A/B testing ads in Google Ads: how to run it
Most 'testing' in local accounts is swapping copy by gut feel with no controlled variables, which never actually answers which version performs better.
What to test first
The headline drives the biggest CTR effect of any ad element, since it's the first thing a user sees in results — test it before the description or extensions.
The second most important element is the call to action: 'submit an enquiry' versus 'get a free quote' can produce noticeably different conversion even at identical CTR, because the wording changes what the user expects from the click.
Test one variable at a time — change the headline and the description simultaneously and it's impossible to tell which edit actually moved the result.
Setting up the experiment correctly
In Google Ads, for a fair A/B test use either even ad rotation within one group or the built-in campaign experiments tool, which splits traffic and tracks statistical significance.
'Optimise for clicks' rotation shifts impressions toward an early leader before enough data accumulates, skewing the test — a fair comparison needs even rotation for the whole experiment period.
Run both variants simultaneously, not one after another — comparing an ad shown on Monday against one shown on Friday gets distorted by audience behaviour differences across days of the week.
How much data you need for a conclusion
A minimum benchmark: at least 100 clicks per variant and at least a couple of weeks of impressions before declaring a winner. Less data produces random noise that's easy to mistake for a pattern.
For conversion, not just CTR, the threshold is higher: 15-20 conversions per variant is the minimum at which the difference stops being statistical noise.
If the difference between variants falls within natural margin of error, the honest conclusion is that the test found no winner, and it's time to test more radically different variants rather than minor wording tweaks.
Common testing mistakes
Stopping a test early at the first noticeable lead by one variant is a common mistake — early results are unstable and often even out as data accumulates.
Testing variants that are too similar — swapping one word in a headline for a synonym rarely produces a statistically significant difference, and such a test just wastes time with no practical benefit.
Forgetting to apply the winning variant after the test ends — an experiment with no result implementation delivers no benefit, and some accounts run a finished test for months with the conclusion never acted on.
What to test after ad copy
Landing pages test on the same principle: two page variants with different form placement or a different headline, an even traffic split, judged on conversion rather than visits.
Bidding strategies can be compared between similar-volume campaigns over the same period, but not within a single campaign simultaneously — the system can't apply two bidding strategies in parallel to one audience.
Once several successful ad tests accumulate, build a library of proven phrasing for the niche — it speeds up launching new campaigns without retesting the basics from scratch.