AGTNEWS
When the answer is the ad: how Gemini reshapes the Search auction
Google is folding paid placement into the model's response itself. The auction mechanics look familiar; the inventory does not.
Google is folding paid placement into the model's response itself. The auction mechanics look familiar; the inventory does not.
For two decades, Google Search sold adjacency: a paid link sat beside the answer, and the answer was a list of links. That separation ended at Google Marketing Live on May 20, 2026, when the company unveiled four Gemini-powered ad formats: Conversational Discovery, Highlighted Answers, AI-powered Shopping ads, and Business Agent for Leads. Each one places paid content inside the AI's generated response rather than next to it. The sponsored label is still there. The list of blue links is not.
The unit an advertiser is buying has moved from a position on a results page to a citation inside a model answer. The change is small in copy and large in consequence, because the auction, the eligibility rules, the conflict of interest, and the publisher economics all shift at the same time.
Conversational Discovery ads use Gemini to generate ad creative tailored to a specific query, then pair it with an "independent AI explainer" that the model runs alongside the ad to evaluate and synthesize product and service context. Both remain marked "Sponsored," and the explainer is the same model that produced the rest of the answer. Search Engine Land reported that the formats are being tested in the United States on mobile and desktop. Highlighted Answers insert what Google describes as highly relevant, high-quality ads directly into AI Mode recommendation lists. The Register, citing Google, says these use the same auction mechanics as existing Search ads and the same ad-quality filters. The continuity is doing a lot of work in the announcement: it positions the new surface as a slot rather than a new kind of persuasion.
AI-powered Shopping ads pull from merchant feeds and assemble product recommendations inside the answer. Direct Offers is being expanded with promotion bundling, native checkout for merchants using the new Universal Commerce Protocol, travel deal integrations, and AI-generated offer recommendations. Business Agent for Leads, an AI chat agent trained on an advertiser's website and embedded in lead-gen ads, is moving to open beta for U.S. advertisers.
Google's pitch is that the auction is unchanged. Per The Register's reporting, Highlighted Answers "use the same auction mechanics as existing ads." The relevance filters are the same. The ad-quality policy is the same. What has changed is the canvas: the model's generated text is the page, and the paid placement is woven into the prose of the answer rather than displayed beside it. The sponsored label is still present, per Google's Marketing Live announcement, but the visual grammar that once separated paid from earned is gone.
That is where the conflict of interest lives. In the old Search, the sponsored slot and the editorial result were both lists, and a reader could pattern-match the difference with a glance. In an AI answer, the model does the synthesis, and the line between citation, recommendation, and paid placement is a function of how the model has been prompted to render them. The training pipeline that decides which answer is "highly relevant" and which ad is "high quality" is the same pipeline. The explainer is the same model that produced the recommendation. The sponsored label is a tag, not a partition.
The behavioral claim Google is leaning on is that 75% of people in AI Mode report making faster, more confident decisions, according to Google's own survey copy. That is a Google-cited number, not independent measurement, and any draft that turns it into "users prefer AI ads" is overreading the source.
The new formats are not available to every advertiser by default. Eligibility flows through Performance Max and AI Max, Google's campaign types that automate bidding and creative across surfaces. Search Engine Land's coverage notes that being present in the new answer-style surfaces requires buying into the AI Max stack, which is where Google is moving the automation that decides which creative, which landing page, and which bid gets assembled for a given query. For an enterprise advertiser with a measurement team, AI Max is a cost of entry. For a small business running a single search campaign, it is a forced migration: the campaign type that used to be enough to win a sponsored slot is no longer the surface where the ad will appear. The long tail of advertisers is being asked to hand creative assembly, audience targeting, and bid shaping to the same system that generates the answer. That is good for Google's automation business. It is a different deal for the independent agency or the in-house team that used to write the ads by hand.
Publishers have been watching the answer surface eat their traffic for a year. Digiday reported on Digital Content Next survey data showing a median drop in Google search referral traffic of roughly 10% year over year for publisher members, with news down 7% and non-news down 14%. The new formats accelerate that pattern. When the citation is the ad unit, the publisher's role shifts from being a destination the model links to to being a source the model quotes, and the quote may be the only thing the reader sees.
Per Digital Applied's summary of Google statements, AI Mode had reached roughly 75 million daily active users and over 100 million monthly active users by late 2025, was processing more than a billion queries a month in the U.S. and India, and AI Overviews were appearing on 48% of queries. The same summary cites a 25.5% rate of ads in AI-generated results, a 394% increase, and a reported max drop in organic click-through rates of 61%. Those numbers come from an SEO agency citing Google executives, not from Google's own filings, and they should be read as directional, not audited.
The publisher question is whether the citation economy, where a publisher's content is summarized in a model answer that may or may not link out and may or may not place a paid answer next to it, has a workable commercial model at all.
The unit being bought is no longer a click. It is a presence inside an answer. That breaks a generation of measurement assumptions, because the click was the cleanest signal an advertiser had, and a citation inside a model response is not a click. Advertisers will end up measuring second-order signals: branded search lift, direct traffic, post-answer site visits, and whatever proxy the AI Max stack offers. Vendors that sell attribution rather than auction optimization are the ones best positioned to grow in this environment, and the agencies that survive will be the ones that can explain, in plain language, whether their client's name appeared in the model answer and what context surrounded it.
Three things will tell us whether this is a slot or a re-platforming. First, whether Conversational Discovery and Highlighted Answers roll out broadly in the U.S. or stay in a long test. The current state is testing, per Google's announcement and Search Engine Land's coverage, and the gap between test and general availability is where the regulatory questions live. Second, whether the "independent AI explainer" remains a model that is separately tuned from the ad-generating model, or converges with it. If the explainer is a distinct system, the format has a chance of being read as a labeled comparison. If it is the same model with a different prompt, the "independent" framing is mostly a label. Third, whether any advertiser or publisher files a complaint, in the U.S. or abroad, that the answer/ad conflation crosses a line the existing ad-disclosure rules were not written for. The UK's Competition and Markets Authority has already been told by the Professional Publishers Association that Google's AI surfaces are reshaping publisher economics, and the U.S. conversation is moving in the same direction.
The ad is no longer next to the answer. It is the answer's co-author, signed at the bottom. The question for the next year is whether the auction, the disclosure rules, and the publisher economy can be rebuilt fast enough to keep up with it.