How AI assistants choose which hotels to recommend

The short version

Nobody outside the AI companies knows the exact recipe, and it changes. Be wary of anyone who says they do.
An assistant that can search the web builds its answer from pages it finds: your website, booking and review sites, maps and open data, tourism boards and the press.
You can't control all of that. You can control the facts on your own pages, whether they agree with each other, and whether your hotel is described the same way elsewhere.
You can see part of the picture yourself in ten minutes. The test is at the end.

What happens when a guest asks

A guest types something like "Which hotel near the cathedral has parking and an early breakfast?" Depending on the assistant and the question, it may search the web, read some of the pages it finds, and write an answer that names a few hotels and often shows the pages it used. Sometimes it answers from what it learned in training, without searching at all.

That means two things for a hotel. What the assistant found and read this time matters, and so does what it already "knows" about you. Both come from public information about your hotel.

Where the information comes from

Your own website. The best place to state facts clearly: parking and its price, breakfast times, check-in and out, cancellation, how to get there.
Booking sites. They publish rates, policies and photos for your hotel. If your own pages don't say something, an assistant may quote the booking site instead.
Review sites. What guests say, and whether the hotel replies.
Maps and open data. Map listings and open databases such as OpenStreetMap and Wikidata can describe where you are and what you are.
Tourism boards and city guides. "Where to stay" pages and local listings.
Press and travel articles. Lists and features that name hotels, which may name your competitors and not you.
Forums. Threads where travellers ask where to stay, such as Reddit and Quora.

Why the same hotel can be named one day and missed the next

Answers vary from one run to the next, even for the same question. A single answer proves very little. That's why anything useful should ask each question more than once and show a range, not a single score, and why we say plainly that a sample from one assistant on one day isn't a ranking.

What a hotel can influence

Facts on your own pages. If a guest's likely question isn't answered, the assistant has nothing of yours to quote.
Agreement. Two check-in times, two phone numbers or an old name on a listing give an assistant a reason to be wrong, or to leave you out.
Your listings. Booking sites, maps and directories should say what your website says.
Your reviews. Replying, specifically and kindly, to critical reviews shows how you treat guests.
Credible mentions. Being included, honestly, in local guides and articles. Not paid placement dressed up as editorial, and never fake or incentivised reviews.

What is not proven

That adding structured data (schema markup) makes an assistant recommend you. It can help accuracy. We have seen no good evidence that it makes an assistant recommend a hotel.
That an llms.txt file helps. It's an informal proposal, not a standard, so not having one is never a fault.
Any exact "ranking formula". Treat confident claims about one with suspicion, including ours if we ever made one.

Try it yourself in ten minutes

Open an AI assistant that can search the web. Use the same one each time.
Ask five ordinary questions a guest might ask about your area, without your hotel's name. For example: "Best 4-star hotel near [landmark] with free parking?" or "Where should I stay in [town] for a weekend with a dog?"
For each, write down whether your hotel was named, where in the list, and which hotels were named instead.
Open the pages the assistant cited. Which are booking sites, which are guides, and which, if any, are yours?
Ask each question again later, or in a new chat. Note what changes.
Then ask about your hotel by name and check what it says against the facts. Is the phone number right? The check-in time? The parking?

You've now seen who gets named instead of you, and where the assistant went to find out. That is most of what a first look should tell you.

Where Presence Pro fits

We do that test properly: more questions, each asked more than once, with the sources checked and the evidence kept. Then we turn it into a short plan. Get a free one-page snapshot to see it for your hotel, or read the checklist to do the basics yourself.

Common questions

Is this the same as SEO?

They overlap: clear, accurate, well-organised pages help both. But assistants also quote booking sites, reviews and press, and their answers vary from run to run, so the work is wider than ranking one page.

Should I block AI crawlers?

That's your decision. If you block them, an assistant may lean on other sources for facts about you. Check what your robots.txt says, so that what it does is what you meant.

Can I pay to be recommended?

We don't know of any reliable way, and we'd be cautious about anyone selling one.

This guide describes how things appear to work from public information and our own testing. The assistants change often. Statistics about how many travellers use AI are on our numbers page, with sources.