Google Ads now shows advertisers a confidence estimate before an experiment launches, not after it fails. The feature, called Campaign Guidance, surfaces an Experiment Power score that predicts whether a test is likely to reach statistical significance given its current setup. It is live for every Performance Max experiment, and for Broad Match experiments running inside Google Search campaigns.
The problem it addresses is common and expensive. Most account-level A/B tests fail on power long before they produce an actionable read. Advertisers run tests for two weeks on low-volume campaigns with a 90/10 traffic split, then treat inconclusive results as evidence that a change did not work. The underlying cause is almost never the change itself. It is an underpowered experiment that never had a realistic chance of reaching significance.
Experiment Power scores fall into three bands: low (0 to 49 percent), medium (50 to 79 percent), and high (80 to 99 percent). Google’s own confidence framework generates these estimates inside the experiment setup workflow, giving advertisers a pre-flight check rather than a post-hoc regret.
Four factors drive the score:
- Campaign selection: campaigns need sufficient historical volume for a reliable baseline.
- Campaign variability: consistent performance over time reduces noise in the results.
- Traffic split: a 50/50 split between control and treatment arms accelerates result confidence.
- Experiment duration: minimum run time scales with campaign volume and the type of change being tested.
The last two factors carry a cost that Google’s framing underplays. A 50/50 traffic split and a longer minimum duration mean more budget committed before the experiment delivers a trustworthy read. A high power score is not free. It is purchased with additional spend held in the test arm and additional time before the winner can be deployed at scale. Advertisers running experiments on campaigns with tight budgets or short planning cycles will find that reaching an 80 to 99 percent power band requires real tradeoffs, not just configuration changes.
Personalized recommendations accompany the score. Google will suggest increasing experiment duration, shifting to a 50/50 split, raising budgets, or substituting a higher-volume campaign. These are structural fixes, not surface-level prompts. An advertiser who acts on them is running a materially different test than the one they configured first.
The feature was first spotted by Hana Kobzova at PPC News Feed and subsequently documented in Google Ads Help. Search Engine Roundtable reported it on June 9, 2026. Google frames the benefit as helping advertisers sidestep inconclusive reads and overlong test windows, though those two outcomes sit in tension: avoiding inconclusiveness requires longer durations and more volume, not shorter ones.
For practitioners, the immediate action is to open any active or queued experiment and check the power score before the test goes live. A low score on a PMax or Broad Match experiment is not a reason to cancel the test. It is a signal to extend the run window, rebalance the split, or move the experiment to a higher-volume campaign if the budget allows. Treating a low-power result as valid data has always been the more common and costly mistake. This feature makes the diagnosis explicit before the spend is made.
Reported by Search Engine Roundtable (seroundtable.com), authored by Barry Schwartz, published June 9, 2026, based on original spotting by Hana Kobzova via PPC News Feed and documentation in Google Ads Help.
