AI Max for Search: what the data says, not the marketing
Google promises +7 % conversions; independent measurement shows more volume at a higher cost per conversion. What AI Max actually does, who it suits, and how to switch it on without the budget running away.
- Google reports an average +7 % conversions with the full feature set at a similar cost per conversion; earlier beta figures were higher.
- An independent analysis of 250+ campaigns found median revenue +13 % but median CPA +16 %, with only 22 % of campaigns hitting their original ROAS target.
- Below roughly 30 conversions a month it behaves erratically. Without negative keywords and search term checks it can get expensive fast.
AI Max for Search is a set of features that adds looser query matching, automatic ad text adjustment and final URL expansion to a classic search campaign. Google is also folding Dynamic Search Ads into it: campaigns with dynamic ad groups move to standard ad groups, with settings and data carried over into AI Max.
For an advertiser that means two things. First, for some accounts the “switch it on or not” decision will be made for them. Second, the centre of gravity of the work shifts: instead of keywords, you manage guardrails.
What AI Max actually does
- Looser query matching. The campaign shows on queries you do not have as keywords, but which the system judges relevant to your pages and assets.
- Ad text adjustment. The system generates and adapts headlines and descriptions to the specific query. This can be steered with text guidelines — rules about what it may and may not say.
- Final URL expansion. Instead of the page you specified, it can send the visitor to another page on your site that it considers more suitable.
Each of those three can be enabled separately. That matters, because they differ enormously in risk.
What Google says and what others say
Platform figures: AI Max campaigns using the full feature set get, according to Google, about 7 % more conversions or conversion value at a similar CPA or ROAS compared with query matching alone. Earlier communication carried higher numbers — +14 % from the May 2025 beta against keyword-match-only campaigns, and +27 % in February 2026 for exact and phrase-heavy campaigns using text guidelines.
They all come from Google, they are averages across very different accounts, and each compares something different. “+7 % against query matching alone” is not the same as “+7 % against your current campaign”. For a decision they are indicative, not a promise.
Independent measurement looks more mixed. An analysis by the agency SMEC covering more than 250 campaigns reports median revenue up 13 % but median CPA up 16 %, with only 22 % of campaigns reaching their original ROAS target. In other words: more volume, dearer. For some accounts that is a good trade, for others not — and what decides is whether the added volume earns at your margin. That is precisely the marginal CPA question.
There are markedly worse individual cases too, such as a B2B account where cost per lead rose from 493 to 850 USD even though clicks tripled. That is not an argument against AI Max as such; it is an illustration of what happens when looser matching is let into an account with no lead qualification and no negative keywords.
When it makes sense and when it does not
| Situation | Recommendation | Why |
|---|---|---|
| Online shop, 100+ conversions a month, broad range | Test, gradually | Enough data to learn and to evaluate; many queries outside your keywords |
| Account below ~30 conversions a month | Probably not | Analyses agree results are unstable below this threshold |
| B2B with expensive, poorly qualified leads | Only with firm guardrails | Looser matching brings volume of enquiries sales will not take |
| Regulated field (health, finance, law) | Carefully, without text adjustment | Automatically generated wording can breach sector rules |
| Account still on Dynamic Search Ads | Prepare for the transition | Google is folding DSA into AI Max; moving deliberately beats being moved |
How to switch it on without pain
- Do not enable everything at once. Start with looser query matching and leave URL expansion off. URL expansion is the riskiest feature, because it decides where the customer lands.
- Deploy it as an experiment, not across the board. Google Ads lets you run an experiment against the existing campaign. Otherwise you will never know what AI Max caused and what the season did.
- Have negative keywords ready first. Shared lists covering competitors, informational queries and purchase signals that do not work for you. Accounts that launched AI Max on defaults without negatives report the fastest rise in CPA.
- Read the search terms report weekly. Look for two things: nonsense to exclude, and queries converting at a good price that deserve their own campaign and page.
- Set rules for text adjustment. What must not be claimed, which words to avoid, how to write the brand. It is cheaper than dealing with a complaint about wording you did not write.
- Evaluate on marginal CPA, not volume. More conversions at a higher price can be a win or a loss. Margin decides.
What it changes about the work
A gradual shift that has been under way for years and which AI Max only accelerates: keywords are ceasing to be a control and becoming a signal. The work moves elsewhere — into the quality of landing pages (because the system infers relevance from them), into negative keywords, and into the conversion data the algorithm learns from. Covered in detail in account structure in 2026.
What to take away
AI Max is a volume tool that charges for volume. In accounts with enough conversions, a healthy margin and disciplined negative keywords it often pays. In small accounts, in expensive B2B lead generation, and anywhere nobody reads the search terms report, it reliably gets expensive. Do not decide on someone else’s numbers — run an experiment and measure it.
Common questions
What is AI Max for Search?
A set of features in Google Ads that adds looser query matching beyond your keywords, automatic adaptation of ad text to the specific query, and expansion of the final URL to another page on your site. Google is also folding the previous Dynamic Search Ads into AI Max. The individual features can be enabled separately, which matters because they differ considerably in risk.
Does AI Max work, or does it just get expensive?
Both, depending on the account. Google reports an average of seven per cent more conversions at a similar cost per conversion with the full feature set. An independent analysis of more than 250 campaigns by the agency SMEC, however, found median revenue up thirteen per cent with median cost per conversion up sixteen per cent, and only twenty-two per cent of campaigns reaching their original return target. What decides is therefore not the average but whether the added volume earns at your margin — and you only learn that from an experiment on your own account.
From what conversion volume does AI Max make sense?
Sources agree that below roughly thirty conversions a month results are unstable, because the system lacks data to learn from and you lack data to evaluate. Above that threshold the chance of a meaningful outcome rises, but it still holds that it should be launched as an experiment against the existing campaign, with negative keywords prepared in advance and a weekly review of the search terms report.
- We are upgrading Dynamic Search Ads to AI Max, GoogleGoogle’s primary announcement about moving DSA into AI Max and the source of the average +7 % conversions figure with the full feature set. A platform figure, not independent measurement.
- Google Ads AI Max For Search In 2026: Setup, Real Performance Data, And When To Actually Use It, GroasSecond-hand source for the SMEC analysis of 250+ campaigns (median revenue +13 %, median CPA +16 %, 22 % on target ROAS) and the B2B case with CPL rising from 493 to 850 USD. An agency blog; we did not verify the original analysis in primary form.
- Google AI MAX for Search Campaigns: Implementation Checklist and Performance Analysis, ALM CorpSource of the guidance on the roughly 30 conversions a month threshold and of the setup recommendations. A vendor blog.