Meet the AI Concierge: How AI-Powered Luxury Travel Actually Works

Most of the artificial intelligence on travel websites is a chatbot that can tell you the opening hours. Ours runs the front desk: it checks the calendar, quotes the exact total, takes the booking and hands you a payment link. This is what it does, what it deliberately does not do, and what happens to what you tell it.

• What it actually does
• What it deliberately does not do
• Where the humans are
• Asking it from ChatGPT or Claude
• Speaking to it, and rating it
• What happens to what you tell it
• How to test it properly
• Questions we are asked most

What it actually does

Message the concierge the way you would message a well-connected friend: two of us, near the Wharf, the last weekend of October. In one conversation it will do all of this.

Check real availability, live against the booking calendar, not against a brochure written last season. If the dates are gone, it says so and offers the ones that are not.

Quote the exact total. The nightly rate, the cleaning fee and the five per cent service fee, added up, for your dates. Not "from" a number. If something is charged on top, a stay with a dog for instance, it tells you that too, before you decide.

Compare across everything we have. Our own apartments in Canary Wharf run from The Mellish Residence at £275 a night to The William Jessop at £580 for a two-bedroom sleeping seven. Elsewhere in London it arranges an apartment from our curated list, and it says whose apartment it is.

Take the booking. It creates the reservation and hands you a secure checkout link in the chat. You pay on our own page, the calendar blocks the moment the payment goes through, and the confirmation reaches your inbox.

Brief the aviation desk. Tell it a route and a date and it captures the brief for private jet charter, which we arrange as an agent with flights operated by licensed carriers.

It does this at three on a Sunday morning exactly as well as at three on a Tuesday afternoon, which is the entire point.

What it deliberately does not do

It does not invent. Every price and every date comes from the live system. It has no ability to offer a discount that does not exist, invent availability, or confirm something that has not happened. When it does not know, it says so and passes the question to a person.

It does not take card details. Ever, in any form. Payment happens on our own checkout page, operated by Stripe, and nobody at Luxury Experience AI will ask you for a card number in a chat, an email or a phone call. If something claiming to be us does, it is not us.

It does not pretend to be a person. It tells you it is an AI assistant, because the alternative is a small deception that poisons everything else.

It does not hold anything without payment. A quote is not a booking and a conversation is not a hold. Dates are held when the payment completes, first come, first served, which is a plainer answer than most of this industry gives.

Where the humans are

Behind it, from 09:00 to 22:00, London time, and reachable outside those hours for anything that has already gone wrong.

The division is deliberate. The things a machine does better, checking a calendar in a second, quoting a total without arithmetic errors, answering at four in the morning in whatever language you prefer, it does. The things a person does better, a supplier who needs persuading, a family with a difficult set of requirements, a complaint that deserves a human voice, go to the team.

Every conversation is readable by that team, and they read them. When the concierge answers badly, the instructions change. That feedback loop is the difference between an assistant that improves and one that annoys people in the same way for two years.

Asking it from ChatGPT or Claude

Newer than the chat window, and the part most people have not seen: you can now reach the same catalogue from inside the assistant you already use.

We publish a connector built on the Model Context Protocol, the open standard for connecting assistants to outside services, supported by both Claude and ChatGPT. Connected, an assistant can search our apartments, read one in detail, check dates, quote a stay and, if you ask it to, start a booking and hand you a link to our own checkout.

Three things are true of that by design. The assistant asks your permission before it does anything that creates a booking. No card details pass through it, because payment happens on our page. And nothing an assistant says is a binding quote: our confirmation email is the contract, which our terms state plainly.

It is early, and the assistants themselves are still learning what to do with a connector like this. But a guest who plans a trip inside ChatGPT can now get a real price from us there, rather than a guess assembled from an old page.

Speaking to it, and rating it

Two smaller things that change how it feels to use.

You can talk to it. Press the microphone and speak, and it transcribes what you said; press the speaker under any reply and it reads the answer aloud. Useful when you are cooking, driving is not, and genuinely useful for anyone who finds typing on a phone a chore.

You can mark an answer good or bad. There is a thumb under every reply. It is not decoration: the team reads the ones marked down, and where the concierge was wrong or clumsy, the instructions change that week. If it gives you a poor answer, press it. That is worth more to us than a polite exit.

Neither costs anything, and neither requires an account. You can use the whole thing, from first question to payment link, without creating a login. An account is useful if you want your bookings in one place, and it is optional.

What happens to what you tell it

Worth knowing, and easy to find out, which is not true everywhere.

Your messages are processed by Google or Anthropic, whose models write the replies. Under the developer terms we use, they do not train their models on what we send them. We keep a record of the conversation on our own servers so the team can follow up and improve the answers, anything that looks like a card number is blanked before it is stored, and a conversation that leads to nothing is deleted after ninety days. If you leave your details, they go into an enquiry, not a mailing list somebody else rents.

All of that is set out, with the retention periods and the providers named, in our privacy policy.

How to test it properly

Do not ask it the weather. Ask it something that would embarrass a chatbot.

What would five nights in a two-bedroom cost in October, for four adults and a dog? It has to check availability, apply the right apartment, add the fees, and know which of our apartments takes dogs.

Can I have the apartment from the 3rd to the 7th, and is that cheaper than the 10th to the 14th? Two live checks and an honest comparison.

I need somewhere in Mayfair, not Canary Wharf. It should tell you that we do not own an apartment in Mayfair and offer to arrange one from the curated list, naming whose it is.

If an assistant cannot do those three things, it is decoration. Ours is meant to be the front desk.

Questions we are asked most

Am I talking to a person or a machine? A machine, and it says so. A person reads the conversation and takes over when it matters, from 09:00 to 22:00, London time.

Can it actually book, or does it just take a message? It books. It creates the reservation and gives you a secure payment link, and the dates are held once that payment completes.

Will it ever ask for my card details? No. Payment happens on our own checkout page. Anything asking for a card number in a chat is not us.

Does it make up prices? It cannot. Prices and availability come from the live system, and the total it quotes is the total you pay, with anything extra named separately.

Can I use it from ChatGPT or Claude? Yes, through our connector. The assistant will ask your permission before it starts a booking, and you still pay on our own page.

What happens to my conversation? It is kept on our servers so the team can follow up and improve the answers, with card-like numbers blanked, and deleted after ninety days if it leads to nothing. The detail is in our privacy policy.

Try it on something real

Give it a brief with a constraint in it: a dog, a budget, a date that might not be free, a part of London where we do not own anything. The answer will tell you more about how we work than this article can.

Ask the concierge whatever you were going to ask a hotel, and ask it the awkward version of the question rather than the polite one.

Last checked against our catalogue on 23 September 2026. Providers and retention periods change, so we say what we checked and when.