What a month of work can and cannot do
OpenAI has not published how ChatGPT decides which hotels to name in an answer, and its answers change from one run to the next. So nobody, including us, can promise that your hotel will be recommended by a given date. Anyone who sells a guaranteed ChatGPT placement is selling something they do not control.
What you can control is the material AI works from: whether it can read your website, whether the facts on it are easy to quote, whether your listings and guest reviews tell the same story, and whether any page answers the questions guests actually ask. This plan spends one week on each of those, in an order that puts measurement first. It is written so a GM can hand it to a marketing manager and check progress at the end of each week.
A note on terms. LLM stands for large language model, the kind of AI behind ChatGPT, Gemini and Claude. GEO, generative engine optimization, is the industry name for the work in weeks 2 to 4: making a business easier for those models to find and cite. Schema is a block of code on your website that states facts about the hotel in a format machines read without guessing.
Week 1: measure where you stand
Start with a baseline, because without one you will not know whether the next three weeks changed anything. The fastest way is the free Staylight scan: one Radar scan per account, 15 real guest questions about your city and type of hotel, checked on ChatGPT, Google AI Overviews and Google AI Mode, with no card required. It shows which questions name you, which do not, and who gets named instead.
If you would rather do it by hand first, write down five questions a guest would ask before they know your name, such as "best hotel in Prague for a honeymoon" or "hotels in Prague city centre". Ask each one in ChatGPT and at least one other assistant. Record whether you are named, where in the list, and which hotels or websites take the slots you wanted. Keep the exact wording, because you will ask the same questions again in week 5.
The most useful part of the baseline is the split between questions you win and questions you lose. That split decides what week 4 is about, so do not skip it in favour of a single overall percentage.
What the baseline usually shows
In our own scans, the questions a hotel misses follow a pattern. Golden Well Hotel, a boutique hotel in Prague, was named in 21 of 295 non-branded questions, or 7.1%. It came up for occasion questions like "best hotel in Prague for a honeymoon" and was missing from broad ones like "good hotels to stay at in Prague". Mosaic House Design Hotel, also in Prague, was named in 25 of 200, or 12.5%. It was named for identity questions like "unique hotels to stay in Prague" and missed generic location questions like "hotels in Prague city centre".
The engines also disagreed. Golden Well was named in 5 of 60 ChatGPT answers, 9 of 60 on Gemini, 7 of 60 on Perplexity, and none on Claude or Google AI Overviews. Mosaic House had 8 of 40 on ChatGPT and 11 of 40 on Gemini. A check on ChatGPT alone would have given each hotel a different picture from the one its guests actually see across assistants.
When these hotels were left out, the sources AI cited most were Booking.com and TripAdvisor. For Golden Well, AI also cited a French travel blog and thehotelguru.com. That list tells you where AI is reading about your market, which matters in week 3.
Week 2: make your website readable and quotable
Week 2 belongs to whoever looks after the website. The first check takes five minutes. Open yourdomain.com/robots.txt and confirm it does not block OAI-SearchBot, the crawler OpenAI uses for ChatGPT search. OpenAI's own documentation says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers, and that a robots.txt change can take about 24 hours to take effect. GPTBot, which collects content for training models, is a separate setting, so a hotel can allow one and block the other. Check for broad wildcard rules too, since a blanket block catches every bot at once.
Next, add Hotel schema to the homepage: the hotel's name, address, phone, check-in and check-out times, star rating and main amenities. If you have an FAQ page, mark it up as well. Our guide to AI crawler readiness walks through both, and the schema generator on this site builds the code for you.
Then fix the copy, so the page states facts an assistant could repeat. A made-up example: "40 rooms, a 10-minute walk from the main station, parking on site, airport transfer on request" gives an assistant something to repeat. "An oasis of calm in the heart of the city" gives it nothing. Put rooms, distances, parking, breakfast, pets and cancellation terms in text on the page, not only inside images or a booking widget that loads after the page.
Keep this week in proportion. Golden Well had the highest readiness score of the four hotels we published, 77.1%, and was named less often than Mosaic House, which scored 60.9%. A readable site removes a reason to be skipped. It does not make AI choose you on its own.
Week 3: make your listings and reviews agree
AI assistants read other sites about your hotel as well as your own. In our Prague scans they cited Booking.com and TripAdvisor most. If your name, address or phone number differs between your website, Google Business Profile, Tripadvisor and Booking.com, an assistant can pick up two versions of the same hotel. Keeping them identical is what the industry calls NAP consistency, for name, address and phone.
Hand this week to front office or revenue. Make one reference sheet with the hotel's official name, address, phone, room count and main amenities, then go through each listing and correct it to match. Start with Google Business Profile, since Gemini, Google AI Overviews and Google AI Mode draw on Google's own data, then Tripadvisor and the OTAs your guests use.
Then reviews. Reply to recent reviews on Google and Tripadvisor, and reply with specifics: the room, the breakfast, the walk to the old town. Google's own help pages explain how owners read and reply to reviews on a Business Profile. Recent reviews that name real things give an assistant current, third-party detail to work with, which our article on reviews covers in more depth.
Week 4: answer the questions you are missing
Go back to the week 1 baseline and list the questions where you were not named. For Golden Well, that meant broad questions about where to stay in Prague. For Mosaic House, it meant questions about location, such as staying in the city centre. Both hotels were already named where their identity was obvious. Our reading is that the missing questions were the ones where nothing on the hotel's own site tied it to the question.
For each missed question, write or improve one page that answers it directly in the first paragraph. If you miss "hotels in the city centre", say plainly how far you are from the main square, the station and the main sights, and in which district. If you miss family questions, state the room sizes, cots, connecting rooms and what children eat at breakfast. Write it in the guest's words and in the languages your guests search in.
Keep each page factual and current, with the date you last checked prices or opening hours. Do not create pages that claim things the hotel cannot deliver, because guests who arrive on that promise will say so in their reviews.
Re-check, then keep going
At the start of week 5, ask the same questions on the same engines and compare the result with week 1. Use identical wording. A new set of questions would measure something different, and because AI answers change from run to run, one answer moving in your favour proves little. Look for a change across several questions and engines.
Expect some engines to move before others, and some not to move at all inside a month. If a blocked crawler was the problem, the fix can show up sooner than anything else in this plan. Expect listing corrections and new reviews to take longer to show. After that, a monthly check on the same questions tells you whether the work is adding up, and which missed questions to take on next.