CPA came in 40% lower, but the search terms report is what matters
At a mattress retailer, we found an angle almost nobody was working: the workers’ compensation program covers therapeutic sleep products for people recovering from an occupational injury. With no competitor showing up in the auction, cost per click was low and the vertical was quiet.
That made it a clean place to test AI Max. I ran two campaigns with the same offer, landing page, and daily budget: one was built the traditional way around a keyword list, while the other used AI Max with brand inclusion for the program terms. I let them compete against each other for about four weeks, then consolidated the budget into the winner.
The results
| AI Max | Standard Search | |
|---|---|---|
| Conversions | 4 | 3 |
| Spend | CA$743.83 | CA$932.40 |
| CPA | CA$185.96 | CA$310.80 |
| Avg. CPC | CA$5.64 | CA$7.90 |
| Conv. rate | 3.03% | 2.54% |
| CTR | 9.83% | 9.78% |
| Impressions | 1,343 | 1,207 |
AI Max spent 20% less and brought in an extra conversion. CPA came in 40% lower.
Seven conversions total is thin, and I would rather say that up front. All of them were phone calls tracked through call tracking. What makes the test worth writing up is not the CPA gap, it is what sits underneath it.
The CTR advantage was temporary
At the four week mark, AI Max was at 13.92% CTR against 8.99% for the standard campaign. By the end, the two were separated by five hundredths of a percentage point.
The gap closed because early on AI Max was matching a small, very hot set of queries. As it spent more it widened out, and the average came back to earth. Anyone who called the test at four weeks would have reported a CTR story that does not exist.
Where the money actually went
The standard campaign spread its spend across the keyword list. AI Max did the opposite: it found the head terms and bought them cheap.
On the program’s main query alone, the standard campaign paid CA$9.92 a click and brought in one conversion. AI Max paid CA$4.46 and brought in three. The ad, account, and test window were the same.
Both campaigns were losing most of their impression share to budget rather than rank, 63% and 58% respectively. Neither one was being held back by ad quality.
The search terms report comes out cleaner
The standard objection to AI Max is automation drift, the idea that it will spend on anything vaguely related. This test showed the opposite.
Across 96 search terms, AI Max went off theme once, at zero cost. Everything else stayed inside the vertical. Meanwhile the keyword campaign was leaking into generic mattress terms with no connection to the program, things like “orthopedic mattress” and searches for a competing retail chain. Traffic related to the product, but without the intent the offer depended on.
The difference comes from the inputs each system uses. A broad match keyword campaign starts with the keyword list, which is an approximation of what people search. AI Max uses the account’s full signal set, landing page, assets, and conversion history. In a vertical where the offer is specific and the page says so plainly, that signal set describes the audience better than any list I would write by hand.
That is why I argue for this campaign type in accounts with a well defined offer. The lower CPA is a consequence. What I am actually buying is a campaign that needs less maintenance because it requires fewer reactive negative keywords every week and wastes less budget before I notice.
What holds up
I consolidated the budget into AI Max and paused the standard campaign. On seven conversions, I would not defend any thesis about CPA, but search terms quality can be observed without conversion volume. The 96 terms show that here.
The next test is on an account with the opposite shape, high intent branded queries, almost no ambiguity about who is searching. There the keyword list is clean by nature, so the advantage, if there is one, will have to show up somewhere else.
For your own AI Max test, download the PPC Drift Checklist. It gives you a way to review the signals, search terms, and controls before you decide what the result means.