Which sources does ChatGPT use about your hotel? How to find out

When ChatGPT answers a guest's question with a web search, it builds the answer from a few web pages and lists them as sources. The same goes for Perplexity, Gemini, Claude, Google AI Overviews and Google AI Mode. These tools run on large language models (LLMs), software trained on huge amounts of text that writes an answer by predicting what a good one looks like. When an answer needs current facts, such as which hotels are open or what a room costs, the tool searches the web first and cites what it read.

So the practical way to find out which sources ChatGPT uses about your hotel is to ask it the questions your guests ask, without naming your hotel, and write down every link it gives. Do the same in two or three other engines. Before long the same domains come back, and that short list is worth more than any general advice about AI.

Expect the list to be dominated by booking platforms and review sites. In the hotels we have scanned, the hotel's own website was rarely the cited source, even on questions where the hotel was a good answer.

ChatGPT has shown the sources for a search answer under a Sources button at the foot of the response, which opens a side panel of links. Search Engine Roundtable reported in August 2026 that the option sometimes shows up in the three-dot More actions menu instead, as View sources. A September update said the free version now shows Sources by default. If you cannot see the button, check that menu.

Perplexity puts numbered citations in every answer. Its help centre says each answer includes numbered citations linking to the original sources, so you can check a claim against the page it came from.

Google AI Overviews and Google AI Mode show links to supporting websites alongside the generated answer. Google's documentation for site owners says both features may run several related searches across subtopics while they write, and can show a wider set of links than a classic results page. A page has to be indexed and eligible to appear in Google Search with a snippet to be shown as one of those links.

Gemini shows a Sources button at the bottom of a response, and selecting it opens a side panel with the links. Google's help page adds that if there is no Sources button, Gemini did not provide links for that response.

Claude attaches citations when it uses web search, so you can check the original pages. Anthropic made web search available on all Claude plans in May 2025.

Two things follow for a hotel. First, an answer without visible sources tells you nothing about where the engine got its view of you, so run the question again or use an engine that shows links. Second, a link title tells you little about what the page says, so open the pages themselves before you decide a source matters.

When AI engines answered Prague hotel questions without naming Golden Well Hotel, a boutique hotel, they cited Booking.com and TripAdvisor most often. A French-language travel blog, prague-secrete.fr, and the hotel editorial site thehotelguru.com came next. Golden Well was named in 21 of 295 non-branded questions, 7.1%, and mostly for occasion questions such as "best hotel in Prague for a honeymoon".

For Mosaic House Design Hotel, also in Prague, the cited sources were Booking.com, TripAdvisor, Expedia and thehotelguru.com. Mosaic House was named in 25 of 200 questions, 12.5%. It appeared for identity questions such as "unique hotels to stay in Prague" and was missing from location questions like "hotels in Prague city centre".

Outside Prague the mix changed. For Olympia Hotel in Mariánské Lázně, a spa hotel with 0% detection across 299 checks on five engines, AI cited Booking.com and local tourism directories such as marianskelazne.cz, and a competing spa hotel brand's own website was cited by name in answers where Olympia could have appeared. For Bouda Mama, a family mountain hotel in Pec pod Sněžkou, AI mostly cited other local hotels, regional tourism sites, the town's own tourism site and TripAdvisor, and leaned less on the big international OTAs than the Prague answers did.

These are four real hotels, not a representative sample, and each scan ran at one point in time. The pattern is still useful: in a big city you compete with platforms for the citation, and in a small destination you may be competing with the hotel down the road.

An OTA page answers a broad question for hundreds of hotels at once, changes every day and is linked from all over the web. Your website describes one hotel and changes a few times a year. For "good hotels to stay at in Prague" the engine picks the source most likely to cover the whole question, and that is usually the platform. Our article on why AI recommends OTAs covers this in more detail.

Some of it is technical and can be checked in minutes. OpenAI says sites that opt out of its search crawler, OAI-SearchBot, will not be shown in ChatGPT search answers, and a hotel can block it by accident through a security plugin or an old robots.txt rule. Google only links pages that are indexed and eligible for a snippet. Facts that only appear inside images or long marketing paragraphs are also harder for an engine to quote.

Fixing the technical side does not settle it, though. Golden Well had the highest readiness score of the four hotels, 77.1%, and was detected in 7.1% of questions. Mosaic House scored 60.9% and was detected in 12.5%. Readiness did not predict visibility in this set. One likely reason is that the engines in our scans mostly cited independent sites, and a hotel describing itself is a single voice next to several independent pages.

Start with the sources you already control. Make sure your Booking.com, TripAdvisor, Expedia and Google Business Profile listings carry the same name, address, phone number, room count and main amenities as your own site. Engines can read all of these records, and if they disagree, an engine may pick up the wrong details. Our guide to NAP consistency and citation authority walks through that check. NAP stands for name, address and phone number.

Then keep your review profiles current and reply to guest reviews. Review sites are among the most cited sources in our scans, and a review that names a feature gives an engine something concrete to repeat.

Next, sort the domains you collected into three groups: your own site, booking platforms and other hotels, and everything else. The third group is your outreach list. Local guides, city tourism sites, regional press and editorial sites like thehotelguru.com are places a hotel can realistically get listed or written about. Booking.com does not need persuading, and a competitor's website will not mention you.

Last, give engines a reason to cite your own site on narrow questions. Write plain answers to what guests ask before booking: distance to the station, whether breakfast is included, what the family rooms sleep. On those questions a booking platform's generic page is a weaker source than yours. None of this guarantees that an engine will cite you or name your hotel. Nobody can promise that, and anyone who does is guessing.

AI answers vary between runs, and so do the sources behind them. The same question asked twice in a day can cite different pages. A single check tells you what one answer looked like at one moment, which is a weak basis for deciding where to spend a marketing budget.

What works is a fixed set of questions in the languages your guests use, asked on the same engines on a schedule, with every cited domain recorded each time. Over a few weeks you see which sources are steady and which come and go, and whether your own site starts to appear after you change something. Outreach is slow because it depends on other people publishing, so measure in weeks and months.

Staylight's free scan shows where AI gets its information about your hotel: which questions named you, and which sites the engines cited, sorted into your own site, booking platforms and editorial or other sources. It needs no card. The paid plans repeat the scan on a schedule, so you can watch that list change.