OpenAI expanding ChatGPT ads globally and building out its ad stack is not a small product update. It is the announcement that a chatbot is becoming a media property, with everything a media property needs: formats, targeting, a way to sell placements at scale rather than one experiment at a time. Google is running a parallel experiment from the other direction, testing search ads with restrictive match types inside AI Mode, its answer-first search product. Two companies that used to compete for the same query on the same results page are now both trying to sell a placement inside a conversation. That single fact rewrites how creative gets made, how targeting works, and how anyone proves an ad did anything at all.
Restrictive match types are a warning, not a footnote
Search advertisers built entire account structures on broad and phrase match, letting the algorithm find adjacent queries a human planner would never have typed. Google testing restrictive match types specifically for ads served inside AI Mode says something plain: the loose matching that worked next to ten blue links does not work inside a synthesized answer. An AI Mode response is already an interpretation of the user's intent. Layering a broad-match ad on top of that interpretation risks contradicting or duplicating what the answer just said, which is a worse failure than an irrelevant ad on a normal results page, because it makes the assistant itself look confused. Restrictive matching is Google's way of keeping the ad tethered to something narrow and provable, which means media buyers who have spent years loosening their match types are about to be told to tighten them again, this time inside a completely different surface with different rules about what "relevant" even means.
The practical effect is that account structures built around volume, broad terms feeding a large pool of impressions, will not transfer. A conversational surface rewards precision over reach, because there is only one answer being generated, not ten links competing for attention. Planners who treat AI Mode placements as an extension of their existing search campaigns, rather than a distinct inventory with its own rules, are going to see spend disappear into ads that never qualify to show.
Getting cited is not the same as getting paid
The reporting on getting a product into ChatGPT versus getting it through checkout draws the exact line that ad buyers now have to plan around. A brand can win a citation inside an AI answer and still lose the sale, because the conversational interface sits between the recommendation and the transaction in a way a search results page never did. Click a blue link and you land on a site built to convert you. Get named inside a chatbot's answer and you are still one more step from a cart, a login, a checkout flow that has to work inside or alongside the conversation itself rather than on a page you control.
Ads inside ChatGPT and inside AI Mode inherit this same problem. Buying the placement is the easy half of the job, and the harder half is building a path from "the assistant mentioned us" to "the person bought something," which is a user-experience and engineering problem as much as a media one. Agencies that have spent years optimizing landing pages for search traffic now need to think about optimizing for a handoff that starts inside somebody else's chat window and only later, maybe, arrives on a page at all.
Citations are becoming the new keyword
Winning citations inside AI answers is now treated as its own discipline, the subject of ongoing work from Ahrefs on how brands earn a mention when there is no results page to rank on. That is the unpaid version of the same battle OpenAI is now selling paid access to. If a mention inside a generated answer is valuable enough that agencies are building a practice around winning it for free, it is valuable enough that OpenAI can charge for the paid version of the same real estate.
Semrush moving AI visibility under its Spotlight product is the measurement industry's admission that this is now a category worth a dedicated dashboard, not a side metric bolted onto rank tracking. The tools that used to tell you where you sat on page one are being rebuilt to tell you whether you got mentioned inside an answer at all, and soon, whether the paid placement inside that answer actually moved anyone toward a purchase. Expect the same vendors that built rank trackers to spend the next stretch of time building citation trackers, and expect clients to start asking for both numbers in the same report.
The traffic pressure is already showing up in the numbers
None of this is theoretical to publishers. USA Today has overhauled its audience strategy citing search pressure directly, and reporting on the biggest winners and losers in U.S. website traffic over the past year shows referral traffic already reshuffling around who gets surfaced by AI systems and who doesn't. When the traffic that used to arrive by ranking well starts arriving, or not arriving, based on whether an assistant chose to cite you, the incentive to buy a paid placement inside that same assistant becomes obvious. OpenAI expanding ChatGPT ads globally is not arriving into a calm market. It is arriving into a market where publishers are already restructuring around the fact that the old click is not showing up the way it used to, and where the next source of revenue may have to come from a surface that didn't exist as an ad channel a year ago.
What this means for the people planning the media
Three things need new playbooks at once. Creative needs a version that works when there is no headline field or description line, just a sentence competing to be the one the assistant chooses to surface inside its answer. Targeting needs a version built for restrictive matching against a paragraph-length prompt instead of a two- or three-word keyword, which changes what "intent" even means for a bidding system. And measurement needs a version that can tell the difference between a citation, a click, and an actual completed purchase, because those three things are no longer the same event happening in the same order they used to happen in.
None of this means the keyword is dead or that search ads are finished. It means the buy is moving from the page into the conversation, one platform at a time, and every part of the workflow that assumed a page, from creative production to attribution modeling, has to be rebuilt for a surface where the ad has to survive inside somebody else's sentence.