Ads in Google AI Mode: why your product feed now writes the ad

If you know Shopping ads, you think in campaigns, bids and audiences. For AI Mode that shifts. Google describes the new formats as built with Gemini inside Search and tailored to the concrete question a person asked. The ad does not go looking for a keyword; the conversation calls up the product, and what the ad is made of is your product data.

1. What Google announced

Ads for AI Search were introduced at Google Marketing Live 2026. From Google's own announcement:

• Ads in AI Mode: sponsored answers that appear inline while someone is asking follow-up questions.

• Direct Offers: product or offer prompts attached to an AI generated answer when the intent is commercial.

• Delivery runs through existing campaign types, above all AI Max for Shopping and Performance Max, not through a dedicated AI Mode targeting.

Google gives as an order of magnitude that advertisers using AI Max or Performance Max see more conversions at a similar ROAS. We pass that on as a vendor figure, not as our own measurement.

2. The real break: the feed does the work of the ad copy

The sharpest sentence from the practitioner reporting sums it up: your feed does the work the ad copy used to do. Gemini decides from the product data whether your product fits a question and how it explains the connection. Data quality beats campaign settings.

In practice a feed entry is no longer just a catalogue field. It is the material the ad is written from. A title like "men's shirt blue" carries less than a title that names cut, fabric and occasion. If a piece of information is missing from the feed, it is missing from the ad.

3. Getting the feed ready

From the reporting, and consistent with Google's focus on the feed:

• Make titles and descriptions specific: concrete detail instead of generalities.

• Fill the conversational attributes, above all question and answer pairs in the style of an FAQ. Exactly the questions a customer asks in a dialogue.

• For products that need explaining, give fabric, fit and durability in full.

• Use supplemental feeds so you do not disturb the entries already running.

• Start with your highest revenue items and bestsellers, not with the whole catalogue.

4. The other half: on the Search side your page writes the ad

The feed is only one of two rails. On the Search side a parallel shift applies the same mechanism to your website. It affects merchants who have not booked a single AI Mode format.

In AI Max for Search two settings work together. Google puts it like this: when Text customization and Final URL expansion are both on, Google may replace your given landing page with a more relevant one and "generate a dynamic headline, description, and additional assets to match your landing page content". Final URL expansion is on by default once you opt into AI Max, and it requires Text customization.

Translated: not just the feed, the text of your subpages becomes ad copy too. Any page Google considers a better fit can become the landing page. A thin category page with no substance is no longer a cosmetic flaw but a candidate for a weak ad.

None of this stays optional. Google has announced that from September 2026 existing Search campaigns with automatically created assets and with campaign level broad match will be upgraded to AI Max automatically. For Dynamic Search Ads, Google has since moved the date to February 2027, and new DSA campaigns are to become impossible to create through Google Ads, the Editor and the API. So the dates move; the direction does not. Run one of those settings and you get the change without ordering it.

You can still steer. Google introduced AI Brief for that. Messaging Guidelines set what your ads should and should not say. Matching Guidelines bound the searches you want. Text Disclaimers force required text to appear even when Final URL expansion swaps the landing page. At URL level you keep page feeds, rules on URL contains, and the exclusion of individual URLs.

As an order of magnitude Google cites 7 percent more conversions or conversion value at a similar CPA or ROAS when the full feature suite is used rather than search term matching alone. That is Google's own analysis for non retail advertisers, not an independent measurement, and we pass it on as exactly that.

5. The difference to ChatGPT Ads, precisely

The direct comparison is worth making, because the two platforms ask different things of you:

• Google: two rails, both driven by data. In AI Mode structured product data is central, the feed is the ad. Without a maintained Merchant Center feed there is nothing for Gemini to build an ad from. On the Search side the content of your subpages joins in, because Final URL expansion derives both landing page and ad copy from it.

• ChatGPT Ads: there the conversational context counts, and the landing page is read as a relevance signal. OpenAI explicitly does not require structured data on the landing page; it helps for product feed campaigns but is not a condition.

That is not a contradiction but a division of labour with two different levers. For Google you invest in the depth and accuracy of your product data. For OpenAI you invest in describing the buying situation and in landing pages that genuinely answer the question. If you want both channels, you need both.

6. Measurement is still unfinished

Update, 31 August 2026. The paragraph below still holds for the paid side. On the organic side something changed on 31 August, and it belongs here, because otherwise it sounds as if nothing exists at all: Google rolled out its Search Console report for generative AI features worldwide. It covers AI Mode, AI Overviews and generative features in Discover. Search Engine Land reports that it shows impressions, pages, countries, devices and dates, but no click data. That is not our own measurement, and we pass on the click limitation as a report's claim rather than as something we verified.

In the same move Google rolled out a control every site owner should know about. Under Settings, Search generative AI, there is a choice between including, excluding, and inheriting from the parent property. On exclusion Google states: "links to your site and your site's content won't appear in Search generative AI features", and crawled content is no longer eligible as input for an AI response. Important for the fear everyone now has: "this control isn't used as a ranking or inclusion signal affecting other parts of Search." Opting out costs generative visibility, not ordinary ranking.

The awkward part is that this setting is invisible from the outside. No checking tool in the world, ours included, can measure it. To know where it stands, you have to look in your own Search Console.

There is still no dedicated AI Mode reporting on the ads side. The available approximations are the conversational query data from AI Max and the movement in branded search. Google is testing AI Performance Insights in Merchant Center for share of voice metrics, initially in selected markets. If you know your organic starting point today, you can say later what the paid delivery moved. If you do not measure, you cannot go back and do it.

Two limits belong in the picture, because they set the expectation straight. According to Google's own help, ads in AI Overviews are currently available in English in thirteen markets, and Germany is not among them. There is no way to target that placement directly, and no reporting broken out by placement either: the clicks run in with Top Ads. The work on the feed and on the pages is still worth doing now, because it is the precondition rather than the booking.

7. What Klariton contributes

The raw material Google now reads out of the feed is the same one we surface from your material. The question and answer pairs a good feed needs are exactly the buying decision questions Buying Answers pulls out of your knowledge. The depth of product data that matters here is what Owned Intelligence builds. And whether you appear in AI answers at all is what Prompt Monitoring measures.

What we do not promise: that a well maintained feed guarantees a particular ad performance. Google's system decides that. What we deliver is the work that comes before it and holds regardless of platform: know the questions, make the product data accurate and complete, and do both so that a human and a machine can use it straight away.

Sources

• Google: Google Marketing Live 2026 collection

• Google Ads Help: See what we announced at Google Marketing Live 2026

• Search Engine Journal: How To Advertise In Google AI Mode For Ecommerce

• Google Ads Help: About Final URL expansion in Search (retrieved 27 Aug 2026)

• Google: Dynamic Search Ads are upgrading to AI Max (retrieved 27 Aug 2026)

• Google: Steer performance with new AI Max features (retrieved 27 Aug 2026)

• Google Ads Help: About ads and AI Overviews (retrieved 27 Aug 2026)

• Google Search Console Help: Search generative AI (retrieved 31 Aug 2026)

• Google: New controls for website owners (retrieved 31 Aug 2026)

• Search Engine Land on the global rollout (retrieved 31 Aug 2026)