How to advertise on ChatGPT: a step-by-step guide for hotels

To advertise on ChatGPT you open an account in Ads Manager at ads.openai.com, add a payment method, build a campaign with one or more ad groups, and wait for review. A hotel can do it on its own. The work before it, on pages and measurement, decides whether the test tells you anything.

Any business whose legal entity is based in a listed country can buy, and the list has 89 countries, Italy and Switzerland among them. The rule looks at the company that advertises and is billed, not at where the hotel stands. An agency should be invited into the hotel’s account as a user, so that the hotel keeps the account and its history.

3 things surprise hotels here.

That second point matters more for a hotel than for a shop. A hotel on Lake Garda sells to Germans and Americans, and its account may at first reach only Italy. OpenAI explains the rule on its campaigns page. If your guests come from abroad, check it before you plan around them; the European details are in ChatGPT ads in Europe.

3 things have to be ready: a page that answers each situation, access for OpenAI’s crawlers and a way to count bookings. OpenAI says landing pages must not block OAI-AdsBot and OAI-SearchBot. A security rule on your site can block them without anyone noticing, and the ad can then be rejected, so ask whoever runs your site to check.

The page matters twice. The ad system reads it to decide whether the ad fits, and on a phone it is what the guest sees next. Campaigns published so far run mostly on mobile: 92.3% of impressions at Hotel Klimczok in Poland, around 80% in a UK test and about 88% of spend in the apps across 3 clients of an Italian agency. Open your landing page on a phone and count the screens before the answer.

OpenAI itself advises against defaulting to a generic homepage.

For bookings, create a data source and send events by Pixel, by the Conversions API or both. No booking event exists for hotels: a completed booking maps to order_created, the only event that feeds sales figures in reports. The click reference, oppref, has to travel with the guest to your booking engine, and an engine on another domain can lose it, so test the path with a real booking. Where consent is required, set the Pixel’s consent to false before it starts and to true after the guest agrees.

Build one campaign with 3 to 5 ad groups, one need and one landing page each. The structure has 3 levels: a campaign holds the objective, budget, dates and countries; an ad group holds a need, a bid and the hints; an ad holds the title, text, image and link.

OpenAI asks for many distinct ads per offering, and at 15 euros a day each extra ad shares the same impressions. In the Slovak hotel’s test one ad of 4 took almost two-thirds of impressions, so its agency’s lesson is to give variants their own group and budget if you want to compare them. A Liechtenstein agency saw the same thing by ad group, with one of 5 groups getting about 3% of impressions.

Set the countries explicitly. Excluding others does not restrict delivery to the one you want, and accepting every available location buys conversations with people who will never come to you. Outside feed campaigns you can also target regions, cities, markets and postal codes where available.

Each campaign also has a trap on dates: on your very first campaign you may not see start and end dates before publishing, so do not press publish just to reach them.

Write each hint as one natural phrase on one idea, saying what you offer, who it helps or when it is useful, and add something the ad and the page do not already say. That is OpenAI’s guidance. An ad group can hold up to 2,000 hints, and OpenAI adds that you do not need to fill the limit. I would start with 5 to 10.

Style is still argued over. The Slovak hotel’s agency says short phrases worked better than sentences, while OpenAI asks for natural phrases, and nobody has published a controlled test on bookings. 2 findings are worth keeping. An Italian travel campaign with 11 hints got its best results on a hint about people deciding whom to book with: 1.52% click-through at €1.65 and about 45% of the spend, against 0.48% for people comparing dreams. And in the same test a group with no hints clicked less and cost a third more than groups with them. Those rates come from one test on one account, so read them against each other only.

Bulk edits replace the whole list of hints, so keep your list in a document. Examples for 8 situations are in ChatGPT ads examples.

Decide this before launch, because since 30 October 2026 they are on by default. OpenAI may rewrite your title and text, translate them, resize images and send the guest to another page of your site, and there is no preview of each version. You stay responsible for the result under its terms.

For an independent hotel with accurate pages, leaving them on may be reasonable: it is one way to reach a guest in another language without writing the ad yourself. A hotel with a brand book, a chain or a property with rates that change often should switch them off, or at least read the landing page line by line first, because that page is what the rewriting draws on. The setting is under Manage settings, and a campaign setting beats the account default. With them off, an English ad stays in English, so write each language.

Your ads go through review, then the auction decides how much they run, and for the first days you should mostly wait. OpenAI checks each ad against its policies before it goes live, and approval, billing, status, budget and schedule all have to allow delivery.

Then change one thing at a time. OpenAI’s reports refresh clicks about every 15 minutes, and a hotel will be tempted to touch something every morning. There is no report of the conversations behind your ad, so your 2 instruments are the booking engine and your patience. The guide to ChatGPT ads for hotels sets out what to measure and when to stop, how much ChatGPT ads cost gives the sums, and our study of 3,288 answers shows what the same pages do for you without any ad.