Your hotel website stopped being how she finds you. It became how she checks you.

The finding everyone is quoting is true. When an AI assistant recommends a hotel, it almost never cites that hotel's own website — under 10% of citations, across 695 unbranded searches run through ChatGPT and Google's AI Mode by Tharro, published June 23. Tharro sells visibility tracking, and it is one month's snapshot. Your site is not how you get into the answer.

The conclusion drawn from it is wrong. The finding describes how the machine learns which hotels to name. It says nothing about the next ten minutes, when she opens your website to find out whether any of it was true.

She goes because she does not trust what she was told. Phocuswright's research across the UK, France and Germany puts trust in AI-generated results at roughly 17 to 20 percent while usage climbs, and we reported in June what follows: 58% go straight to the property's own site, and the travelers who trust AI most verify at 62% against 63% for the neutral ones.

She is coming. The question is what she finds.

She arrives with a total, not a rate

She has just come from an OTA listing, and what she read there was not a nightly rate. It was a number for her whole stay with the taxes and the fees already in it, on one screen, without her asking twice. When Skift put Meta's new AI agent on a New York hotel search on September 9, the figure it pulled off Expedia was $543 a night and $627 with taxes and fees — the all-in number, first screen, nothing clicked. That is the number in her head when she reaches you.

Now look at what your site does with it. It shows a rate from, and wants her dates before it shows anything real. Tax arrives at the third step, the fee at the fourth, and the figure she could actually hold against the OTA tab still open behind your window takes four clicks and a form.

She concludes you are more expensive. She is not wrong about what she saw. Nothing leaked, nobody undercut you, your rate was fine all along. The OTA page answers before it asks. Your booking engine asks before it answers, and she is not booking yet. She is checking.

How you meet a total you were never given

The objection is fair: a booking engine needs dates, and she did not bring any.

Answer the question anyway, approximately, before she has done any work.

Publish what a stay actually costs. Not a teaser rate, a worked total. Three nights in a double in October, from this much, VAT and city tax included. A short table by month and length of stay does more for a verifying traveler than any amount of photography, and almost no independent has one.

Put the taxes and fees in the first number you ever show her, everywhere you show it. If the OTA quotes all-in and you quote nightly-plus-extras, you have handed away a comparison you were winning on the real figures.

Name the fees in words on the room page. Resort fee, city tax, parking, breakfast in or out. She will find them eventually, and finding them at step four of a booking form is what makes a person close the tab.

And answer the parity question before she tests it, because she will — the OTA page is still open. A plain sentence saying this is the lowest available rate for this room, and asking her to say so if she finds otherwise, is checkable in fifteen seconds. That is the point. She is in a checking mood.

What buys the rest of her effort

Entering dates into a booking engine is work, done on speculation. She does it only if the page has already suggested the answer is worth having.

So clear the list she brought. The assistant made claims about you: the distance to the station, the pool, breakfast, parking, the view. Those claims, answered plainly on a page she does not have to hunt through, are worth more than any brand copy on the site.

Show the room at the price. If your cheapest rate is a courtyard double, photograph the courtyard double and call it what the OTA calls it, because she is cross-referencing two pages and a mismatch reads as a trick rather than a naming convention.

Put the cancellation terms in a sentence.

And give her somebody. Phocuswright's reading of its own data is that human sources win not because they know more but because they can be held to account. A person who recommends a hotel can be asked about it afterwards. An algorithm cannot, and neither, in practice, can an OTA call centre.

Which makes the chat window on your site the right instrument, filled by most hotels with the wrong thing. She came to your page to check what a machine told her. Putting a second machine there to answer her hands her the same problem in your brand colours. What she needs is someone at the property, on the chat or the phone, who knows the pool is closed in October because she walked past it this morning. That is a verification device.

The habit is a human one

An agent does not verify. Meta's compared properties on location, subway access, review scores, refundability, bed type and fees, chose one, and never opened a hotel's website. It had no reason to. The doubt that sends a person to your page is doubt a machine does not carry.

The window is open because travelers are still checking for themselves. Use it while it is.

Run her check before she does

Open an OTA listing for your own hotel and write down the first number a traveler sees there. Then open your own site and count what stands between her and the same figure. That gap is the part you can close this week, and it costs nothing but a decision about what your first screen is for.

Then the harder half. Ask an AI assistant for hotels like yours in your town and write down everything it says about your property, the true and the wrong together. Take that list to your own pages and time each line. Whatever you cannot settle on one screen, she will settle somewhere else, on a page that sells the room.

Something she only half believes told her to consider you. She came to your website to decide, and she brought her card.