Ten prompts cover the daily load: replies to booking enquiries, pre-arrival emails, early-arrival and late-arrival answers, service-recovery messages and post-stay thank-yous, plus rate-plan descriptions, package copy, plain-English cancellation terms and upgrade offers. Each works only when you paste in your own facts and leave guest names and booking details out.
ChatGPT knows nothing about your hotel, so left alone it fills the gaps with a plausible spa, a free car park or an award you never won. Every prompt below therefore starts from a fact sheet that you write once, tells ChatGPT to put [ask reception] where a fact is missing, and ends with a "Facts used" list, so you can tick each claim against the sheet in under a minute.
One fact sheet, written once, pasted above every prompt
Most bad hotel copy from ChatGPT traces back to a missing fact. Picture a sheet that says only "Parking: 12 spaces". The draft will call the parking free, or bookable, because both appear in plenty of hotel emails, and a guest arrives expecting what the email promised. Add the price and the rule and the problem disappears. Below is a filled-in sheet for an illustrative 24-room hotel. Every figure is made up for the example, so replace each one with your own.
PROPERTY FACT SHEET (checked 1 Oct; review monthly)
Hotel: [Hotel name], 24 rooms: 14 standard doubles, 6 superior doubles (garden view), 4 family rooms (sleep 4).
Ground floor: 4 rooms (2 standard, 2 superior). All other rooms are up stairs.
Check-in from 15:00. Check-out by 11:00. Reception staffed 07:00 to 22:00.
Arrivals after 22:00: key safe; we give the code by phone on the day.
Rates per room per night, Flexible: standard $149, superior $169, family $199.
Advance Saver: standard $134, superior $152, family $179. Paid in full at booking, non-refundable.
Flexible cancellation: free until 15:00 two days before arrival; after that the first night is charged.
Breakfast: 07:30 to 10:00, served at the table, $14 per adult, $7 per child under 12.
Parking: 12 spaces, $10 a night, first come first served, cannot be reserved.
Dogs: welcome in the 4 ground-floor rooms, $15 per stay, two dogs maximum.
Luggage store: free, before check-in and after check-out.
Not available: spa, evening restaurant, room service, lift.
Voice: warm, brisk, plain English, no exclamation marks. Sign off "[Your first name], front desk".Four habits make the sheet work.
- List what you don't have. The "Not available" line does more than any other, because it stops ChatGPT adding the facilities that a typical hotel email mentions.
- Put the rule beside every price: "first come first served", "cannot be reserved", "per stay, not per night".
- Date it and re-read it on the first of each month. Breakfast hours, parking charges and seasonal rates drift.
- Keep it to one page. A long sheet buries the fact you needed under the ones you didn't.
Keep the sheet somewhere it gets pasted for you. In ChatGPT, a Project can hold it as instructions or as an uploaded file, so every new chat starts with it. OpenAI offers Projects on the free plan too, with a lower file limit, so check what your plan allows. Claude has Projects that work the same way. Don't build a custom GPT for this: creation has ended, and existing ones stop running on 11 December 2026.
The other half of the habit is what stays out of the chat. Copy the guest's message into a notes app first, and strip whatever identifies them before you paste it.
| What the guest's message contains | What to paste instead |
|---|---|
| Name, email address, phone number | [guest], or nothing at all |
| Booking reference | [ref], which you add back yourself after the draft is done |
| Card or payment details, passport or ID numbers | Nothing. These never go into any prompt |
| Medical, dietary or access needs tied to a named person | Only the need you must answer, as "[access need: ground floor]", without the reason |
| Other guests' names, staff names | A role, such as "night porter" |
A personal ChatGPT account has a model-training switch in its privacy settings, and business plans don't train on your content by default. Neither fact is a reason to paste names, and the merge tags in the prompts below mean you never need to: your booking system or email tool fills [first name] from the reservation when the message is sent.
The wrapper that stops ChatGPT guessing
Each prompt in this tutorial is the TASK part only. Paste the wrapper first, put your fact sheet into its slot, then add the task underneath.
You are drafting an email for [Hotel name]. Use only the facts in the FACT SHEET below.
If the task needs a fact that is not on the sheet, write [ask reception: what is needed] instead of guessing.
Do not add facilities, discounts, awards, times or reassurances that are not on the sheet.
Write in British English, in plain words, with no exclamation marks.
After the draft, list "Facts used": every figure, time and rule you relied on, quoted from the sheet.
FACT SHEET:
[paste the sheet here]
TASK:
[paste one of the ten tasks below]Three lines do the heavy lifting. The [ask reception] instruction turns a gap into a visible flag instead of a quiet invention. The "do not add" line is a general ban that catches extras you didn't think to forbid. And the "Facts used" list changes the check from rereading the whole email to ticking five lines against the sheet: if the list quotes something the sheet doesn't say, you've found an invention. ChatGPT can still slip, which is why every example below ends with what to fix.
Six guest emails, from first enquiry to thank-you
These six cover most of what a front desk writes. Each shows the task, a sample of what might come back (illustrative; yours will differ) and the lines that need fixing before it goes out.
1. Reply to an enquiry from your own availability
Enquiry replies are the emails most likely to turn into a booking, and the slowest to write, because the answer needs sums. ChatGPT can set out the options and the tone; you supply the availability you have checked and the rates from the sheet.
TASK: Reply to the enquiry below. Offer the two options on my availability line, with the total for the stay worked out line by line (nights, dog fee, parking).
Say we will confirm anything that is not on the fact sheet by the end of the working day. Mention a benefit of booking direct only if the fact sheet lists one. 130 words maximum.
ENQUIRY (identifiers removed):
"Hi, two adults and a dog, Friday and Saturday night next month. Anything on the ground floor? What would it cost, and is parking easy?"
AVAILABILITY I HAVE CHECKED: ground-floor standard double and ground-floor superior double, both free both nights.A typical draft:
Subject: Friday and Saturday night with your dog
Hello [first name],
Thank you for getting in touch. Both ground-floor rooms are free for those two nights, and both welcome dogs. The standard double is $149 a night ($298 for two nights). The superior double has a garden view and is $169 a night ($338 for two nights).
Add $15 for the dog, charged per stay, and $10 a night for parking. That makes $333 for the standard double and $373 for the superior. Parking is first come, first served, although there is usually space on a Friday.
You can book directly on our site [link] and we guarantee our best rate. Anything else you would like to know, just ask.
[Your first name], front desk
Facts used: standard $149, superior $169, dog $15 per stay, parking $10 a night first come first served, dogs in ground-floor rooms.
Two lines aren't on the sheet: "there is usually space on a Friday" and "we guarantee our best rate". The sheet says parking is first come, first served and lists no direct-booking benefit, and the prompt said to mention one only if the sheet had one. The Facts used list gives both away, because nothing in it supports either claim. Cut them, then redo the sums yourself: 2 × $149 = $298, plus $15 for the dog and $20 for two nights' parking makes $333; 2 × $169 = $338, plus the same $35, makes $373. ChatGPT usually adds small numbers correctly, but this is money you are quoting.
2. The pre-arrival email guests read to the end
Send it five days out, with one question at the end. A list of ten facilities gets skimmed; what guests scroll back for on the day is how to get in, where to park and when breakfast stops. If you would rather schedule these, automating pre-arrival and post-stay emails covers the sending side.
TASK: Write our pre-arrival email, sent five days before arrival. Cover, in this order: check-in time and what happens if they arrive after 22:00; the parking rule and price; breakfast times and price; the luggage store.
End with one question only: what time do they expect to arrive?
Subject line under 50 characters. 150 words maximum. Leave [first name] and [room type] as merge tags, exactly as written.
BOOKING DETAILS: [room type], [nights], extras booked: [dog / parking / none]What came back:
Subject: Your stay with us: arrival details
Hello [first name],
We are looking forward to welcoming you. Here is what you need on the day.
Check-in is from 15:00, and reception is staffed from 07:00 to 22:00. If you arrive later than that, phone us on the day and we will give you the key-safe code. Your [room type] will be ready and waiting for you.
Parking is $10 a night, first come, first served, so we cannot reserve a space. Breakfast is served at the table from 07:30 to 10:00 and costs $14 per adult. You are welcome to leave bags in our free luggage store before check-in or after check-out.
What time do you expect to arrive?
[Your first name], front desk
The slip is "ready and waiting for you". The sheet says check-in is from 15:00, so a guest who reads that line, lands at 11:00 and finds no room has a screenshot to show you. Pre-arrival emails are where a stray promise does the most damage, because guests keep them. Delete the sentence, or replace it with the rule: "Your room is ready from 15:00." The merge tags did their job: ChatGPT never saw a guest's name, and your email tool fills [first name] and [room type] from the reservation at send time.
3. Local tips from a list you have checked yourself
This is where "fact-check the local tips" matters most. ChatGPT has no idea which café closed in the summer or which path floods in autumn. So the prompt gives it your list and forbids it to add anything. (For a fuller method, see how to give local tips without wrong answers.)
TASK: Write an "Around the hotel" paragraph of 90 to 110 words for the pre-arrival email, using only the places in MY CHECKED LIST.
Do not add dishes, views, prices, opening hours or opinions that are not in the list. Include each opening-hours note as written.
End with: "Hours can change, so please check before you set out."
MY CHECKED LIST (checked 24 Oct, by phone or on each venue's own page):
- Bakery next door: 1 minute's walk; opens 07:00; closed Mondays
- Riverside path: flat; 25 minutes on foot to the old bridge
- Pub on the corner: 4 minutes' walk; food served 12:00 to 21:00, Tuesday to Sunday; kitchen closed Mondays
- Cinema: 10 minutes' walk; times are on its own websiteA typical result:
Next door, the bakery opens at 07:00 and is closed on Mondays, a one-minute walk from reception. If you would like to stretch your legs, the riverside path is flat and takes about 25 minutes on foot to the old bridge. The pub on the corner is four minutes away and serves food from 12:00 to 21:00, Tuesday to Sunday, with the kitchen closed on Mondays. The cinema is ten minutes on foot, and its times are on its own website. The bakery's sourdough is famous locally, so go early. Hours can change, so please check before you set out.
"Famous locally, so go early" is not in the list. It reads as friendly, which is why it survives if you skim, and it is exactly the kind of line that sends a guest to a queue or a closed door. Strike it. Then do the part ChatGPT cannot: before each season, ring or check each venue's own page, update the list and change the "checked" date. Keep that date on your list, not in the email, and treat any entry older than a month as unverified.
4. Early arrival and late arrival, with the decision made first
The decision is yours; ChatGPT only words it. Put the decision and your reason in the prompt, and ban the soft phrases that turn a "no" into a "maybe".
TASK: Reply to a guest asking for an early check-in. Our decision is below. Write it kindly and plainly, in 70 words or fewer.
Do not promise anything that is not in the DECISION line. Do not use "try", "hopefully", "do our best" or "as soon as possible".
GUEST REQUEST: "We land at 10 am. Any chance of getting into the room when we arrive?"
DECISION: No early check-in (house is full the night before). Room ready from 15:00. Free luggage store, usable whenever reception is open.What came back:
Hello [first name],
Thank you for asking. We are full the night before, so we cannot offer an early check-in, and your room will be ready from 15:00. You are welcome to leave your bags in our free luggage store as soon as you arrive; reception is open from 07:00. We will do our best to have the room ready a little earlier if we can.
[Your first name], front desk
The last sentence undoes the decision, and it appears despite the ban on "do our best". Models sometimes ignore a negative instruction, so add a sweep to your routine: search every draft for "try", "hopefully", "best", "soon" and "possible" before sending. For a late arrival the same task works with a different decision line, for example: "DECISION: guest arrives about 23:30 by train; give the key-safe code by phone; ask them to ring when ten minutes away." Note that the sheet says we give the code on the day, so the draft must not put the code itself in an email.
5. Service recovery that offers only what you decided
Complaint replies fail in two ways: they argue, or they promise. This prompt puts the facts and your offer in a fenced block and forbids everything outside it. The wider method sits in service-recovery scripts for hotels.
TASK: Write a reply to the complaint below. Structure: thank them for telling us; apologise for what they experienced (not for anything I have not confirmed); state the facts from FACTS FROM OUR LOG; make the OFFER exactly as written; invite a phone call. 120 words maximum.
Do not mention any action, repair, refund, discount or future stay that is not in OFFER or FACTS.
COMPLAINT (identifiers removed): "Cold shower on Saturday morning and nobody told us what was going on. Poor for $169 a night."
FACTS FROM OUR LOG: Cold water reported by phone to reception at 07:10; boiler reset by 08:15; no other room affected; no record of a call back to the guest.
OFFER: Refund of breakfast for both mornings ($28 in total) to the card used for booking.A first draft:
Hello [first name],
Thank you for telling us, and I am sorry your Saturday began with a cold shower. Our log shows you phoned reception at 07:10 and the boiler was reset by 08:15. Only your room was affected. We should also have called you back to explain what was happening, and we did not.
I would like to refund breakfast for both mornings, $28 in total, to the card you booked with. We have also had the boiler fully serviced, so this will not happen again.
If you would like to talk it through, please phone me on [number].
[Your first name], [role]
"Fully serviced, so this will not happen again" is an invented repair and an unkeepable promise, and both sit outside the OFFER and FACTS lines, which is the part the prompt tried to forbid. Cut it. The line "We should also have called you back" is different: ChatGPT drew it from "no record of a call back". Whether to admit that is your decision, not the model's. Keeping it is honest and often calms a guest down; just make it a choice. Anything involving injury, illness, an allergic reaction, discrimination or a legal threat does not go through ChatGPT at all.
6. The post-stay thank-you that goes to every guest
The review invitation needs the most care. Google's policy on prohibited and restricted content lists, among behaviours that count as rating manipulation, offering incentives such as payment, discounts or free goods or services, discouraging negative reviews or selectively soliciting positive ones, and asking staff to collect a set number of reviews. So the email goes to every guest, offers nothing, and does not steer the rating. For replies once reviews arrive, see replying to hotel reviews without sounding canned.
TASK: Write a short thank-you email for the morning after check-out, sent to every guest. 90 words maximum.
Thank them for staying. Say we would like to hear how it went. Give the review link [link] and a private route for feedback [reply to this email].
Offer nothing in return. Do not ask for a particular star rating. Do not write "if you enjoyed your stay" or anything else that only invites happy guests to respond.What came back:
Subject: Thank you for staying
Hello [first name],
Thank you for staying with us. We hope it was a five-star stay, and whatever you thought, we would like to hear it. If you have a few minutes, you can leave a review here: [link]. If you would rather tell us privately, simply reply to this email and it will reach the front desk.
[Your first name], front desk
"We hope it was a five-star stay" steers the rating, so it goes. The rest is fine, and the discipline matters more than the wording: send it to every guest, not a list filtered by how the stay went. Search each draft for "enjoyed", "star", "discount" and "free" before it leaves.
Four pieces of rate copy that must match your numbers
Rate copy is where a small slip costs real money, because guests quote your page back to you. The prompts here run on the same sheet, with one addition: the rate ladder below, which all four of them draw on. The Advance Saver is priced about 10% below Flexible, rounded down to whole dollars.
| Room | Flexible, per night | Advance Saver, per night | Saving per night |
|---|---|---|---|
| Standard double | $149 | $134 | $15 |
| Superior double (garden view) | $169 | $152 | $17 |
| Family room | $199 | $179 | $20 |
Deciding the rates is your job, and so is the cancellation rule. The prompts below only describe them, which is why every number in the prompt has to come from you.
7. Rate-plan descriptions for the booking engine
Booking engines give each rate plan a name, a short summary and a "good to know" line, usually with tight character limits. Check the limits in your own engine, because they differ, and put them in the prompt.
TASK: Write booking-engine copy for three rate plans on the superior double: Flexible, Advance Saver, and Bed and Breakfast.
For each give: a name (30 characters maximum), a summary sentence (140 characters maximum) and a "good to know" line (220 characters maximum) that begins with the cancellation rule.
State every saving in dollars per night and for two nights, not as a percentage. Show the sums you used.
RATES, superior double per night: Flexible $169. Advance Saver $152, paid in full at booking, non-refundable. Bed and Breakfast = Flexible plus $24 per night for breakfast for up to 2 adults (the table price is $14 per adult).
CANCELLATION: as on the fact sheet.A first pass, set out as a table:
| Plan name | Summary | Good to know | Two nights |
|---|---|---|---|
| Flexible Rate | Our standard rate: change or cancel free until 15:00 two days before you arrive. | Free cancellation until 15:00 two days before arrival, then the first night is charged. Pay at the hotel. | $338 |
| Advance Saver | Pay now and save 17% against the Flexible rate. | Paid in full at booking. Non-refundable: no refunds or changes. Garden-view superior double, room only. | $304 |
| Bed and Breakfast | Flexible rate with breakfast for two included, $4 a night cheaper than paying at the table. | Free cancellation until 15:00 two days before arrival, then the first night is charged. Breakfast is served 07:30 to 10:00. | $386 |
Three checks catch the problems here. First, "Pay at the hotel" appears nowhere on the sheet; delete it or replace it with your real payment rule. Second, "save 17%" ignores the dollars instruction: $169 − $152 is a $17 saving per night and $34 over two nights, which is about 10% (17 ÷ 169), not 17%. Models sometimes turn a dollar figure into a percentage without noticing. Third, redo the totals: 2 × $152 = $304, and 2 × ($169 + $24) = $386. The breakfast claim holds: two adults at the table cost $28 a night, so $24 saves $4. ChatGPT also miscounts characters, so paste each line into a spreadsheet and use =LEN() against your engine's limits before you trust the counts.
8. A package page with the saving worked out
A package earns its place only if the saving is true, and guests check it. Give ChatGPT the parts and their separate prices, and make it show the subtraction.
TASK: Write a web page section for our "Midweek Escape" package: a heading (60 characters maximum), a description of about 90 words, a bullet list headed "Included", a bullet list headed "Good to know", and two email subject lines of 45 characters or fewer.
Use only the PACKAGE FACTS. Show the saving against buying the parts separately, worked out line by line.
PACKAGE FACTS: 2 nights; arrival Sunday to Thursday only; superior double (garden view); breakfast for 2 adults each morning; check-out until 13:00, subject to availability on the day; $355 per room for the stay; not valid 20 December to 3 January; cancellation as the Flexible rate.
BOUGHT SEPARATELY: 2 nights at $169 = $338; breakfast for 2 adults on 2 mornings at $14 = $56.A typical result:
Heading: Midweek Escape: two nights, breakfast and a later check-out
Two nights in a superior double with a garden view, breakfast for two every morning and a later check-out, for $355 a room. Arrive on any day from Sunday to Thursday. Bought separately, the room and breakfasts come to $394, so the package saves $39 before you count the late check-out. A complimentary welcome drink on arrival sets the tone, and our free luggage store means you can explore before your room is ready.
Included: two nights in a garden-view superior double; breakfast for two adults each morning; check-out until 13:00, subject to availability.
Good to know: arrival Sunday to Thursday only; not valid 20 December to 3 January; cancellation as the Flexible rate.
Subject lines: "Two nights, breakfast and a later check-out" (43 characters); "Midweek Escape: $355 for two nights" (35).
The sums are right: $338 + $56 = $394, and $394 − $355 = $39. The invention is the "complimentary welcome drink", which is on neither the sheet nor the package facts. Guests do not forget a free drink you never meant to give. Also check the heading length yourself: it comes to 59 characters, inside the 60 you allowed but with no room to edit. If the late check-out is only "subject to availability", the copy should say so wherever it appears, as this draft does.
9. Change and cancellation terms in plain words
Guests email to ask "what happens if I cancel?" because the terms read like a contract. A plain summary above the full terms answers that, as long as it means exactly what the contract says. The prompt asks for a comparison list so you can check.
TASK: Rewrite the clause below as a plain-English summary for the booking page: three short sentences at most, plus a one-line version for the confirmation email. Keep every condition, time limit and charge. Do not soften, add or remove any.
After the summary, make a two-column list: each condition in the original, then the words you used for it.
OUR CLAUSE: "Cancellations received less than 48 hours before 15:00 on the day of arrival will be charged the first night's accommodation. No-shows will be charged the first night. Advance Saver bookings are paid in full at the time of booking and are not eligible for refund or amendment save at the discretion of management."What came back:
In short: you can cancel for free until 48 hours before you check in. After that, or if you do not arrive, we charge the first night. Advance Saver bookings are paid in full when you book and cannot be refunded or changed, unless we agree otherwise.
One line for the confirmation email: free cancellation until 48 hours before check-in; after that, one night is charged.
The drift is in "48 hours before you check in". The original pins the deadline to 15:00, two days before the day of arrival. A guest who arrives at 18:00 could read "check in" as their own arrival time and cancel three hours later than your clause allows. This is why the comparison list matters: put the original condition and the new wording side by side and tick each one. The plain version is a summary, so label it "In short" and link the full terms, which remain what binds. Ask whoever drafted your terms, or a qualified adviser, to confirm that the summary doesn't contradict them. ChatGPT is not a source of legal advice, and nothing here is.
10. The upgrade offer, with a limit that is real
Upgrade offers work best in the pre-arrival email, and they backfire when the scarcity is invented. The pricing ladder and timing are covered in upselling room upgrades and late checkout; this prompt only writes the offer.
TASK: Write a 70-word upgrade offer to add to the pre-arrival email for guests booked in a standard double: the superior double with a garden view for $20 more per night. Ask them to reply to claim it.
Use only the facts below. Do not state how many rooms are left.
FACTS: Upgrade available on request; up to 3 upgrades a night, first come first served; confirmed by reply within 24 hours; the only difference we promise is the garden view.What came back:
Fancy waking up to the garden? For $20 more a night, you can upgrade your standard double to a superior double with a garden view. Only two rooms are left for your dates, so reply to this email to claim yours, and we will confirm within 24 hours.
"Only two rooms are left" ignores the instruction, and it has no source: the facts say three upgrades a night, not how many remain on a given night. Real scarcity is fine, but it must come from your booking system on the day you send, never from the model. Either take the number from the system or leave the claim out. The draft was right to promise only the garden view, which is all the sheet supports.
A five-minute routine that catches what a quick read misses
Every example above had a slip, and most would get past a quick read. A fixed routine finds them faster than a careful read does.
- Tick each line of the "Facts used" list against the fact sheet. Anything unsupported is an invention.
- Search the draft for the words in the table below.
- Redo every sum on a calculator, including nights, extras and savings.
- Check the merge tags: [first name] and [room type] appear exactly as your system expects, and no real guest name has crept in.
- Read the draft aloud once. If it sounds like a brochure, cut the adjectives.
| Word or phrase to search for | Why it appears | What to do |
|---|---|---|
| guarantee, best rate | A promise copied from other hotels' emails | Delete unless the sheet says it |
| free, complimentary | Extras you do not offer | Match against the sheet; keep only what is listed |
| usually, normally, rarely | Made-up reassurance | Replace with the sheet's actual rule |
| ready and waiting, as soon as possible, do our best | Soft promises about timing | Delete, or state the fixed time |
| only [number] left | Invented scarcity | Use your system's count, or remove |
| award-winning, famous, popular | Puffery | Delete |
| save [number]% | A dollar saving restated as a percentage | Recalculate, and state it in dollars |
Keep a short edit log in a notes file: the prompt number, what you changed and why. After twenty entries the repeats will be obvious, and each repeat becomes a new line in the wrapper, such as "Never write 'usually'." That is how the wrapper improves with use. Once the ten prompts work for you, store them where the whole desk can find them; building a shared prompt library shows one way.
A week at the front desk: 202 minutes against 93
Take an illustrative week at the 24-room hotel: 18 enquiries, 3 early or late arrival requests and 1 complaint. Suppose a reply takes 9 minutes by hand for an enquiry, 5 for an arrival request and 25 for a complaint. With the prompts, assume 4, 2 and 15 minutes, because pasting, ticking the Facts used list and redoing the sums still takes time, and the complaint still needs thinking.
| Message type | Count | By hand | With the prompts |
|---|---|---|---|
| Enquiry replies | 18 | 18 × 9 = 162 minutes | 18 × 4 = 72 minutes |
| Early or late arrival | 3 | 3 × 5 = 15 minutes | 3 × 2 = 6 minutes |
| Complaint | 1 | 25 minutes | 15 minutes |
| Total | 202 minutes | 93 minutes |
That is 109 minutes saved, a little under two hours a week. The numbers are placeholders: time five of your own replies each way before you believe any of them. The pre-arrival email and the thank-you add nothing to the weekly count once they are templates with merge tags. Two other measures tell you more than minutes. One is the share of drafts that go out with no edits; if it is high, check that you are really checking. The other is the number of "can I cancel?" and "what's included?" emails after the plain-English terms and rate copy go live. If those do not fall, the copy is not clearer than what it replaced.
Where these prompts should stop
- Safety, injury, illness and allergic reactions. A draft can admit or promise something that matters later. A person writes these.
- Legal threats, chargebacks and data requests. Owner or adviser only.
- Accessibility questions. "Is the bathroom step-free?" must be answered from a measured access statement you wrote. ChatGPT will happily agree with whatever the question implies.
- Group, wedding and corporate contracts. It can draft the cover note; the contract wording is yours or your solicitor's.
- Anything about another guest. Do not paste it, and do not ask for a reply that mentions them.
Inside those limits, the pattern is the same ten times: your facts in, a labelled gap where a fact is missing, a list of what was used, and a person who ticks it. The prompts save time on the writing. The sheet and the check stay yours.
Questions hotel owners ask before they start
Do I need a paid ChatGPT plan to use these prompts?
No. The prompts run on any plan, and paid plans mainly raise usage limits. What matters for a hotel is data handling: on a personal plan, switch off the model-training setting in privacy settings; business plans don't train on your content by default. Either way, keep guest identifiers out of what you paste.
Should guests be told an email was written with AI?
An email a member of staff has read, edited and sent is your message, and the responsibility for it is yours either way. A chatbot that answers guests without a person checking is different: tell guests it is automated. Check any disclosure rules that apply to your business, because they vary.
Can I use these prompts for guests who write in another language?
Yes. Ask for the reply in the guest's language, then have someone who reads that language check every figure, time and condition before you send it. Mistakes in translated cancellation terms or arrival instructions cost more than mistakes in tone, so keep the English version beside it.
Can I run these prompts in Claude or Gemini instead?
Yes. The wrapper and the ten tasks are plain instructions, so they work as written. Claude Projects hold the fact sheet the same way ChatGPT Projects do, and in Gemini a Gem (becoming a skill from November 2026) can carry it as standing instructions. Whichever you use, keep the Facts used list and the five-minute check, because every model can add a detail that isn't on your sheet.
Further reads
- Hotel Email Marketing With AI: Past-Guest Segments and Subject Lines — Take the same drafting method to past-guest marketing emails.
- How Independent Hotels Win More Direct Bookings With AI — Where direct-booking copy fits in a wider plan.
- How Accurate Is ChatGPT? What Owners Should Expect by Task — Which tasks ChatGPT gets wrong most often, and how to check.
- Is AI Dynamic Pricing Worth It for a Small Hotel? — Setting the rates themselves, which these prompts leave to you.
- ChatGPT for Hotels: What It Can and Cannot Do at the Front Desk — The front-desk jobs ChatGPT should never be given.
- How Small Hotels Use AI to Answer Guest Messages Around the Clock — Answering guest messages overnight, with a checked fact sheet.
- How to Use ChatGPT Projects to Keep Client Work Separate — Setting up Projects so each fact sheet stays in its place.
- AI Consulting for Hospitality: What a Hotel Should Ask Before Hiring — Twelve questions to put to a hotel AI consultant, the month-one deliverables to expect, and a sample plan with numbers you can test.
- Hotel Guest Feedback: Turn Post-Stay Surveys Into Fixes With AI — How a hotel can turn a month of post-stay survey comments into a ranked fix list with AI labelling, spreadsheet counts and named owners.
- Hotel Rate Shopping: A Weekly Competitor Rate Sheet Built With AI — A weekly competitor rate sheet for a small hotel: pick five rivals, fix the search, capture rates by hand, then let AI summarise and flag what moved.
- WhatsApp for Small Hotels: Meta's AI Rules, Templates and Handover — What Meta's rules allow when AI answers hotel guests on WhatsApp: the free app versus the Platform, templates, handover, opt-in wording and a test plan.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Maps user-generated content policy (prohibited and restricted content, re-checked October 2026); OpenAI's help pages on Projects and the custom GPT retirement. All hotel figures in the examples are illustrative.