Google's advertising documentation answers three questions about ads in AI Overviews with the same word. Can you target the placement? No. Can you opt out? No. Will you get segmented reporting? Not currently. Publishers looking at the same feature from the supply side get one switch, in or out, and a report that counts impressions but not clicks. The most consequential new surface in search arrived with its control layer stripped from both sides at once.

  • Google Ads documentation confirms advertisers cannot target ads in AI Overviews, cannot opt out of them, and get no segmented reporting, because in-unit placements are counted as Top Ads.
  • Publishers received a domain-level toggle documented in July 2026 offering inclusion or exclusion, plus Search Console reports that expose impressions without clicks or click-through rate.
  • The open-web search economy ran on granular controls: match types, negative keywords, page-level directives, query-level reporting. AI Overviews retired that layer for buyers and suppliers in the same year.
  • A placement that cannot be selected, excluded or counted cannot be priced, and unpriced inventory always settles in favor of the party operating it.
  • The first granular controls on the way are regulatory rather than product: the CMA requires page-level grounding controls by March 3, 2027.

What a control layer is, and what ads in AI Overviews removed

A control layer is the set of levers that let an advertiser choose where a message runs, let a publisher choose whether its content is used, and let both check the result afterwards. In classic search that layer was granular and symmetric. Buyers had match types, negative keywords and placement exclusions. Publishers had noindex, nosnippet and page-level directives. Both had query-level reporting. The AI Overview ships without any of it.

The auction you could steer, the answer you cannot

In the link economy, control and price moved together. You bid on a unit you could name, excluded what you did not want, and the report told you what you bought. In the citation economy, Google's systems decide whether an ad belongs inside the generated answer by weighing the query and the text of the overview itself. Relevance replaces the bid as the deciding signal, and the buyer never sees the decision.

What PPC Land documented, and what we checked

PPC Land's explainer, published August 25, 2026 by Luis Rijo, assembles the record: query fan-out, twelve English-language markets for in-unit ads against more than 200 for ads above and below, and the three constraints taken straight from Google Ads Help. Two figures needed updating. The piece puts AI Overviews at over 2 billion monthly users; Sundar Pichai put it past 2.5 billion at I/O in May 2026 and Alphabet restated that with Q2 2026 results, so we use 2.5 billion. It also cites a Semrush prevalence figure of 13.14 percent of results, while BrightEdge tracked overviews on roughly 48 percent of monitored queries by February 2026 and Similarweb put the figure above 40 percent of Google searches in August 2026. Treat 13.14 percent as a reading from an earlier phase.

Why good SEO is good GEO answers the wrong question

The case that no new playbook is needed

Google's position deserves stating properly. Brendon Kraham, VP of Search and Commerce for Global Ads Solutions, wrote on Think with Google in June 2026 that generative features run on the same ranking and quality systems and the same index as classic Search, so foundational SEO remains the launchpad and no special markup or AI files are required. Google also reports more than 10 percent query growth on the query types that trigger an overview. On that reading, marketers mourning the old dials are mourning labor, not outcomes: automated matching has been beating manual keyword control in paid search for years.

Eligibility guidance is not a market structure

All of that describes how to deserve a placement, not how to buy one. A market needs an addressable unit, an exclusion, and a report. Performance Max at least returned a conversion to price against. Here the in-unit placement is folded into Top Ads, so a buyer cannot separate spend that reached the new surface from spend that did not. Similarweb launched a third-party dataset on August 17, 2026 precisely because no public ad library covers these placements, while Meta, Google's own display business and TikTok all have one. When outside vendors build the transparency layer a platform declines to ship, the market is naming what it lacks.

What this means for brands and for publishers

For CMOs, media buyers and agencies: fund the surface you can still steer

Treat AI visibility as a media line rather than an SEO chore, then accept that one part of that line is currently unmanageable. Inside Google's own surfaces you influence eligibility and nothing else, so rely on third-party detection instead of platform reporting to know whether you appear at all. Put working budget where an addressable unit still exists: the publisher pages AI engines cite, where a placement can be selected, counted and priced.

For publishers: a binary switch is not a control panel

The July 2026 toggle offers inclusion or exclusion and nothing in between, and the June 2026 Search Console reports show impressions without clicks or click-through rate. That is leverage over Google, not revenue from Google. The asset that stays entirely yours is the page itself, where you can sell a placement, count it, and get paid whether the reader is a person or an agent.

Three signals that decide who prices the answer

First, the CMA calendar: main publisher controls on December 3, 2026, page-level grounding controls on March 3, 2027, and an obligation to supply impressions, clicks and click-through rate through Search Console. The first granular dials in AI search are being written by a regulator, not shipped by a platform. Second, independent measurement: Similarweb reports sponsored ads in 26 percent of ChatGPT responses and nearly 30 percent of ad-eligible AI Mode queries, less than a year after Adthena detected in-unit Google ads at 0.052 percent across 25,000 results pages in November 2025. Third, format maturity: Adthena's 29-million-query analysis describes the mature in-unit format as a product carousel matched through feeds and priced per click. Formats standardize, then prices follow, and whoever holds the controls at that moment sets them.

Conclusion

Hold on to this: ads in AI Overviews removed targeting, exclusion and reporting from the buy side in the same period that publishers were handed a single on-off switch, and a placement neither side can steer or count is a placement neither side can price. The category that closes that gap is Generative Engine Advertising, native ads for AI agents placed on the publisher pages AI engines actually cite, buyable by brands as a nameable unit and paid through to the media sources feeding the answer. That is the market Smalk AI is building on both sides. What to watch next: whether the disaggregated click data Google owes under the CMA requirement becomes a public benchmark, because the moment the answer carries a measurable click it acquires a price, and whoever holds the controls will set it.