Same question, three right answers

"Is there a pool?" has one fact behind it and many right answers. Your AI needs your tone and your house rules, not just your facts.

That is the part hoteliers underestimate when they look at AI on their chat. Accuracy is the floor, not the goal. A guest who gets a true fact in the wrong voice, at the wrong level of detail, or with the wrong number attached to it, still walks away.

Voice is part of the answer

Three properties, one fact, three correct replies:

• A lifestyle hotel brand wants its assistant warm, personal and a little witty: "Yes, and it's a very good reason to pack a swimsuit."
• A boutique chain wants short and to the point, with a tasteful emoji at most: "Yes, heated. Open 07:00 to 21:00."
• A family attraction wants festive excitement: "How fun that you're celebrating with us!"

The same sentence in the wrong voice feels like a different brand. Put the witty line on the boutique chain's chat and it reads as flippant. Put the clipped line on the family attraction and it reads as bored.

Your website copy already encodes this. Your chat usually does not, because nobody has ever written the rules down. The front desk simply absorbs them by sitting next to someone who has them.

A tone brief does not need to be long. Three or four lines will do:

• How long is a normal answer? One sentence, or two and a link?
• Emoji: never, sparingly, or freely?
• Do you use the guest's first name? Do you use "you" formally or informally in your other languages?
• Which words do you never use? Many properties ban "unfortunately", "unable" and "policy".

That last one is worth doing properly. A ban-list is the fastest way to make an assistant sound like your brand rather than like a call centre.

Who is talking?

One chain introduces its assistant by name as a digital agent and adds that a human team is available. Another brand gives the assistant a name and a character and never calls it digital.

Both are valid choices. Neither is the default. The important thing is that the choice is yours and that the assistant holds it consistently, including when a guest asks directly, "am I talking to a robot?"

Decide now, because the answer to that question will be screenshotted eventually. Write the exact sentence you want used, rather than leaving the assistant to improvise a philosophical position on your behalf. It is a bigger decision than it looks, and it deserves an answer of its own.

The same applies to the handover. If a human is taking over, does the guest get told? A brand that presents its assistant as a named character usually wants the transition to be invisible. A brand that says "digital agent" usually wants it announced. Pick one.

Price is never "just the price"

This is where a generic assistant does the most damage, because every decision below is a house decision:

• Do you show the total for the stay or the nightly rate? A high-value five-night stay looks very different depending on which you lead with.
• VAT. Foreign tourists and locals may pay different amounts. The guest's language is not a reliable signal: an Israeli may write to you in English, and a tourist may write in Hebrew. If the difference matters to your pricing, the assistant has to ask rather than assume. You also need to decide how it asks without sounding awkward.
• Do you add a line such as "prices are correct at the time of this quote"? If your rates move with demand, you probably should.
• Some businesses keep prices out of the chat entirely and send them in a quote document instead, because the quote is the thing they want to control.

None of these is a preference you can leave to a supplier's default. Write down, on one page: which figure you lead with, what is included, what is excluded, what the caveat sentence says, and when the assistant must stop quoting and hand over.

What not to say

A guesthouse sends one check-in morning message containing a precise map link, the after-hours gate code, pool and breakfast hours, and the WiFi password. It deliberately leaves out the room number, because rooms change at the last minute.

That omission is not an oversight. It is the most carefully considered part of the message. Sending a room number that later changes creates a guest standing outside the wrong door with a code that does not work, at eleven at night, with nobody on reception.

Every property has one or two of these. Facts that are true, that the guest would like to have, and that you must not send early. The usual suspects:

• Room or unit numbers before the room is firmly allocated.
• Upgrade hints before the upgrade is confirmed.
• Arrival times for a transfer you do not directly control.
• Anything about another guest's booking, however innocuous.

Make that list explicitly. An assistant will not infer it, because every fact on it is genuinely true.

What to do on Monday

Take your ten most common chat questions and write the answer you would be happy to see on a screenshot. Not the facts. The answer, in your words, at your length. Then write the ban-list, the identity sentence, the price rules and the do-not-send list on the same page.

That page is worth more to an AI project than a hundred-line FAQ, and it takes an afternoon.

Takeaway

A correct fact in the wrong voice or the wrong format still loses the guest. Your AI needs your facts, your tone and your house rules. Only you can write the last two, and they belong on the same page as the line where the assistant stops selling and calls a person.