You can't run a clean A/B test inside a responsive search ad
Google recombines your headlines before anyone sees them. A one-variable matrix inside a single RSA isn't attributable to anything.
Responsive search ads take up to 15 headlines and 4 descriptions and assemble combinations at auction time. That's the feature. It's also why the testing habit people bring from Meta produces numbers that can't be read.
The problem with a one-axis matrix
On Meta you can hold everything constant and change one line, because the ad that serves is the ad you built. In an RSA, the headline you're testing appears alongside a different second headline each time. When performance moves, you can't attribute it to the line you changed.
What to do instead
- Test at the ad level, not the asset level. Two RSAs with genuinely different angles beat fifteen headlines in one.
- Use asset performance ratings for what they are: a signal about which assets are being chosen, not a controlled result.
- Pin sparingly. Pinning restores control and gives up most of the reason to use the format.
Negative keywords are the other half
Search is the one place where deciding who not to reach matters as much as the targeting. Negatives don't close-match, so plurals, misspellings and variants each need their own entry. A negative for "free" won't block "freely".
AdPlaybook generates negative keywords rather than leaving them to you, and only for platforms that actually have search terms to exclude. See the Google Ads specs.