Why Brands Should Test Turning Down Google Ads, According to Rand Fishkin

In a recent LinkedIn post, Rand Fishkin discusses why brands should test reducing paid advertising to determine whether campaigns generate incremental sales. His post points to an experiment involving Performance Max (PMAX) campaigns and argues that platform-reported attribution can overstate the sales impact of ads. The following article reports on Fishkin’s claims and analysis; it is coverage of his post, not an article written by him.

“This is why, as we urge in Zero Click Marketing, you have to TEST TURNING DOWN YOUR ADS.”

An experiment to measure incremental sales

Fishkin cites a campaign experiment by Craig Graham. According to the details in Fishkin’s post, PMAX was paused in 26 U.S. states for 28 days while remaining active in 22 other states. The comparison was intended to estimate what sales might have occurred without the ads, rather than relying solely on the platform’s attribution reports.

Fishkin says sales in the paused states came in about 18% below the predicted level, equivalent to 1.17 times the net dollars not spent on advertising. He contrasts that result with a break-even figure of 2.22 times and Google Ads’ reported return of 2.3 to 2.8 times. On his reading of the experiment, the measured return was roughly 40% to 55% of the platform-reported figure.

“Google Ads will take credit for sales that would have happened anyway. PMAX perhaps even more than others.”

Attribution and incrementality are different measures

Fishkin’s central point is that an attributed sale is not necessarily a sale caused by an ad. A platform may report conversions associated with a campaign, while an incrementality test asks whether those outcomes would have happened without the campaign. He argues that the distinction matters when marketers assess whether a spend is producing additional business.

“Brands think they’re getting attribution when they buy performance ads,” Fishkin writes, but he says the reported estimate can diverge substantially from true incrementality. The figures in his post are from one campaign experiment; they illustrate his concern, though the post does not provide further methodological detail with which to assess how broadly the result applies.

“But what they’re really getting is an estimate that’s very different from the true incrementality numbers.”

Human judgment in automated campaigns

Fishkin also connects the measurement issue to the expanding use of AI in marketing and advertising. In his view, agencies and consultants will remain valuable because campaign automation does not remove the need to test performance claims and interpret results.

He cautions marketers against relying on automated systems without independent evaluation. The post’s practical recommendation is to run controlled tests, such as pausing ads in selected markets, and compare observed outcomes with platform reports. Fishkin presents that approach as a way to distinguish attributed performance from incremental sales—and to make budget decisions with a clearer view of what advertising is contributing.