Ads inside AI assistants: what a hotel should know before paying for a sponsored answer

Ads in AI assistants are paid, labeled placements shown below an AI chat answer, sold at auction and kept separate from the answer itself. One large provider has tested them in the United States since 9 February 2026 (OpenAI Help Center, "Ads in ChatGPT", accessed 7 November 2026) and lists Türkiye as available for advertisers (OpenAI Help Center, "Ads Manager Availability", accessed 7 November 2026), but has not said whether travelers in Türkiye see them.

An agency calls and offers your hotel visibility inside AI assistants. The ad format is real and the invoice will be real. The question nobody has answered for you is whether the travelers you want are in the room to see it. This article separates what is actually sold from what agencies imply, tells you where to check availability for a hotel in Türkiye, explains how a click is measured, and gives you a small test with a budget cap and a stop rule. The arithmetic is the same one you already use for every paid channel.

What has actually been announced about ads in AI assistants?#

One large assistant provider began testing ads in the United States on 9 February 2026 and is rolling them out gradually to eligible users in select regions, according to its help center article on ads (opens in a new tab) (OpenAI, accessed 7 November 2026). The format is a single labeled ad unit below a response, sold at auction.

The format is narrow and specific. One ad unit, which may hold one or several images, appears below a response. The provider is also testing a multi-image visual format during image generation. Ads are clearly labeled and kept separate from the response. They are not placed near sensitive topics such as personal health, mental health or politics, and political advertising is not allowed. Ads are selected by relevance and advertiser bids.

Advertisers receive only aggregated reporting such as views and clicks, never users' chats. That limit matters more than it first appears, and we return to it in the measurement section.

It is worth separating this from booking news, because the two get confused. In October 2025 a plan for third-party apps inside chat was announced (Skift, 6 October 2025), and in August 2026 a search company began rolling out hotel booking in AI Mode for US travelers (Skift, 27 August 2026). An ad slot is neither of those. Buying a labeled placement under an answer does not put your hotel inside the answer, and it does not let a traveler book through the assistant.

No verified public figure exists for ad share, for the share of hotel bookings made through assistants, or for any count of hotels using the channel. Treat any such number an agency quotes as a claim, not a fact.

Who actually sees an ad#

Only some users see ads at all. Users on the Free and Go plans may see them. Plus, Pro, Business, Enterprise and Edu accounts have no ads. Accounts identified as belonging to people under 18 see none, and Temporary Chats show none. Ads in logged-out experiences are described as suitable for all audiences. Free users can switch to an ads-free option with lower limits.

For a hotel, the practical reading is that your most engaged, highest-spending travelers, the ones who pay for premium AI tools for work, are the least likely to see your ad. The audience you can reach skews toward free-plan users.

What the provider says it is selling#

The ad and the answer: can a paid placement change what the assistant says?#

According to the provider, no. The answer is composed by the chat model and the ad is selected by a separate advertising system, so a hotel buys a labeled slot under the answer, not a sentence inside it.

In plain words, two systems are at work. The model writes the response to the traveler's question. A separate advertising system picks an ad unit by relevance and bid and places it below the response, clearly labeled. The provider states that ads do not influence the answers, that they run on systems separate from the chat model, and that advertisers cannot shape, rank or alter responses. This is the provider's own statement, not the result of an independent test, and it should be read as such.

What this buys is a position. What it does not buy is presence in the answer text. You cannot pay to be described accurately, favorably, or at all inside what the assistant says about, for example, family hotels in your town.

There is an uncomfortable half to this separation. The same wall that keeps you from buying a sentence means a paid slot can sit next to an answer that gets your address, your rates or your facilities wrong. That is money spent beside an error, and no amount of ad budget fixes it. The free work comes first: publish one accurate, consistent set of facts about the property where assistants read it, before any paid budget. We cover how travelers asking instead of browsing changes your website in AI search and hotel websites.

On the traveler's side, the design is defensible: ads are labeled and kept separate, and a user who does not want them can move to a paid plan or to an ads-free free option with lower limits.

Can a hotel in Türkiye buy ads in AI assistants?#

On the buying side, yes: the provider's availability page lists Türkiye as available for self-service Ads Manager. On the delivery side nobody can say, because neither of the provider's pages lists the countries where users see ads.

Split the question in two, because agencies blur it. First, can a Turkish company buy? Second, will travelers in Türkiye be shown the ad? These are different questions with different answers.

On the buying side, the Ads Manager availability page (opens in a new tab) (OpenAI Help Center, accessed 7 November 2026) lists Türkiye as available for self-service Ads Manager, and states that the legal entity that will advertise and be billed must be based in a country listed as available. On the delivery side, the ads article says only that ads are rolling out gradually to eligible Free and Go users in select regions. Neither page lists countries, so neither says whether travelers in Türkiye see ads.

One European detail often confuses Turkish advertisers. Advertisers in the new European markets start through the provider's ads solutions team, agency partners or technology partners, and personalized ads are not initially available in the European Economic Area or Switzerland. Türkiye is not in the EEA, but the pattern shows how regional the rollout is, and why a written answer matters more than an agency's confidence.

Ask for three things in writing from the provider or the agency: which legal entity is billed, which regions the campaign may serve, and whether reporting shows the geography of the people who saw or clicked. If the country list is not published, an agency's assurance about Turkish travelers is a claim, not a fact. The agency cannot know either. And if the answer is "not yet", the correct move is to prepare the account and the facts, not to spend. Before signing anything with an agency on this topic, work through our agency checklist for AI search.

Who would you bid against, and how is the price set?#

You bid against anyone targeting the same questions, and the auction gives no published list of who that is. Ads are sold by auction and the winning bid sets the price, so there is no published rate card to plan against.

A campaign is charged on clicks or on impressions, depending on its billing setting, and the two carry different risk for a hotel with a small budget. The strategy timeline matters too, because it changes what you should build. Intelligent bidding, Max Bid and Max CPC exist now. Target CPA and Target CPC replace them for new campaigns from 9 November 2026, and Max Bid and Max CPC retire for existing campaigns on 1 January 2027 (OpenAI Help Center, "Bidding strategies and how pricing works in ChatGPT Ads", 2026). Build new campaigns on the strategy that survives, and do not let an agency lock you into a setting that disappears.

Relevance and bid together decide who appears, and you cannot see who else bids on the same questions, and the auction publishes no bidder list. No public figure exists for ad share or typical prices in this channel, so any percentage an agency quotes is a claim to be tested. The number that decides the channel is cost per booking measured against what the hotel keeps, the same comparison used for every other channel, which we set out in distribution cost per booking.

How do you measure a click from an AI assistant to a booking?#

You measure it like any other referral: a tagged link, a landing page that matches the question, and a booking recorded against that source in your own system. The provider's report stops at the click.

The mechanism is ordinary web analytics. The click arrives as a normal visit to your site, so campaign tagging in the link is what lets your analytics and your booking engine recognize where it came from. Without tagging, the visit lands in the same bucket as everything else and the spend is unreadable.

The limit is structural. Advertisers receive only aggregated reporting such as views and clicks, never users' chats, so the provider's numbers end at the click and cannot follow a traveler to a reservation. Close the gap yourself: match the landing page to the question, keep one page per offer, and make the booking path short enough that the visit is not wasted.

Plan for the assisted path too. A traveler reads in the chat, then books later by searching the hotel's name or by calling the desk. Ask at booking how the guest found you and keep the answer in one field. Define the result before launch: a booking visible in your own system with a value attached, not a click and not a session. A high click count with no bookings is a cost, not a signal of progress. For the full setup, see measuring visits and bookings from AI assistants.

What does a small, honest test look like?#

A capped test on one offer and one date window, with a stop rule written before launch and a result defined as bookings in your own system.

Set the scope first: one destination question, one date window, one offer, so the result can be read without argument. Set the cap: a fixed total small enough that losing all of it teaches you something and hurts nothing. Write the stop rule before launch: stop when cost per booking passes a stated figure, or after a set number of days with no booking at all.

Define the result in bookings with a value, and say who checks it and on which date. Compare against the alternative: the same money placed in the channel you would otherwise have used, with the same cost per booking arithmetic. For example, a hotel that knows its average room rate and its target acquisition cost can state in advance what a pass and a fail look like.

Record the null result too. A written note that the channel was not yet available for your market, or produced no bookings at that budget, is a real finding and stops the next agency conversation from starting from zero. Do not scale on click-through rate alone, and do not renew a test because the dashboard looked busy.

What to do next#

Write the one-page test sheet before you take any agency call: scope, cap, result definition, stop rule, comparison channel and review date. Then get the availability answer in writing, because a Turkish company can open an ads account while the provider still does not say whether travelers in Türkiye see ads. The order matters: fix the facts the assistant reads about your hotel first, measure the click to the booking second, and only then decide whether a sponsored slot earns its place next to the answer.