AI Max for Shopping: Google may write titles and choose landing pages

With a German-only feed there is currently practically nothing to turn on

AI Max for Shopping is a beta, and text customisation is limited to English-language feeds, as the German version of Google's help page also states. Automatic landing page selection, called final URL expansion, requires this text customisation. If you only submit German-language data in the DACH region (Germany, Austria and Switzerland), the question is postponed (as of November 2026). If you run an English feed for other markets, it already arises there. My advice in both cases: do not treat the single switch as one decision.

The one click hands over two different things

According to Google, the one-click upgrade keeps your product targeting and bidding flexibility. That settles which products you advertise and with which bid. What is said about the product and where the click leads, on the other hand, moves to Google. These are two decisions with different risks, even though Google recommends using both settings together.

The feature details come from Google's help page on AI Max for Shopping.

With titles, the control comes after the ad has run. According to Google, a generated title only serves if it is expected to perform better than the original, and it is meant to be based on your data. Google's own example turns a plain sofa title into a variant with "Made in USA", which is a claim about origin. If that is correctly stated in your data, fine. If your website carries an outdated claim about material or suitability, it becomes a claim in your ad. I assume that you, not Google, answer for it to customers and competitors; for sensitive product ranges I would have that checked by a lawyer. And 25 terms will not cover a product range. Use them for the most expensive mistakes and phrase the rest as messaging restrictions, for example "no claims about origin".

Landing pages bring two further side effects. The link to Target ROAS means that if you currently bid differently, you would have two changes in the same test, so stabilise the bidding first (Deriving target ROAS from your margin). And according to Google, campaign priority applies to Shopping formats, not to final URL expansion. So priorities cannot be used to contain the new traffic.

Hardly any shop can verify Google's 5%

According to Google, AI Max for Shopping delivers on average 5% more conversions or conversion value at a similar CPA or ROAS, based on internal data for retail worldwide. Two calculations put that into context.

First, the value, as a worked example with invented figures. Suppose a shop makes €200,000 in Shopping revenue per month at a ROAS of 5 and a 40% contribution margin before ad spend. An extra 5% is €10,000 in revenue, €4,000 in contribution margin and €2,000 in ad spend. That leaves €2,000 a month, and only if the entire increase is genuinely incremental. A nice amount, but no reason to rush.

Second, measurability. My rough calculation for comparing two arms of equal size, with conversions as counts, 95% confidence and 80% power:

Conversions per arm = 7.84 × (2 + u) / u², where u is the relative uplift.

That is the lower bound. Conversion value varies more than the plain count, and comparing two product segments is less clean than a random split. So in most shops the test will not prove a 5% uplift, but it can make damage visible. Stop criteria are therefore more valuable than a success threshold.

Segments instead of a test arm

Google's experiment with a control arm and a test arm within one campaign is described for Search campaigns. For Shopping I therefore plan with a segment comparison. If your account offers a genuine test arm for Shopping, use that.

1. Define the segments. One category or margin tier as the test, a comparable one as the control, outside your peak season. Choose segments that hardly share a category page, because such clicks later sell control products too. AI Max is controlled per campaign, so the test segment needs its own campaign. Split it off before measuring, as the restructuring disrupts bidding. smec advises testing only text customisation on a clearly defined segment first. I agree.
2. Set damage limits. Term exclusions, messaging restrictions and the URL exclusion list are in place before launch, as are the stop values below.
3. Title phase. Text customisation only. Google advises changing nothing for at least two weeks. After that I would evaluate for at least four more weeks and check the title sample against the product data every week.
4. Landing page phase. Only once the title phase has been passed. Clarify beforehand which campaign would give up the new traffic, meaning the overlap with PMax, broad match and DSA. Then open a dynamic landing page with your tracking template: according to Google, an unsuitable template can lead to 404 pages, and {product_id} stays empty for these ads. If your margin calculation depends on it, switch it over first: Google recommends {extensionid} and {adgroupid} appended to {lpurl}.
5. Evaluate at account level. Google advises this too. My reason: a single campaign can look good simply because it takes conversions away from PMax or Search. Add up the contribution margin of the test segment across all campaigns and compare how it develops with the control segment.

These signals end the test

The values are my starting values, not benchmarks. Adjust them to your risk before the test runs.

According to Google, final URL expansion can be switched off at any time. But a switch does not undo damage to your product claims, which is why the title check comes first.

Broad accounts can skip the test

The possible gain depends on what you already run. smec sees the greatest benefit in accounts with Standard Shopping and exact match, while PMax with all assets plus broad match probably already covers these search queries. I find that plausible. According to Google, on the other hand, AI Max for Shopping is designed to capture additional mid-funnel queries. I am not aware of any robust independent test data specifically for Shopping.

I would leave it out if your account is already broadly covered, if your volume would not show damage below 20% either, if incorrect product claims would be legally expensive, or if your bid strategy or tracking is being rebuilt right now. To place the campaign types, see the choice between PMax and Standard Shopping, and for the parallel in search, AI Max for Search campaigns.

The preparation pays off even without the switch

As long as German feeds are excluded, your time is best spent on what Google will later read from your data. Clean up product claims in the feed and in the shop, because both form the basis of the generated titles (Fixing product titles in the shop). Change tracking templates so that they use {lpurl} correctly: according to Google, a static URL without this placeholder sends users to the fixed address instead of the dynamic landing page. Keep one list of pages that must never become a landing page and one of claims that must never appear in an ad. Both lists will save you a scramble when the language restriction is lifted.