Getting Found by ChatGPT: How to Enter the Answers of AI Assistants (Hotels and Tour Operators)

One of your potential guests has just asked ChatGPT “where should I stay in Sorrento with the children, near the sea, under €150”. They got four names back. They did not open Google, did not look at Booking, did not see an advertisement. They read four names and picked one. If yours was not among them, as far as that customer is concerned you do not exist — and you did not notice, because no lost visit shows up in your statistics.

It is the quietest change in tourism of the past two years. There is no drop in traffic to raise the alarm: it is simply that part of the decision is now taken somewhere else, before anyone reaches your website. This guide explains how it actually works, what can be measured and what can be done — without promising shortcuts that do not exist.

What changed: from “searching” to “asking”

For twenty years the route was always the same: the user typed something, Google showed ten links, the user clicked. All the SEO we know exists to get into those ten links.

With a conversational assistant the shape of that route changes. The user describes a situation — not a keyword — and receives one answer, not a list. Inside that answer there are two or three names, five at most. There is no second page.

Where an assistant gets its information

Simplifying, from three places:

1. What it learned during training. General information, often not very recent, about well-known places and properties.
2. What it reads at the time of asking. Almost every assistant, when the question calls for current data, goes and reads web pages in real time. This is where the game is played.
3. What it finds in aggregated sources. Portals, directories, reviews, third-party listings.

Point two is the only one you can act on directly, and it is also the one that weighs most on the concrete questions — prices, availability, amenities, location.

It is not an entirely new phenomenon: it is the same direction taken by voice search, where the user already asks a question and receives a single answer. Conversational assistants have carried that model out of the kitchen and into the moment of the buying decision.

They are not all the same

“Getting found by AI” is a simplification: the assistants behave differently, and it helps to know which one you are trying to reach.

• ChatGPT generates the most volume, and when the question concerns availability or prices it goes and reads the web then and there. It is also the one that, in the logs of the sites we look after, passes by most often.
• Perplexity cites its sources explicitly, with the links on show: it is the assistant most likely to send you an actual visit, because the link is clickable.
• Google AI Overviews is the summary that appears above the search results. It is not a separate assistant: it draws on the same pages Google has already indexed. Here the SEO you have already done continues to count.
• Gemini and Copilot follow logic close to Google’s and Bing’s: indexing and structured data are what matter.

The good news is that the work to be done is largely the same for all of them. The bad news is that there is no lever for “optimising for ChatGPT” separate from everything else.

How much is really happening: the numbers from a real site

On one of the sites we look after — a tour agency in Campania — we measured the reads made by AI assistant crawlers. Over roughly two months: just under 65,000 reads, 78% of them from ChatGPT, spread across some thirty different agents. Before that period, the same measurement returned zero.

The figure that concerns people, on the same site, is a different one: 0.7% of real visits arrive from an AI assistant. That is not much. But two years ago it was zero, and the curve is rising.

This is the honest picture: today the assistants read a great deal and send very little. Anyone who gets organised now does so while the cost is low and the competition is looking elsewhere.

What makes a site readable by an assistant

There is no trick. There are conditions that make a page usable by a machine which has to summarise it in two lines.

1. Answer the question instead of circling it

An assistant looks for sentences that answer on their own. “Check-in is from 2pm to 8pm; after 8pm on request, with a €20 supplement” is quotable. “Our staff will be delighted to welcome you at whatever time suits you best” is not: it contains no information.

The same holds for everything: prices, distances, cancellation policies, pets, parking, opening hours. Every piece of information a customer would ask for on the telephone should exist on the site as a clear, self-contained sentence.

2. Structured data, done properly

Structured data (schema.org) is the standard way of telling a machine “this is a hotel, this is the price, this is the location, this is the star rating”. For an accommodation business you need at least:

• Hotel or LodgingBusiness with address, coordinates, telephone and price range
• FAQPage covering the questions you actually receive
• Product or Offer for rooms and packages, if you sell direct
• Review and AggregateRating, if you have reviews of your own

3. Do not block the crawlers without knowing it

Many sites block the assistants’ bots in their robots.txt file, sometimes because somebody copied a configuration found online, sometimes because a security plugin does it automatically. The user agents worth checking today are at least GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and Applebot-Extended.

Blocking them is a legitimate choice — but it has to be a choice, not an accident. If you block them, you accept not appearing in the answers.

4. Content somebody would quote

Assistants favour sources that answer specific questions precisely. For a property in Campania, pages like “How to get to Sorrento from Naples airport” or “Where to park in Positano” are worth more than an “About us” page: they are exactly what somebody asks an assistant, and you can answer better than a general portal because you live there.

It is the same material local SEO needs, and the same that feeds an assistant on your own site: written once, it works on three fronts.

5. Be consistent everywhere

If your site says check-in is at 2pm, your Google Business Profile says 3pm and Booking says 4pm, an assistant has no way of knowing whom to believe — and it tends not to cite a source that contradicts itself. Address, telephone, opening hours and amenities have to match across every listing.

6. Have reviews, and have them visible

When an assistant has to choose between two equivalent properties, it looks at public judgement: average score, number of reviews, what they say. It is not a factor you can touch by writing code — it is built up over time by managing your reviews. But it is worth knowing that it counts here too, not only on Booking.

How to check whether they name you

There is no mature tool yet, nothing like Search Console. In the meantime, two methods that work:

The manual test. Put the questions one of your customers would ask to three or four different assistants. Not “hotels in Sorrento”, but “I’m going to Sorrento in June with two small children, looking for somewhere near the sea without spending too much”. Write down who gets named. Repeat every month: it is a crude measure, but it is a measure.

The server logs. Crawler reads leave a trace. Filtering the user agents shows which assistants come by, how often and which pages they read. It is the most solid data available today, and it is where the figures quoted above come from.

What does not work

A certain amount of nonsense is already circulating. For the avoidance of doubt:

• You cannot buy a position inside ChatGPT’s answers the way you buy an advertisement on Google. Anyone offering you that is selling something else.
• Stuffing your pages with “AI”, “artificial intelligence” and the like does nothing. It is the 2026 edition of keyword stuffing, and it works about as well.
• Mass-generated text published without review makes things worse. Assistants are trained to recognise generic content; a page that adds nothing does not get cited, and meanwhile it dilutes your site.
• It is not a substitute for SEO. Google remains by far the first channel. This is one more channel, to be cultivated while the other keeps working.

Where to start, in order

If you have to choose what to do first:

1. Check your robots.txt. Ten minutes. If you are blocking the assistants’ crawlers without meaning to, nothing else you do will have any effect.
2. Put structured data on the home page and on the room or tour pages. That is the foundation.
3. Rewrite the “service” pages so that they contain real answers: hours, prices, distances, rules. No courtesy phrases standing in for information.
4. Align the external listings — Google Business Profile, Booking, TripAdvisor — with what the site says.
5. Write three pages answering concrete local questions you get on the telephone.
6. Measure, and measure again in three months.

The questions we are asked most

Should I block the AI crawlers to protect my content?

It depends what you sell. If you live on original content — guides, itineraries, photography — blocking them is a defensible position. If you sell rooms or tours, the content on your site exists to get you chosen: blocking whoever reads it means giving up on appearing. For an accommodation business, in the great majority of cases it is better to let them in.

How long before I see an effect?

Crawlers come back in days or weeks, depending on how often the site is updated. Being cited in the answers is slower, because it also depends on external sources and on the competition for that particular question. Reasonably: a few weeks to be re-read, a few months to measure a difference.

Do I need a plugin or a paid tool?

No. Structured data, a correct robots.txt and well-written pages are done with what you already have. There are AI citation monitoring tools, useful if you manage many sites; for a single property, the monthly manual test and the server logs are enough.

I’m on Booking — don’t they find me already?

They find you, but as a line in somebody else’s list. Assistants readily cite the big portals, and in that answer the name that sticks is the portal’s: the booking goes through them, with the commission that follows. Appearing in your own right is exactly the work that increases direct bookings.

Does AI-written content help you get found by AI?

No, and it usually does damage. An assistant is looking for information it cannot find elsewhere: opening hours, prices, real distances, the advice of somebody who lives in the place. Generic text generated in bulk contains none of that, and meanwhile it lowers the average quality of your site.

In short

AI assistants are already reading your site — or skipping it. Today they bring a small share of visits, and anyone telling you otherwise is overstating it. But it is a share that did not exist two years ago, and the conditions for being cited are built now, with work that is worth doing anyway: clear information, tidy data, consistency across channels.

It is the same work that makes a site useful to a person. The difference is that somebody else is now reading it first.