Your Guest's AI Will Message You Before They Do

You are on the phone with a frequent guest. She is halfway through asking for a late checkout when she says: actually, let me just have my assistant sort this out.

Two seconds later, a perfectly formatted WhatsApp message arrives. Late checkout request. Her usual room preference. A question about parking. A request to be notified if a spa slot opens between five and seven.

Polite. Precise. Not from her.

This is already half true

Travellers already use assistants to plan trips, compare rates, check whether a cancellation policy is flexible enough for a changeable flight, and handle routine requests - ask the hotel if I can check in early and whether they have vegan breakfast.

Today that usually ends with the assistant drafting a message for the guest to send. The next step is the assistant sending it directly, reading the reply, and acting on it. That step is small, it is being built by every major assistant vendor, and it changes what a hotel's messaging has to be.

Guest messaging stops being an engagement channel and becomes an interface between two automated systems.

What changes when the other side is a machine

A human guest forgives a slow reply. Tolerates a slightly inconsistent answer. Rephrases when the first attempt is misunderstood.

An agent does none of that. It sends a precise request and expects a precise answer. If the reply is slow, it notes the latency. If the answer contradicts what your website says, it notes the inconsistency. If the reply is a menu of buttons, it cannot proceed at all.

And then it decides. Quietly, without the guest seeing the exchange, it may recommend a different property next time - one whose messaging answered faster and matched its own website.

That is the shift. Your messaging is now being scored, and the scorer does not tell you the score.

Why the last generation of chatbots fails here completely

Most hotel messaging flows were built for a human clicking. Predefined buttons, scripted decision trees, flows that already break when a guest free-types something unexpected.

An agent does not click. It writes a structured request in natural language - early check-in on the 14th for a reservation under this name; confirm parking availability and the pet policy - and expects a structured answer.

A decision tree cannot parse that message. It matches one keyword and shows a menu. The agent cannot select from the menu. The exchange ends, and the agent records that this property could not answer.

A language-model assistant grounded in the property's facts handles the same message in one reply: early check-in available from 11am for a fee; parking on-site and free; pets welcome in ground-floor rooms. That is the difference between being bookable by an agent and being invisible to one.

The two things an agent will test

Precision. Is the answer specific, consistent with your other sources, and complete? An agent cross-references. If your site says check-in is 3pm and your messaging says 2pm, it notices, and it trusts neither.

Honesty when unsure. When the assistant does not know - a question outside the knowledge base, a request needing a human decision - does it say so and hand over, or does it improvise? An agent will act on an improvised answer, and the consequence lands on the guest at the desk.

Both are the same things a good messaging assistant needs for human guests. The agent simply tests them harder and more often.

What to do now

Nothing exotic. Two things that are already worth doing.

One clear, current, consistent source of every answerable fact about the property - hours, policies, prices, amenities - structured so machines read it and kept the same everywhere it appears. This is what an AI search engine needs to recommend you and what an agent needs to transact with you.

A messaging assistant that answers precisely from that source and escalates honestly when it cannot. Not a tree. Something that understands a structured request in any language and either answers it fully or routes it to a person with context.

Do those, and you are ready for the guest who arrives with an assistant. Skip them, and the first you will know is a slow decline in bookings from the guests who used to be easiest to serve.