Smec’s Mike Ryan pulled 383 million ecommerce Search impressions from accounts running AI Max. He checked a single number: how much of the traffic billed to exact match keywords was still, in fact, exact. His finding, posted to LinkedIn in late July, put the answer at just under three quarters. AI Max is the Google feature that expands ad delivery beyond a literal keyword list. It now supplies close to 29 percent of what shows up in exact match reporting, up from nothing at the start of last year. Exact match exists so an advertiser can decide precisely which queries trigger an ad. Ryan’s chart shows that decision being overridden more often every month, and the pace has quickened sharply since spring.
This publication reported on August 7 that the shift will soon stop being optional. Google plans to switch AI Max on automatically starting September 1 for Search campaigns already using broad match together with automatically created assets, with no advertiser action required. Ryan’s sample only covers accounts that made that choice themselves, and his data collection ends before the deadline. Whatever the Smec chart shows for opted-in advertisers today becomes the default condition for a much larger set of accounts once the automatic switch lands.
Google has published four different performance numbers for AI Max within about fourteen months, each measured a different way. The feature launched in beta in May 2025 carrying a claimed 14 percent lift in conversions. Eleven months later, at general availability, the company lowered that figure to roughly 7 percent. That number covered the whole feature bundle rather than AI Max alone, and a footnote excluded retail advertisers entirely. A month after that, Google told advertisers running exact and phrase match heavy accounts that switching to AI Max produced 27 percent more conversions without hurting CPA or ROAS. Then, on Alphabet’s July 22 earnings call, Chief Business Officer Philipp Schindler offered a fourth framing: an average 15 percent gain in conversions or value across AI powered campaigns generally. Google’s Search ad revenue climbed 17 percent to $63.3 billion on that same call. Every one of those figures originates inside Google.
Two measurements from outside Google point at a gap between that framing and account-level results. A separate piece of Smec’s own research, published in November 2025 and covering upward of 250 retail campaigns, is the first. It put AI Max’s return on ad spend roughly 35 percent under what traditional match types delivered within those same accounts. Conversions cost more, and average order sizes came in lower. A second test, conducted in August 2025 and spanning about 30,000 individual search terms under AI Max, found that 99 out of every 100 impressions produced no conversion at all. Spend also leaked onto Search Partner Network inventory. Neither result comes from Google. Neither has been countered publicly with independent numbers of similar scale.
Verifying either side is hard because of how the platform reports the traffic. Google acknowledged in December 2025 that AI Max’s search term matching works from inferred intent, not the literal text of a query. It also confirmed that when an account has no broad match keyword to catch that traffic, the system logs it against an exact or phrase match keyword instead. A single row in a standard performance report can therefore blend two products with different economics. Nothing in the interface currently separates them.
Some advertisers have already built around the problem rather than waiting for Google to fix the reporting. Martin Große, who heads paid search and programmatic work for Suchmeisterei GmbH, keeps AI Max out of his exact-match-heavy accounts for that reason. New traffic goes into a campaign built specifically to hold it. Once a budget starts flowing toward unrequested queries inside an existing campaign, he has found it difficult to claw back.
Any team relying on exact match for query control should run Ryan’s audit on its own account before September 1. Pull the search term reports, compare them against the exact match line item, and measure how much of that “exact” traffic is already algorithmic broadening. Wait for the automatic switch, and the audit happens after the baseline has already shifted, not before.
Luis Rijo reported this analysis for PPC Land on August 9, 2026, based on Mike Ryan’s LinkedIn post for Smarter Ecommerce.
