Google’s blog cites internal 2026 data, from non-retail advertisers: AI Max campaigns with the full set of features have on average 7% more conversions or conversion value, with similar CPA or ROAS, compared with search term matching alone. It’s a vendor figure, with no public sample or methodology. To decide in your account, test with experiments and measure real cost per result.
What Google claims, and what it doesn't
The claim is: on average, +7% conversions or conversion value, with similar CPA or ROAS, for AI Max campaigns with all features, compared with search term matching alone, in internal 2026 data from non-retail advertisers.
It doesn’t claim: that it’s the typical result, that it applies to online shops, that it applies to Portugal, that it applies to small accounts, nor the period, sample or dispersion. “On average” hides variation. “Non-retail” excludes precisely the type of business with feeds and tight margins, where many Portuguese SMEs are.
Six questions for any vendor figure
1. Who measured? (here, Google itself). 2. With what sample and period? (not disclosed). 3. Against what? (term matching alone, not against campaigns without AI). 4. Which metric? (conversions or value, which mixes different things). 5. Who does it apply to? (non-retail). 6. What dispersion? (not indicated).
A figure that doesn’t answer these questions is a clue, not proof. It doesn’t mean it’s false: it means it’s not enough to decide.
AI Max according to Google (internal data, 2026)
+7%
conversions or value, on average, with similar CPA or ROAS
Non-retail advertisers, AI Max with all features versus term matching alone. No public sample or methodology.
What “similar” means
“Similar CPA or ROAS” means cost per result didn’t worsen relevantly. But similar can be 3% more or 3% less, and a 7% rise in volume with a 5% rise in cost per result may not pay off, depending on margin.
So what matters to your business is profit: more conversions at a slightly higher cost per conversion may be good if margin is high and bad if low. Always calculate with your maximum cost per customer.
How to test in your account, rigorously
Use Google Ads experiments to create two versions of the same campaign: one with AI Max and one without (or with fewer features), with the same budget, the same audience and the same period. Define beforehand the metric (cost per qualified lead, cost per purchase) and duration, preferably one to two conversion cycles.
If the account has few conversions, the test may take weeks or be inconclusive. In that case, the decision may be for control: keep the version that requires less review work and has less risk of surprises.
A figure that only shows the good side is advertising.
Measuring quality, not just volume
More conversions isn’t necessarily better if they are of lower quality. In lead businesses, ask sales for feedback: of leads from AI Max, how many are qualified? In shops, look at margin and returns.
Search term matching may bring conversions on broader searches, with less intent. If close rate falls, the 7% volume may hide a loss of value. Measure through to sale whenever possible.
When to trust a vendor figure more
When it’s confirmed by independent third parties, with public methodology and large samples; when it’s consistent with what you see in your account; and when the vendor also presents cases where the feature didn’t help. A figure showing only the good side is advertising.
The presence of practitioner criticism, such as that cited by the Meta report for the +22% ROAS, is a healthy sign of scrutiny. Read it, even when you disagree.
What to do in Portugal this week
Choose a Search campaign with reasonable volume, create an experiment, define metric and duration and record them. Write text guidelines and brand and location controls before switching on. Schedule a review in two weeks.
If you prefer not to test yet, that’s fine: record the decision and the reason. The worst scenario is letting automation decide for you, without knowing what changed. If you want help designing the test, I can do it with you.
Summary in three lines
The +7% is a clue from Google, not a guarantee for you. Test with experiments, measure real quality and cost per result, and decide for your business. If a supplier presents this number as proof that you “must activate AI Max”, ask them for the test in your account.
Test checklist
Written hypothesis; two versions identical except for AI Max; same budget and audience; metric and duration defined; record of dates; confirmation with real sales or sales feedback; decision by written criterion; review in two weeks. With these eight points, the test is defensible and repeatable, and the decision stops being an opinion.
One final question before deciding
Before switching on a feature because of a vendor figure, ask: if the result were zero, what would we lose? If the answer is little, run a small test with a written criterion. If it is a lot, ask for more proof and protect the account with limits.
A figure is an invitation to test, not a decision made.
Frequently asked questions
Can I expect +7% conversions?
No. It’s a Google internal average, from non-retail advertisers, with no public methodology. Your result may be higher, lower or nil.
Does the figure apply to online shops?
Google refers to non-retail advertisers. For shops, test with your feed and your margin.
Should I switch on all the features?
Google recommends it. A prudent approach is to switch them on one by one, with measurement, and choose by real cost per result.