For twenty years a search ad was one thing: a block of text you wrote, placed beside the organic results. The two formats Google introduced in May 2026 at Google Marketing Live break both halves of that definition. You do not write the copy, and it does not sit beside the answer — it sits inside it.
The two formats, briefly
Conversational Discovery ads
These answer the user’s specific question with creative Gemini builds for that individual query, using the assets and data the brand has made available. It is not a variant picked from a list: it is a composition made in the moment out of the material you uploaded.
Highlighted Answers
These place relevant ads inside the recommendation lists AI Mode produces. When the generated answer is a list — "best X for Y", "what you need for Z" — an ad can appear as one of the entries, labelled as such.
In both cases the ad carries a Sponsored label and is accompanied by an explainer Gemini writes independently alongside the creative: text the brand does not control, there to tell the user why that result is present.
The difference that matters. In classic Search you buy a position and put your message in it. Here you supply raw material and the system composes the message. Control moves upstream: not over what is shown, but over what the system has available to compose it from.
Where the rollout stands
Conversational Discovery ads and Highlighted Answers are in testing in the United States, on mobile and desktop. Industry tracking published during 2026 puts ads in roughly 25.5% of AI Mode results — a meaningful density, which explains why "when does this reach Europe" is the most frequent question and the least answerable one.
The honest answer is that there is no public date for most markets. The useful answer is that the preparation work does not depend on the date.
What actually changes in the day job
If creative is composed on the fly, three things you currently treat as secondary become central.
1. Assets stop being ads and become raw material
In the RSA model an asset is a headline that may appear verbatim. In the generative model an asset is a piece of information the system may rephrase. That changes the writing criteria: brevity and punch matter less, specificity and coverage matter more. A generic headline like "Quality since 1985" gives the system nothing to work with; "Next-day delivery nationwide, free returns within 30 days" gives it two facts it can slot into the answers to two different questions.
2. Questions matter as much as keywords
Queries in AI Mode are longer and more conversational than classic Search queries. The material that makes you eligible to appear inside an answer is the material that answers questions, not the material that repeats product categories. FAQ pages, buying guides, complete spec sheets and comparison pages acquire paid value, not just organic value.
3. Product data becomes creative
If you sell online, what is in the feed — availability, sizes, materials, delivery times, returns policy — is what the system builds the answer from. It is the same reason we devoted an article to the Merchant Center product data specification: an incomplete feed is no longer only a coverage problem, it is a representation problem.
The measurement problem
The least-resolved part of all this is how you measure it. An ad inside a generated answer participates in a journey where the user can get what they came for without ever leaving the page, or leave it much later in the decision process than a traditional click implies.
Two practical consequences:
- The click is worth more and you will see fewer of them. It is the same pattern we documented analysing the impact of AI Overviews on CTR: less traffic, better qualified. Anyone judging on click volume reads the phenomenon backwards.
- Attribution leans harder on the data you upload. Offline conversions, true customer value, first-party data: these are the only things that stay stable when the interaction surface moves inside the results page. We wrote the guide to closing that value loop in lead generation.
What to do now, before it arrives
| Area | Action | Why now |
|---|---|---|
| Text assets | Replace generic claims with specific, verifiable facts | Improves RSAs and AI Max today, and becomes raw material for generative formats |
| Site content | Pages that answer real questions, not only category pages | It is the material the system builds answers from |
| Product feed | Complete attributes, not only the required ones | In conversational commerce the data is the creative |
| Measurement | True customer value, offline conversions | Holds up when the click becomes rarer and later |
| Brand | Guidelines on what may be said about the brand | The system rephrases; without boundaries it rephrases freely |
An honest perspective
There is a strong temptation right now to treat every Google Marketing Live announcement as a paradigm shift to chase immediately. It is worth resisting. These formats are in testing in one market, in one language, across part of the traffic. No budget should move today on their account.
What is worth doing today is the work you need anyway: specific rather than generic assets, content that answers questions, complete feeds, measurement based on real value. Those are the same things that improve your current campaigns. That they also happen to be the prerequisites for the future formats is a bonus, not the reason.
If you want to know how ready your material is for this transition, the free audit looks at assets, content and tracking with exactly this lens. You can also see how we work.