotel the three that cost the most when they go wrong are check-in, room turnover and the night audit. Each has one person who does it better than anyone else, and that person is your source. The AI is the interviewer and the typist.

Connect on LinkedInSagnik Bhattacharya

Pick the owner for each: the receptionist whose guests rarely complain for check-in, the head housekeeper (or the attendant the head housekeeper would copy) for room turnover, and the night auditor who has run the audit longest. Book 45 minutes with each, away from the desk, and tell them the SOP is a record of how they work, not a performance review. Staff who feel audited describe the official version of the job, and the real version is the one you need.

Subscribe on YouTube@sagnikteaches

Fix the layout before the first interview, so that all three SOPs match and the AI doesn't invent its own. This skeleton suits a 28-room hotel and a 90-room one:

HEADER     SOP code, title, owner, version, date approved, next review date
PURPOSE    One sentence: what this job protects (first impression, a clean balance, a safe room)
WHEN       What triggers the job, and who does it
NEEDED     Systems, keys, forms, supplies
STEPS      Numbered, one action each, the screen or object named exactly as it appears
IF WRONG   Table: situation / what to do / who decides
CALL       Names and numbers for escalation, by shift
RECORDS    What to log, where, and for how long
SIGN-OFF   Owner, tester, manager, date

Guest data, codes and passwords stay out of the chat

An interview about a hotel job drifts towards real guests, and real guests don't belong in a chat tool. Before anyone starts, agree four exclusions: guest names and contact details; card, passport or ID numbers; safe, door and alarm codes and every password; and staff personal data. Say "the guest", "room 12" or "the corporate account" instead. If the auditor has to describe a screen, describe the field, not what is typed into it.

Two settings matter. On personal ChatGPT and Claude accounts, find the model-training switch in privacy settings and turn it off for this work; business plans don't train on your content by default. And keep the house style and your drafts in a Project (ChatGPT and Claude both have them, and Claude's free plan allows up to five), so every new chat starts from the same instructions rather than whatever you typed last week.

An interview is also a recording of someone's words. If you voice-record it, say so at the start and delete the audio once the text has been checked.

The interview prompt, and how to run it

Paste this into a new chat in your Project, change the three lines at the top, and hand the keyboard to the owner:

You are helping me write a standard operating procedure (SOP) for a hotel. I will answer as the person who does this job every day.

Job to document: front desk check-in
Property: independent hotel, 28 rooms, one reception desk, one night auditor
System: our property management system (PMS). I will name screens exactly as they appear.

Rules:
1. Ask ONE question at a time and wait for my answer.
2. Start with what triggers the job and what must be ready. Then walk me through the job in order.
3. After each step, ask what can go wrong at that point and what I do about it.
4. If I'm vague ("check the booking"), ask what exactly I look at or click.
5. Never add a step I haven't described. If you think something is missing, ask me.
6. No real guest names, card numbers or door codes. Call guests "the guest".
7. When I type DONE, write the SOP in the layout below, then list every place you were unsure under "Questions for the owner".

Layout: Purpose / When it applies / What you need / Steps (numbered, one action each) / If something goes wrong (table) / Who to call / Records to keep

Run it as a conversation, not a form. The owner answers in their own words, typed or dictated with the keyboard's microphone key, and says "skip" when a question doesn't apply. A good interview takes 12 to 20 questions. When the answers dry up, type DONE and read the "Questions for the owner" list first: it is where the AI admits what it guessed.

Here is how the first minutes might go (illustrative, for a 28-room independent hotel):

AI:  What starts a check-in, and what should already be in place before the guest reaches the desk?
You: At 14:00 I open today's arrivals list and look at which rooms are marked Clean and Inspected.
AI:  What do you do with the arrivals list once you've looked at it?
You: Flag early arrivals and anyone with a note, like a birthday or a late-arrival warning. I print registration cards for anyone who hasn't done online check-in.
AI:  What can go wrong with the room status at that point?
You: The room still shows Dirty. I ring the head housekeeper for a time, tell the guest honestly, and offer to store their bags.
AI:  What exactly do you say, and how does the guest hear when the room is ready?
You: I give them the time the housekeeper gave me, add ten minutes, and text them when it flips to Inspected.

Two habits in that exchange are worth keeping. The AI asked what exactly the receptionist looks at, so "check the booking" became "open the arrivals list and sort by arrival time". And it asked what goes wrong, so the room-not-ready branch came from the person who handles it daily, not from a generic guide.

Check-in: what the first draft got wrong, line by line

The draft the AI produced from that interview read well, which is the danger: fluent text hides wrong steps. The owner read it at the desk with a real booking on screen and changed the following (illustrative):

The pattern repeats in every hotel SOP. Delete what the hotel doesn't do, replace adjectives with something observable, and replace remembered keystrokes with the real menu path. Then watch the owner do three live check-ins before you sign off. The step they perform without thinking, such as glancing at the booking note for an accessibility request, rarely comes up in an interview, and it is the one a new starter will skip.

The exceptions table does most of the work

A new receptionist can follow ten steps unaided. It is the sixth guest with a declined card who needs the page. Ask the AI to build this table from the owner's "what goes wrong" answers, then add the thresholds and the decision-maker yourself (illustrative values):

Housekeeping: describe clean with photos, then check the AI's reading

Housekeeping is where words fail and photographs work. "Make the bed to standard" means nothing to a new attendant; a photograph of the bed made to standard means everything. The method takes about an hour per room type:

1. Have the head housekeeper make up one room of each type to the hotel's standard, then photograph it empty: the bed from the door and from the side, the pillows, the basin and towel display, the shower, the amenity tray, the desk or kettle area, the wardrobe. Keep guests' belongings and people's faces out of frame.
2. Upload the set to an assistant that reads images. As of October 2026, Claude accepts JPEG, PNG, GIF and WebP, up to 20 files a chat; ChatGPT accepts PNG, JPEG and non-animated GIF, up to 20MB each; Gemini takes up to 10 files per prompt. Limits depend on your plan, so check yours.
3. Ask for checkable statements, one per line, and make the model admit what it cannot see.

These photos show a Standard Double made up to our standard by the head housekeeper. For each photo, write checkable statements an inspector could tick or fail from the photo alone: position, count, fold, direction, and what is NOT there. One statement per line, 15 words or fewer. If you can't see something clearly, write [UNCLEAR] instead of guessing. Then list anything in the photos that looks like a one-off rather than a standard.

A typical response needs editing. Illustrative lines, with what the head housekeeper did to each:

• "Two pillows against the headboard, two smaller pillows in front, open ends facing away from the door." Kept, because it matches the standard.
• "Duvet folded back about 30 cm from the headboard." Removed. A model can't measure from a photograph, and the real standard is "folded back to the first pillow line".
• "Hand towel folded in half on the bath sheet." Rewritten. The photo shows a bath sheet, a hand towel and a bath mat, so the line now names all three and where each sits.
• "Half-full water glass beside the kettle." Deleted. A one-off left in the shot; the model itself listed it under one-offs.
• "[UNCLEAR] amenity tray contents." Filled in by the head housekeeper: two tea bags, two coffee sachets, two sugars, two biscuits, one card.

The photographs then sit beside the text in the SOP, one per statement group, so the attendant checks the room against the picture and the inspector checks the attendant.

Turnover and stayover are two SOPs

A departure clean and a stayover service share a vocabulary but not a sequence, so interview for them separately. Here is the departure turnover as one hotel might write it after the interview (illustrative):

1. Knock, announce "housekeeping" and wait ten seconds before entering. Open the window and curtains.
2. Strip the bed into the laundry bag, checking the pillows and under the bed for guests' belongings. Anything found goes into a labelled bag and the lost-property log.
3. Clear the bins, glasses and crockery to the trolley.
4. Bathroom first: toilet, shower, basin, mirror and floor, then the towels placed as in the photograph.
5. Make the bed to the photograph, dust from high to low, and vacuum towards the door.
6. Restock the amenity tray and kettle area to the list.
7. Stand at the door with the reference photograph on your phone and fix anything that doesn't match.
8. Set the room to Clean in the system. Only the inspector sets Inspected.

The SOP also carries a target time, 32 minutes for a Standard Double, measured by the head housekeeper over ten rooms rather than guessed. That figure matters beyond the SOP: it is the number a rota uses to turn a day's departures into attendant hours. Beyond the sequence, the decisions that belong to the hotel rather than the model are the ones to pull out in the interview:

• Do-not-disturb rule. For example: sign still on at 14:00, knock and phone once, log it, tell the duty manager, never enter without the duty manager.
• Lost property. Bag it, label it with room, date and finder's initials, log it, and hold it for the period your policy sets.
• Maintenance. Log the room number and the fault; mark the room Out of Order if the fault affects safety or sleep. The log itself is covered in logging and chasing maintenance requests.
• Inspection sampling. For example: the head housekeeper checks every fifth departure room, plus any room promised to an early arrival, and only the inspector moves a room from Clean to Inspected.

That last rule links the housekeeping SOP to the check-in SOP, which waits for the Inspected status. It is also where separately interviewed SOPs drift: one says "ready", another "made up", a third "clean". Paste all three into one chat and ask for any term used for two different things, or two terms used for one. Settle one set of status words and use it everywhere. Assigning rooms by arrival time is a separate job, covered in allocating rooms by arrival time.

Night audit: make the AI follow your system, not its memory

The night audit closes the hotel's business day. Someone checks the day's arrivals, departures and no-shows against the bookings, checks that room rates and other charges were posted correctly, balances cash, card and invoiced payments (with the card terminal's batch total matching the system), and only then closes the day and runs the reports. That is the common shape in hotel training guides such as Setup My Hotel's night audit and end-of-day process. The names, the order and what is automatic differ by system.

Some systems can close the day by themselves at a set time. HotelRunner's property management system, for example, lets you choose whether unfinished check-ins, check-outs, open cash registers or pending room moves stop the close or are handled automatically. If yours can do this, the SOP must record which setting your hotel chose and who checks what the automatic close did.

This is the one SOP where the model must not write from memory, because a wrong step here costs money rather than comfort. Feed it two sources: your system vendor's end-of-day help text, and the auditor's spoken walk-through of a real audit, typed or dictated. A screen recording transcribed afterwards also works.

Write the night audit SOP using ONLY the two sources below: (1) the vendor's end-of-day help text, (2) the auditor's walk-through. Use screen and button names exactly as they appear in the sources. If a step isn't in either source, don't write it; list it under "Questions for the owner". If the sources disagree, show both versions and mark the step [CHECK]. Add an "If the balance doesn't match" section using only the auditor's answers.

SOURCE 1 (vendor help text):
[paste]

SOURCE 2 (auditor walk-through):
[paste]

The "[CHECK]" rule earns its place the first time the two sources disagree. In one illustrative case, an SOP written from memory told the auditor to "run End of Day". The system's closing wizard had an optional no-show step on a separate tab, and a new auditor skipped it for six nights; the missing no-show charges surfaced at month end. The corrected line names the tab and says to process every unpaid non-arrival as a no-show before closing.

The branch that matters: the balance doesn't match

This is the part of the night that goes wrong most expensively, and it is easy to leave unwritten. Draw the branch from the auditor's answers, then check each line is something they would really do (illustrative):

1. Don't close the day. Print the discrepancy report.
2. Check the three usual causes in this order: a charge posted after the report was printed; a manual card payment entered against the wrong guest account; a rate change or discount with no note.
3. Fix only what the paperwork proves. Write each correction in the audit log with the time and your initials.
4. Still out after 30 minutes? Call the duty manager. Never force the balance with an adjustment entry.
5. If the card batch won't close, leave it open, note the terminal and the time, and tell the duty manager before the morning shift starts.

Two more sections belong in the same SOP. The first is downtime: what the auditor does if the system or the card terminal is down, with pre-printed registration cards, a paper folio sheet, the support phone number, what must be entered afterwards and who signs it off. The second is the handover: what the morning shift receives, which is open items, no-shows processed, rooms out of order and late arrivals expected. Drafting that report from desk logs is covered in hotel shift handover notes with ChatGPT.

The cold-read test and the sign-off chain

A draft isn't an SOP until someone who doesn't know the job can follow it. Give the draft to a person who has never done the task, such as a new starter or a housekeeper trying the desk, and watch them work through it with a made-up booking or a quiet shift, saying nothing. Write down every pause and every question. Each one is a gap in the text. Fix them, then run it again until a pass raises none.

What a test looks like depends on the job. Check-in takes 20 minutes with a dummy booking. For housekeeping, the tester makes up a room from the photographs and the head housekeeper inspects it against the same pictures. For the night audit, the tester works beside the auditor on a quiet night, or in your system's training mode if it has one. A tester who has to ask "which button?" has shown you a missing screen name.

Once the test passes, the document moves through five stages. Keep the evidence from each so that nobody has to take the SOP on trust:

The read-and-sign sheet can be four columns: name, SOP code, version, date. Its value is the version number. When something goes wrong, "she was trained on v1.1, before the incidentals amount changed" is a very different conversation from "I think she was trained".

Version control a busy hotel will actually keep up

Version control sounds like software work, but for SOPs it comes down to four habits.

• One live copy. Keep every current SOP in one shared folder or wiki. The rule is: if it isn't in the folder, it isn't the procedure. Printed copies carry a footer saying they are uncontrolled and to check the folder for the latest version.
• A numbering rule people can remember. Name files like HTL-FD-001 Check-in v1.2 2026-10-28. Move the first number when a step is added, removed or reordered, and the second when only the wording changes. A new first number means everyone re-signs.
• A change log on the last page. One row per version, with what changed and why.
• Review dates with triggers. Review each SOP once a year, and early after a system update, a new card terminal, a policy change, a new room type or an incident where the SOP was followed and still failed.

A change log for the check-in SOP might read like this (illustrative):

If the folder is Google Docs, use its named versions: Google's help page says to open the version history from the Last edit link, pick a version and choose Name this version. You need edit permission to see the history, and a document holds up to 40 named versions, so name only the approved ones ("v1.1 approved 14 Nov") and you can always restore what staff were actually trained on.

Let the AI write the change-log line, but check its work. Paste the old and new text and ask for every difference as one line, with who needs retraining. Models can miss a changed number inside a long step, so compare the figures yourself. Store the reference photographs in a folder named for the SOP version, and reshoot them when the standard changes, for example with new bedding. Old photographs next to new text are how two standards end up in one hotel.

Staff hours and AI cost for the three SOPs

The AI costs little. A free plan covers this work if you stay inside its message limits, and an individual paid plan has been about $20 a month on Claude (Pro, billed monthly) and ChatGPT (Plus), so one month of either is a sensible ceiling. Check current prices before you subscribe. The real cost is staff time. For a hotel of around 28 rooms, an illustrative budget looks like this:

Start with the SOP behind your most expensive mistake of the last six months. If you've had a balance problem, that is the night audit; if guests complain about rooms, housekeeping; if nothing has gone wrong yet and you have no SOPs at all, check-in is the gentlest first run. The general method for writing any procedure from rough notes is in how to write SOPs with AI, and once the three are signed off, training front-of-house staff to work alongside AI covers how to use them in induction. Who is on the desk, in the rooms and in the breakfast room when each SOP is used is a rota question, handled in building a department-by-department hotel rota.

Questions managers ask before trusting an AI-written SOP

Can ChatGPT write a night audit SOP without my system's manual?

It can write something that reads well, but it will guess the screen names, the order of steps and what your system does when a balance fails. Give it your vendor's end-of-day help text or an auditor's spoken walk-through, and tell it to mark any step it cannot source with [CHECK]. Then have the auditor run the SOP on a quiet night.

How often should a hotel SOP be reviewed?

Set a yearly review date in the header, and add triggers that force an early review: a system update, a new card terminal, a change of policy, a new room type, or any incident where the SOP was followed and still failed. Review the three busiest procedures first, because they drift fastest.

Do staff need to sign every SOP?

A short read-and-sign record is worth keeping for check-in, cash handling and the night audit, because it shows who was briefed on which version and when. For lower-risk procedures, a note in the training log is enough. Re-sign only when the procedure changes in a way that alters what people do.

Is it safe to paste hotel procedures into a chat tool?

Procedures themselves are usually fine. Guest names, card or ID numbers, door, safe and alarm codes, passwords and staff personal data are not. Describe the field instead of its contents, and check the model-training switch in privacy settings on personal accounts or use a business plan.

Further reads

• Restaurant SOP Template: Opening and Closing Checklists With AI — Opening and closing checklists for the breakfast room or bar
• How to Create SOPs From Screen Recordings With AI — A second way to capture system steps for the night audit
• Hotel Front Desk Scripts: Can AI Role-Play Difficult Guests? — Rehearse the exceptions from your check-in SOP
• AI for Hospitality: Which Jobs to Delegate and Which Stay With Staff — Which hotel jobs to hand to AI and which to keep
• AI Voice Agent for a Hotel Front Desk: Calls to Hand Over and Keep — A call-type table, night transfer rules for three kinds of hotel and 14 test calls to run before an AI voice agent answers a real guest.
• AI Agents for Hotels: What They Can Do Today and Where They Fail — What a hotel AI agent can handle now, where it fails on bookings, guest messages and housekeeping, and the tests and hand-off rules to set first.
• Hostel Software: Can AI Handle Guest Messages in Several Languages? — Test AI on check-in times, quiet hours and house rules in five languages with a native checker, then store the approved versions in your hostel software.
• How Can AI Help Hotels? Housekeeping Checks and Maintenance Requests — Room-turn checklists, inspection patterns, fault tickets and rota drafts for a small hotel: what AI produces, how to check it and what stays with a person.
• Smart Room Automation Through AI in Hotels: What a Small Property Can Add — What a small hotel can add to its rooms first: thermostat setback linked to the booking system, window and door sensors, and clear guest consent.
• What Software Do Hotels Use? PMS, Channel Managers and AI Add-Ons — A map of hotel software by job (bookings, rooms and payments), how the systems connect, and the few places where AI earns its keep.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Docs Help (version history and named versions); Anthropic, OpenAI and Google help pages on uploading images; Setup My Hotel's night audit and end-of-day process guide; HotelRunner support page on automatic night audit. Step lists, times and figures in the examples are illustrative.