Google’s 7% and the independent 16% never measured the same advertisers.

A media team somewhere has “AI Max migration” written on a September sprint board, under a bullet that says Dynamic Search Ads. The date is right. The campaign type is wrong, and the two mistakes cancel out just well enough that nobody notices until the wrong thing changes.

Here is the correction, from Google’s own announcement. Campaigns using ACA and the campaign-level broad match setting are auto-upgraded starting in September 2026. Dynamic Search Ads sunset and auto-upgrade begin in February 2027. [1] And the door most people believe has already closed is open: Google restored the ability to create new DSA campaigns on June 15, 2026, when it pushed the migration back, and removes it in January 2027. [3][9]

The June reversal is the detail worth reading twice. Google did not only move the date. Its developer blog tells advertisers to use the restored DSA creation to maintain a baseline while they test AI-powered alternatives. [9] That is Google describing the control group it wants you to keep.

What you are switching on, in Google’s own words

Google’s blog counts three features. Google’s help documentation counts two. [1][4] The help page names search term matching as one feature and asset optimization as the other, the second being the panel that contains text customization, formerly known as automatically created assets, together with final URL expansion. [4] The blog counts the contents of the second box as peers of the first.

That is not pedantry, because the packaging tells you what you can decline. One dependency is documented outright: final URL expansion requires text customization to be turned on, so the ad copy matches the landing page. [5] You cannot hand Google the choice of destination page while keeping your ad text under your own control.

The other structural fact is that none of this is a new campaign type. Google states plainly that AI Max is an optimization layer activated within existing Search campaigns. [4] That is why a migration can happen to you without anything in the account visibly restructuring, and why “we did not change anything” and “our targeting changed” can both be true in the same month.

Two numbers that were never about the same accounts

This is the part that gets flattened into “Google says it works, agencies say it doesn’t,” and the flattening hides the only fact that matters.

Google’s claim is an average of 7% more conversions or conversion value at a similar CPA or ROAS from the full feature suite, measured against search term matching alone. Its footnote scopes it: Google internal data, 2026, for non-Retail advertisers. [1] No methodology, no confidence interval. The one scoping fact Google does disclose is the vertical, and it is the fact that breaks the comparison everyone makes.

The most-cited independent read comes from Smarter Ecommerce, across more than 250 retail campaigns, published in March 2026 while AI Max was still in beta. [7][1] A median conversion value uplift of 13%, measured as AI Max revenue against the rest of the same campaign. Incremental conversions costing a median 16% more than the business-as-usual campaign. A spread from 42% above baseline to 35% below it. And a median ROAS difference of exactly 0%, which is the one independent number that does not embarrass Google’s clause. [10] It is one vendor’s analysis of its own retail book, and its author says the obvious thing himself: his retail data contradicts a claim Google made for non-retail campaigns. [6]

So there are two arguments, not one. The first is about price, and it cannot be settled by these two sources because they never overlap in vertical or in time. The second is about whether the extra volume was ever new, and that one has a lead rather than a rate. In two accounts smec examined, AI Max’s broad match arm landed on queries that broad match keywords already covered 49% and 63% of the time. Its author reads that as cannibalization and also names the likely cause: legacy broad-match-modified keywords that report as broad but behave like phrase, in which case the expansion is real rather than recycled. [10] Two accounts are not a benchmark. They are the question to put to your own search terms report.

Two smaller corrections, since both circulate as facts. The earlier 14% figure is not the 7% shrinking. The 14% is what Google said in May 2025 that advertisers typically see after activating AI Max at all, footnoted to non-Retail advertisers whose campaigns drew more than 70% of conversions from exact or phrase match. The 7% compares the full feature suite against search term matching alone, so AI Max is already running in both arms. [11][1] Both Google numbers exclude retail. And the single-advertiser test that gets quoted most, a four-month comparison in which AI Max returned $100.37 per conversion against $43.97 for phrase match, $52.69 for exact, $61.65 for exact close variants, and $97.67 for phrase close variants, is one account, from November 2025. [6][8] It shows what the bad end of a wide range looks like, and the Smarter Ecommerce range says that end exists. It proves nothing about your account.

What the months until February are for

The window between now and February is the last period in which a DSA campaign and an AI Max campaign can exist side by side in your account. Google restored DSA creation for exactly this reason and says so. [9] After the migration you have the new behavior and no baseline, because the thing that would have been your control has been converted into the thing you were trying to measure.

So use it as a measurement window rather than a grace period. Run the comparison at campaign level, with Google’s one-click experiments or a duplicated campaign, and keep an unconverted Search campaign as a control for as long as the calendar allows. [1] Judge it on the two questions the published data actually raises. What did the extra conversions cost, in your vertical, this year? And how much of the added query volume was already covered by keywords you had? A search terms report against your existing keyword list answers the second in an afternoon.

Set the guardrails before you switch anything on, because two of them used to be structural and are now optional. Brand exclusions and negatives matter more once matching expands past your keyword list through broad match and keywordless technology, which is the same brand search problem arriving through a different door. URL scope matters because the destination page is no longer something you chose. And the value you send back is what the system optimizes toward, so what a conversion is worth becomes the input deciding whether an extra 13% of revenue was worth a 16% higher CPA, if your account behaves like that retail book at all. If you have already lived through this pattern with Performance Max, none of the shape will surprise you.

The uncomfortable version is short. The migration is not the risk. Arriving in February with a number that was measured on someone else’s vertical, before general availability, is the risk, and that is a decision you make by doing nothing between now and then.

Sources

1. Google · Google’s Dynamic Search Ads are upgrading to AI MaxThe migration dates, the June 11, 2026, update, the April 15, 2026, exit from beta, the 7% claim, and its footnote scoping it to non-Retail advertisers.
2. Search Engine Land · Google sets AI Max migration timeline for Search campaignsAnu Adegbola, August 14, 2026. The August 3 creation block for broad match and ACA, and the September and February windows. It places the end of DSA ad group creation in February 2027; this article follows Google’s developer blog, which puts campaign creation removal in January.
3. Search Engine Land · Google delays Dynamic Search Ads migration to AI MaxJune 12, 2026. DSA creation restored from June 15, new DSA creation ending January 2027.
4. Google Ads Help · How AI Max for Search campaigns worksTwo main features, and the statement that AI Max is not a campaign type.
5. Google Ads Help · About Final URL expansion in SearchThe documented dependency on text customization.
6. PPC Land · Independent tests show AI Max underperforms traditional match typesThe Smarter Ecommerce retail caveat in Mike Ryan’s words, and the Mantica figures.
7. ALM Corp · What data from 250+ real campaigns actually revealsSecondary reporting that calls the Smarter Ecommerce analysis the most-cited third-party benchmark. It misdates that analysis to November 2025 and misattributes the 63%; the figures here come from the primary at [10].
8. LinkedIn · Xavier Mantica on four months of AI Max testingA single advertiser’s published comparison, quoted as one data point.
9. Google Ads Developer Blog · Dynamic Search Ads automigration delayed to February 2027 and campaign creation restoredJune 11, 2026. Creation restored June 15, removed January 2027, and the advice to keep a DSA baseline while testing.
10. Smarter Ecommerce · The ultimate guide to AI Max for Google SearchMike Ryan, March 23, 2026. The primary for the 13%, 16%, 0% median ROAS, the 42% to 35% spread, and the two-account 49% and 63% broad match finding with its caveat.
11. Google · AI Max for Search campaignsMay 6, 2025. The original 14% claim and its footnote: non-Retail advertisers, campaigns with more than 70% of conversions from exact or phrase match.

MS

Editor-in-chief

Mikołaj Salecki

Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.

LinkedIn ↗Facebook ↗salecki.digital ↗

More articles →