AI Max for Search campaigns: what an online shop should check

The question is not whether AI Max brings revenue, but where from

AI Max almost always reports additional conversions. For an online shop that is the wrong yardstick, because much of the demand that AI Max picks up on top is already served by Shopping, Performance Max and your own exact keywords. My assessment: in a shop with a broad Shopping set-up, part of the AI Max revenue reported is relabelled, and neither the search terms report nor the campaign experiment shows you which part. I would therefore switch on AI Max only as a test, with controls in place and evaluated on the account total.

According to Google, AI Max is not a separate campaign type but an optimisation layer within existing Search campaigns. That has a practical consequence: in new Search campaigns, AI Max is on by default, and so are text customisation and final URL expansion. If you create a campaign and do not make an active decision, you have already started the test, just without any structure.

What AI Max does differently from broad match

Broad match extends a keyword to related queries. AI Max goes further: according to Google, search term matching uses broad match plus asset-based and landing page-based technology. The campaign therefore also finds queries for which no keyword exists at all, derived from ad text and shop pages. Two further functions come on top that have nothing to do with reach: Google rewrites ad text and chooses the landing page itself.

One rule from the matching logic matters for the evaluation: only an exact keyword identical to the search query takes precedence. Phrase, broad and AI Max share the next level, after which relevance and ad rank decide. An exact keyword therefore does not protect a query if it is merely similar, such as a plural or a query with an extra word. AI Max can take over these queries and report them as its own hit. The practitioner Brad Geddes has criticised exactly this: AI Max could bypass existing exact and phrase keywords and book their conversions for itself. Google explains such cases with inferred intent rather than literal wording.

Which controls a shop sets before the start

According to Google, AI Max needs conversion-based Smart Bidding. That means the bid strategy decides how much the additional queries may cost. If your conversion value is revenue rather than margin, AI Max optimises for revenue, even if the new queries bring smaller baskets. It is worth reading margin as conversion value first.

After that, in this order:

1. Brand exclusion. It applies at campaign level. Generic campaigns should exclude your own brand, otherwise AI Max books brand demand as a generic success.
2. Decide on final URL expansion deliberately. If it is on, pinned ad text is no longer taken into account. If you pin mandatory legal information, prices or promotion notices, you lose that control. I would switch it off in the first test and check reach separately from landing pages.
3. URL exclusions, if the expansion is on. They apply only at campaign level and require text customisation and final URL expansion. Typical candidates in a shop: filter pages with a thin range, customer account, basket, guides and blog, shipping and returns pages, outlet with poor margin.
4. Search term matching per ad group. It can be switched off in the ad group settings. Sensible for groups with narrowly defined products, where any expansion is more likely scatter.
5. Check your negative lists. More reach means more scatter. Reconcile your lists before the expansion runs; see negative keywords for shops.

Brand inclusions work in the opposite direction: they restrict a campaign or ad group to certain brands, such as manufacturer brands in your range. An inclusion at ad group level overrides the campaign one.

How to read the search terms report

In the search terms report, "AI Max" appears as its own match type, and a source column shows whether the hit comes from broad match expansion or keywordless matching. You see the landing pages chosen in the landing page report, in the "Selected by" column.

Read the report with the precedence rule in mind. For every AI Max term with revenue, ask two questions:

• Would an existing keyword have received the query? If the term is a close variant of one of your keywords, the revenue has probably been shifted, not created.
• Would Shopping or PMax have served the query? For product and category queries that is the rule in a shop. Here you often just move the click from Shopping to the text ad.

What is new is whatever neither question answers with yes: queries about problems, uses or terms that are missing from your keyword set. You are most likely to find them under the source "keywordless". One limitation applies as well: for AI Mode, AI Overviews, Lens and autocomplete, the search term shown can, according to Google, be an interpretation of the intent rather than the query as typed. The report is then a reading, not a log. How to evaluate it commercially otherwise is covered in evaluating the search terms report commercially.

What the available data suggests

Two analyses point in the same direction. Smarter Ecommerce analysed more than 250 Search campaigns: a median of 13 percent more conversion value, but the additional conversions were more expensive than keyword traffic, and only 22 percent of campaigns were close to the original ROAS target. In its clients' generic campaigns, one agency found around 11 percent of cost on AI Max terms but only around 5 percent of conversion value, at roughly the same CPA. That means the same number of orders per euro, but smaller baskets.

Both figures apply to other people's accounts and cannot be transferred to your shop. They also measure at campaign level. Whether the extra revenue was taken away from Shopping or PMax, they do not answer.

How to test: the account total decides

Google offers AI Max experiments within a campaign, with a control arm and a test arm. Excluded are, among other things, portfolio bid strategies, shared budgets and already active text customisation. The experiment is cleaner than a before-and-after comparison, but it has a blind spot: Shopping and PMax serve the same demand in both arms. If the AI Max arm wins queries that Shopping would otherwise have received, that looks like growth in the experiment.

You therefore also evaluate the sum of generic Search, Shopping and non-brand PMax, over the test period against a comparison period, and convert it to contribution margin. The rule:

Additional orders needed in the account total = extra cost ÷ contribution margin per order before ad spend

A worked example, assumed: in four weeks the AI Max campaign reports 100 additional orders for €4,000 of extra cost. The contribution margin per order before ad spend is €35. At least 115 additional orders would be needed in the account total. If the total rises by only 30 orders, because 70 came from Shopping and exact keywords, the test brings €1,050 of contribution margin against €4,000 of cost. The campaign looks good, and the shop loses around €2,950. If the new queries bring smaller baskets, the contribution margin per order falls and the threshold rises further.

The logic behind this is explained in measuring incrementality in Google Ads.

When this check is not enough

For small accounts, volume is often too low. If your account total fluctuates from week to week by more than the expected effect, you cannot separate growth from noise. In that case I advise a longer test period, or leaving AI Max out for now and extending keyword coverage yourself. Seasonal peaks such as Black Week are unsuitable as a test period.

Conversely, AI Max is most likely to be new demand where your Shopping set-up is weak: for ranges that are searched for by problems or uses rather than product names, or for products that hardly run in Shopping. That is where I would start the test. Status of the features: November 2026.