ChatGPT ads for hotels: how they work, what they cost and when to skip them

ChatGPT ads are sponsored cards that appear below an answer, each with the advertiser’s name and logo, a title, a line of text, an image and a link to a page of your site, as OpenAI describes them. A hotel can buy one itself in Ads Manager, OpenAI’s self-service tool, from EUR 15 a day. Testing began in the United States on 9 February 2026.

Paying does not move your hotel into the answer. OpenAI says ads do not influence what ChatGPT writes: the card sits underneath and is labelled as sponsored. So there are 2 different jobs. The answer is earned by your pages and your reputation, which the guide on how to get your hotel on ChatGPT covers. The card under the answer is bought.

Only part of the audience can see the card. Ads appear on the Free and Go plans, never on Plus, Pro, Business, Enterprise or Edu, and never on accounts identified as under 18. A Free user can also switch to an Ads-Free mode with fewer messages. No public measure says how many of your guests sit on those 2 plans, and for a 5-star hotel that is the first honest doubt.

The booking sites got there first, and they buy under questions that do not mention them. GrowByData, which sells this kind of tracking, read 2,248 logged-out US searches on 150 travel questions between 12 September and 11 October 2026 and found 423 sponsored placements from 117 advertisers.

Trackers disagree, so read any one of them with care. In that sample Booking.com had 0 placements, while SE Ranking gave Booking.com 62.57% of its travel ads on 23 July, in a US study of 20 sectors. Different questions on different days give a different leader. What does not change is the shape: the big sellers are there, and they said so themselves. Expedia told investors in August it was an early adopter of ChatGPT’s latest ad product. Booking Holdings said on its August call that it is in OpenAI’s test group for cost-per-click buying. Radisson has been running campaigns since July, and independents have started too, among them Galgorm Group through Net Affinity and Smarthotel.

One small example shows why this matters to a hotel. GrowByData ran the question “which hotel brands offer the best pet-friendly accommodations?” 6 times: 3 runs carried an ad, 2 of them Expedia’s, and Expedia was never among the brands the answer recommended. The answer can name you and the card below can still belong to someone else.

An ad is matched to what the guest is trying to do, read from the whole conversation, and there are no keywords to buy. OpenAI described the system on 21 October 2026, and it runs in this order.

The signals are the conversation, your landing page, your title and text, your context hints and, where personalisation is on, a person’s wider use of ChatGPT. In the European Economic Area and Switzerland personalised ads are not yet available, so the conversation does the work there.

Context hints are phrases you add to an ad group to say what you offer, who it helps and when. OpenAI is clear that they cannot enforce geography, schedules or exclusions, and a guest does not need to use your words for the need to match. The full mechanism is in how ChatGPT ads work.

3 things follow for a hotel. Your page is read by the system, so a vague page makes a vague ad. A low count of impressions can mean the ad arrived late, with nothing wrong in your bid. And you will never see which questions showed your ad: reports split by country, device, platform and product, and a Liechtenstein agency that spent about €894 on its own ads closed its test saying exactly that.

ChatGPT names hotels well and leaves the price open, so the useful ad supplies the fact the answer is missing. In a panel of 80 guest questions asked on 31 October 2026, 40 in English and 40 in Italian, none of the answers gave a room rate, and most ended by asking for dates or a budget.

The sources tell the same story. Of the 280 links in those answers, 202 went to hotels’ own sites, 52 to tourist boards and event organisers, 25 to guides and magazines, and 1 to a booking site. On 60 luxury questions on 30 October, also through the API, 74 of 148 sources were hotel sites. The same page does 2 jobs: the answer cites it and the ad system reads it. Being named at all is the harder part: in our study of 3,288 answers from ChatGPT, Claude and Gemini, from August 2026, when the question started from the trip and not from the hotel, only 189 of 1,380 answers named one of the world’s 50 best hotels.

That gap is where the card comes in. The guest has just been told “tell me your dates and budget”, and nothing on the screen says what the room costs, whether the dog pays, or which room looks at the lake. So far this is reasoning. Nobody has tested whether an ad written this way brings more bookings.

Write the guest’s situation in the title and one concrete fact in the text, and leave out everything that could describe any hotel. OpenAI asks for clear, specific copy with a different angle in each ad. The limits are 50 characters for a title and 100 for the text, with 16 to 24 and 32 to 48 recommended; on OpenAI’s ad previews at 390 pixels wide, in October, titles were cut after about 33 characters and text after about 62.

These examples are invented, for hotels that do not exist.

Each stronger ad would pass a simple test: you could say it aloud to a friend, and it already gives something. More of them, with hints and the page each should land on, are in ChatGPT ads examples.

One setting needs a decision before launch. Since 30 October 2026 creative enhancements, OpenAI’s automatic rewriting, are on by default, for new and existing campaigns: OpenAI may rewrite your title and text, translate them and change the destination page, with no preview of each version. Under its terms you remain responsible for what runs. A chain with a brand book should choose before the first euro is spent, in Manage settings. With enhancements off, write each language yourself.

A campaign on a daily budget costs at least EUR 15, USD 25, GBP 15 or CHF 20 a day, and you pay by the click or per 1,000 views, never by the booking. Those are OpenAI’s minimum daily budgets; a campaign-total budget is the other option. A daily budget is an average over 7 days: a single day can reach twice the amount, a week never more than 7 times it.

OpenAI suggests a maximum bid of $3 to $5 a click as a starting point. What people have actually paid varies far more than that, and only a few figures come from hotels or from Europe.

What decides the matter is the cost of a booking, which is the price of a click divided by the share of clicks that book. If 1 click in 100 books, a €2.55 click means €255 a booking. If a €600 stay would otherwise pay an assumed 18% commission, that booking is worth €108 to you, and the sums add up only at a click of €1.08 or less. Put in your own rate, your own commission and your own conversion. The full table is in how much ChatGPT ads cost.

No. On 31 October 2026 no hotel has published spend and bookings together at a scale anyone could plan on. The public record is 2 tiny tests and a case study without counts.

The large sellers describe a small channel. Booking Holdings said in August that traffic from AI assistants, paid and unpaid, was well under 1% of its room nights. Hilton’s chief executive called it early and insignificant in traffic and bookings on 27 October, and Hyatt said 2 days later that bookings from these platforms are very small. Both expect it to grow. Anyone promising you a return from ChatGPT ads today is promising something nobody has measured.

Measure bookings in your own booking engine, against your own cost per booking from other channels, and treat everything else as a clue. The Slovak hotel saw why: the ad got clicks at a good price, and its visitors stayed 64 seconds against 223 from Google Ads, with Google returning 20.8 times the spend by its own count in the same 2 weeks. Each source counts click-through its own way (Similarweb’s panel counts clicks per reply carrying an ad), so never hold one source’s rate against another’s.

Give it time before judging delivery. OpenAI asks for at least 24 hours after launch before calling it zero delivery, and clicks refresh about every 15 minutes while spend can lag 7 to 8 hours. A test at the minimum budget tells you whether the channel is worth a bigger test; it will not prove the ads added anything. At 230 clicks a month, if 1 click in 100 books, you see about 2 bookings.

Probably not yet. The hotel feed is a limited beta for approved advertisers, and OpenAI’s page sends you to your account team if hotel options are not on your account. It lets a group upload its properties as a file and run one ad template across them.

For a European hotel the audience line decides it: a Swiss or Italian property could in principle be listed, yet would reach only people in the US. If US guests matter to you and you have an account team, ask. The file itself is easy to prepare: 1 description of up to 5,000 characters per property, 1 to 20 images, a location and, if you want them, guest ratings. That description is worth writing anyway, for your own site.

Skip them when the money would work harder elsewhere or when you could not read the result. I would hold off in any of these cases, and most hotels will recognise at least one.

Spend the first week on your own pages and your reviews, before any money goes out. This order costs little, and the pages stay useful if you never buy an ad.

The step-by-step setup is in how to advertise on ChatGPT, and what changes for Italian, Swiss and other European hotels is in ChatGPT ads in Europe.