Google’s Ads blog has published testing guidance that sets different confidence levels by the size of the decision: 80% for directional changes and 95% for major structural ones. The advice comes from an Ads Decoded podcast episode on how to use A/B tests to validate AI Max, budget shifts and Smart Bidding strategies.

What Google recommends

The post lists three rules. The first is to start with templated setups in the Experiments tab, which change one variable at a time and split traffic 50/50 to reduce setup errors. The second is to measure qualified conversions rather than shallow proxy metrics, and to run tests for four to six weeks to account for conversion lag. The third is to match the confidence level to the scope of the decision: the page says 80% confidence works well for directional tweaks and recommends reserving 95% certainty for major structural overhauls. It closes with: “Explore the Experiments tab in Google Ads, test with confidence, and let data prove the value before scaling.”

Why it matters

An 80% threshold accepts a higher chance that a measured difference is noise than the common 95% standard does. The page gives no statistical reasoning for the lower bar and no caveats beyond the split by decision size. A team that adopts it should write down which decisions count as directional before the test starts, so a result is not graded against whichever threshold favours it afterward.

The insight

The four-to-six-week duration matters as much as the threshold. Conversion lag means a test stopped after a week can read as a win or a loss that later reverses. A marketing leader can turn the guidance into a standing rule: one variable per test, a named qualified-conversion metric, a minimum run length, and a stated confidence level agreed before launch. Related: Wingify’s data on experiment goals and the AI Max search term parameter.

Source: Google Ads & Commerce Blog