You Can Now Buy an Ad Inside ChatGPT. First Ask Who's on the Other Side.

Since August 24, a hotel in 31 European countries can pay to appear under a ChatGPT answer. For the first time there's a price list. And since October 26, for the first time, there's data on who is actually asking.

I've bought twenty years of hotel ads: directories, metasearch, Google, Meta. Every new channel arrived with the same two questions. What does it cost, and who's on the other side? This one finally has answers to both.

What it costs

• Only the free tiers see ads. Free and Go (€7.99 a month). Plus, Pro and Business users never do. Whoever pays for ChatGPT will never see your hotel there.
• A text ad, labeled "sponsored," under the answer. 24 characters of title, 48 of description. One image, square. In Europe it's contextual only: the current conversation, location, language. No personalization, no custom audiences.
• Minimums around €15 a day in Europe, $25 in the US. Bids under $3 a click rarely serve.
• The first hotel numbers, from a US agency's beta: average click at $3–3.50, click-through rate of 1.6–1.9% — far above display — and the ads already make up 31–56% of its hotels' total AI traffic. Everything lands in GA4 as "AI Assistant" or "Referral."

Not expensive. Not cheap either, for a 30-room hotel: €450 a month is a reservations shift. So the second question matters more.

Who is actually asking

On October 26 OpenAI published "Signals," a public dataset of how people use ChatGPT, by country, age and plan. Martin Soler went through it for hoteliers. Three things stand out.

• There is no travel category. No destinations, no hotels. Any "X% of ChatGPT is travel" you'll read in the coming weeks is a guess.
• Information-seeking is 20% of all messages, up from 16.7% in mid-2024, and three quarters of it is non-work. By country: Spain 26.3%, France 26.1%, UK 23%, Germany 22.9%, US 21%. India and Brazil around 16%.
• The older the user, the more they ask. 18.1% of messages from 18–24-year-olds are information-seeking; 25.2% from 55–64-year-olds. That's the guest of a seaside hotel, not of a hostel.

Soler's example: a Paris hotel with mostly American guests. 46% of US messages come from people over 35, about 21% are looking for information. Promising — except nobody knows how many of them are on the free plan, the only one that shows ads. Do affluent travelers pay for software they use casually? Often not. The dataset doesn't say.

Discovery, not closing

The agency that ran the beta says it plainly: treat this as upper-funnel, a discovery channel, and if your hotel is already cited organically in AI answers, put the money in Google Demand Gen instead. A consultant quoted this month reaches the same place from the other side: it suits lifestyle and luxury independents as a way to be seen before the search starts.

Which means the test comes before the campaign. Ask the chat about your hotel, in your top source market's language. If you're already in the answer, the ad is paying for a seat you have. If you're not, that's what the ad is for — and the free fix (the exact facts, everywhere) comes first.

The takeaway: don't decide this with the dataset; decide it with a small test. Soler's recipe, which I'd follow as written: 30 days, one leisure segment, one or two source markets where the age mix and the information-seeking share look like your guest — Spain and France top the list — at the minimum budget, and one metric: bookings tagged "AI Assistant" in GA4 against what you spent. Before you start, ask the chat about your hotel; if you're already there for free, the money goes elsewhere. A new channel is not a new strategy. It's one more door, and you read it like you read the OTA twenty years ago: by the cost of each guest who actually came through it.