In little more than seven months, advertising inside ChatGPT has gone from a test limited to users in the United States to being available in close to forty markets across the Americas, Europe and Asia, and the question now reaches every hotel marketing meeting with the same wording: does it make sense to move part of the online advertising budget into this new SEM channel?
Answering that with any rigour means putting three more questions on the table: what kind of visibility can actually be bought on ChatGPT Ads, what it costs, and what evidence exists that the investment comes back as bookings. With the data available today, none of the three answers justifies an abrupt budget shift, and all three justify starting to test ChatGPT Ads with caution.
What kind of visibility do you actually buy inside ChatGPT Ads?
The first thing most people picture when they hear about ChatGPT Ads is their hotel brand appearing inside users’ conversations with the AI. The mechanics work differently. Advertising inside ChatGPT is always labelled and kept separate from the AI’s answers, and it has no influence whatsoever on what ChatGPT replies. The format sits at the end of the conversation, activates when there is a relevant product or service connected to the context of that conversation, and is excluded from sensitive topics such as health, mental health or politics (OpenAI, Search Engine Land). What you buy, therefore, is an identifiable promotional space inside a conversation, much closer to a classic sponsored result than to an editorial recommendation.
The inventory does not cover the whole user base either. Only registered adults on the Free and Go plans see advertising, while users on Plus, Pro, Business, Enterprise and Education will never see your ads, and anyone under the age of eighteen is also excluded. That means the reachable audience is limited to people using the free version of the tool and those on the cheapest paid plan, which narrows the target audience considerably further.
There is also a nuance in Europe that changes how targeting works. OpenAI has chosen to ask for explicit consent rather than rely on legitimate interest, so the user chooses between personalised or contextual advertising. Personalised targeting uses the topic of the conversation, approximate location, device type, time of day and language, while conversation history and memory stay out of targeting at the European launch. Anyone expecting surgical targeting fed by years of accumulated conversations finds, for now, something much closer to plain old contextual segmentation.
An advertising channel with barely seven months of history and a regulatory layer ahead of it
OpenAI announced the tests on 16 January 2026 and began showing ads on 9 February in the United States, extended to Canada, Australia and New Zealand on 26 March, added the United Kingdom, Mexico, Brazil, Japan and South Korea on 7 May, and left Europe until last, with the announcement of 18 August and effective activation on the 24th across 31 markets (OpenAI, Dataconomy). A Spanish hotel considering an investment today is therefore looking at a channel with ten days of history in its own market, with no data of its own, no reliable local benchmarks and no open platform for buying without assistance, because access still runs through OpenAI’s Ads Solutions team and approved partners while the Ads Manager finishes rolling out.
A regulatory layer has just been added on top of that, and it will shape the channel over the coming months. On 31 August 2026 the European Commission designated ChatGPT as a very large online search engine under the DSA, after the company reported 159.1 million average monthly users in the European Union, with a compliance deadline at the end of December and specific obligations for the advertising business, among them keeping a public ad repository, explaining which parameters determine who sees each creative, and giving up profiling of minors (PPC Land). More transparency for the advertiser, and also more operational friction when launching campaigns.
What campaigns on ChatGPT Ads cost
The channel launched with a CPM of sixty dollars and a minimum spend close to 200,000 dollars, which dropped to 50,000 in April and can now come down to 10,000 if the entry point is a partner such as Criteo (eMarketer). Its technology has improved quickly and already includes CPM and CPC bidding, conversion optimisation, geotargeting, custom audiences, its own pixel and a conversions API (OpenAI), yet that threshold remains out of reach for, say, a property with sixty rooms that spends three thousand euros a month on acquisition and has to justify every euro against its cost of acquisition.
Very little has been published on actual performance, and the success stories deserve a cautious read. The most transparent documented case covers a single account with around 60,000 dollars invested over fifteen days, returning a blended ROAS of 1.49, a cost per click of 1.72 dollars and a conversion rate of 2.35%, with daily volatility swinging between 0.2 and 2.9 ROAS and an explicit warning that the economics of arriving first inflate the numbers and will not hold once competition shows up (according to data from Opascope). As a point of contrast, Google search benchmarks in travel and hospitality sit at around 2.12 dollars CPC, 8.7% CTR and 5.8% conversion (PPC Chief), with fifteen years of history behind them and an attribution model the industry already knows how to read.
The channel’s own financial pulse adds a useful warning. ChatGPT has already passed one billion dollars in annualised advertising revenue and OpenAI is aiming at 2.3 billion this year and 100 billion by 2030, supported by more than a billion weekly users, around twenty per cent of whom show commercial intent. eMarketer, by contrast, calculates that the 2030 target will fall short by roughly ninety per cent, and points to an advertising infrastructure that is not yet ready to absorb large budgets. Between the optimistic headline and the conservative forecast there is an enormous gap, and that gap is where the choice between entering now and entering two quarters from now gets settled.
Is hotel demand already going through artificial intelligence?
Comparing global audiences adds little, however striking the contrast, because Google processes close to 16.4 billion searches a day and holds around ninety per cent of the global search market (according to data from DemandSage). No revenue manager makes decisions based on total market size. What we would need to know is how much booking intent is shifting towards conversational interfaces, and that figure is still not published in a way that can be compared across platforms.
The signals that do exist point in a fairly clear direction. Adobe measured, in a study published in May 2026 and covered by Search Engine Land, 194% year on year growth in traffic reaching travel websites from artificial intelligence systems, with a cumulative 2,215% since October 2024, and describes a visitor who behaves better than average, staying 70% longer per visit, showing 21% more engagement and bouncing 41% less. That same visitor still converts 28% below traffic that does not come from AI, although the gap has narrowed by almost 70% since October 2024. The study includes one figure that matches what we find in audits: only 63% of hotel home pages are readable for AI systems, against 73% of their product pages.
On the traveller side, the consultancy McKinsey puts at 55% the share already using ChatGPT or applications built on AI to plan their trips, with more than 90% trusting the accuracy of the answers and barely 2% willing to hand full autonomy to an agent. What those numbers describe is a tool that already owns inspiration and comparison but does not yet close the transaction, which fits with OpenAI pulling Instant Checkout back in March towards the applications model inside ChatGPT, where Expedia, Booking.com, Skyscanner, Accor and Lighthouse operate, once it became clear that a travel booking drags along price volatility, rate rules, extras, cancellations, modifications and everything that comes after the sale.
Keep an eye on Google’s next moves
Treating Google as the passive actor in this story would be a misreading. Its AI Mode has passed a billion monthly users and the company is using the travel vertical as the showcase for its AI advertising, with IHG and Booking Holdings inside the Direct Offers pilot. On 27 August 2026 it went a step further and enabled full hotel booking inside AI Mode with ten launch partners, among them Booking.com, Expedia, Hilton, IHG, Marriott, Choice, Priceline, Trip.com, Wyndham and Hotels.com, for now only in the United States and in English, keeping the hotel or the platform as merchant of record but without clarifying commissions, ranking criteria or whether sponsored positions will appear inside that flow, and with no date for Europe while the DMA sits in the way.
In parallel, tools such as AI Max hand Google control over match types, creatives and landing page, with an effect worth watching in the hotel case, because a property inside a chain can end up giving traffic away to another establishment under the same brand if URL inclusions and exclusions are not managed. The more automated media buying becomes, the more expensive it gets to have no judgement of your own overseeing it.
Does it make sense to split the SEM budget between ChatGPT Ads and Google Ads?
The defensible position today is to treat ChatGPT Ads as what it is, a test channel with inventory that is still cheap for lack of competition, immature measurement and an investment threshold many independent hotels cannot justify without sacrificing what already works for them, while Google keeps concentrating demand with transactional intent and metasearch and brand keep carrying the bulk of direct bookings.
Before arguing about which platform deserves the money, there are three fronts that return more and cost less. The first is making sure the website and the booking engine are readable for AI systems, because presence inside generated answers cannot be bought with budget and is today the only real way of appearing there. The second is having measurement in order, with a pixel, a conversions API and attribution that does not break when the visitor arrives from a conversation rather than from a search, which in most accounts we review is still not the case. The third is setting aside a limited budget, between 5 and 10% of acquisition spend, to test the channel with a specific hypothesis, a defined source market and a window of at least four weeks, because a single week of data measures volatility rather than performance.
From there the decision gets made with your own data, which is the only evidence that ends up mattering in a profit and loss account. If ChatGPT Ads produces bookings at a defensible cost of acquisition, you scale. If it does not, the budget goes back where it was and the decision gets revisited a quarter later, when the channel has some history behind it. Framing the discussion around being modern leads nowhere, because the traveller does not choose platforms, only the tool in front of them when a trip needs sorting, and the job of a revenue and marketing team is to be present at that moment with the right value proposition and a cost per booking the hotel can defend. Setting up campaigns has been within anyone’s reach for years. Making them profitable is another matter.
