Quick answer: Run Google Ads experiments with a 50/50 traffic split. Test one variable at a time. Run for 2-12 weeks and read results at 70-95% confidence. The slowest part is usually building creative variants. Coinis Revise (Variate) and AI Copywriting help you generate more test hypotheses faster.
---
What Is A/B Testing in Google Ads?
A/B testing in Google Ads means running a controlled experiment between your current campaign and a modified version.
Definition and purpose
You run two versions of the same campaign at the same time. One is the control. One has a single change. You measure which performs better. The result tells you whether that change is worth keeping.
How experiments isolate variables
Per Google's Ads Help Center, custom experiments split traffic between an original campaign and an experimental campaign. The design goal is simple. one change, one measurement. That's what makes the result actionable.
Why 50/50 traffic split is standard
Google recommends a 50/50 split for the most reliable comparison. Equal traffic means equal exposure to seasonality, auction dynamics, and audience behavior. Skewed splits are faster to reach but slower to reach statistical significance.
---
Which Campaign Types Support Experiments?
Not every campaign type supports experiments. Know this before you start.
Supported. Search, Display, Video, Hotel
Google Ads custom experiments work with Search, Display, Video, and Hotel campaigns. These cover most standard performance advertising use cases.
Not supported. App, Shopping
App and Shopping campaigns don't support custom experiments. Per Google Ads documentation, you'll need alternative testing methods for those campaign types.
Campaign-specific experiment types (Performance Max, Demand Gen)
Performance Max has its own experiment framework. You can run uplift experiments, which compare PMax against non-PMax campaigns, or optimization experiments that test variations within a single PMax campaign. Demand Gen campaigns have dedicated A/B experiment tools with their own setup requirements and minimum thresholds.
---
What Should You Test?
Test one thing at a time. That's the whole rule.
Creative assets and formats
Swap one image. Change one headline asset. Test one video against another. If you change multiple assets in the same experiment, you won't know which change drove the difference. Pick the variable you're most uncertain about and start there.
Bidding strategies and targeting
Maximize Conversions vs. Target CPA. Broad match vs. exact match. An audience segment added vs. removed. Each of these is a separate experiment. Running them back to back builds a clear picture of what actually moves performance.
Landing pages and ad copy
Landing page tests often produce the biggest impact. Change the headline. Test a short form vs. a long form. For ad copy, test one headline variation per experiment. Copy changes are quick to make but easy to contaminate if you change too many things at once.
Testing one variable at a time
This isn't optional. It's the core discipline. Google Ads Help Center guidance is explicit. test one variable at a time to isolate the cause of any performance change. Multiple changes in one experiment give you correlation, not causation.
---
How Long Should You Run an Experiment?
Most advertisers end experiments too early. Don't.
2-12 week minimum window
Google Ads experiments run for a minimum of 2 weeks and a maximum of 12. Demand Gen campaigns follow the same 2-12 week window. You need enough time for auction behavior to stabilize and for your data to clear short-term noise.
Factors affecting duration (campaign volume, variability, traffic)
Four factors determine how long you actually need. Campaign volume (higher spend shortens the timeline). Campaign consistency (volatile campaigns need more time to stabilize). Traffic split (50/50 reaches significance faster than 80/20). Test type (some bidding changes take longer to fully register in the auction).
Experiment Power score guidance
Google's Campaign Guidance tool includes an Experiment Power score. This score predicts how likely your experiment is to produce a statistically significant result. A low power score is a signal. extend the duration, pick a higher-volume campaign, or hold the 50/50 split.
When to end early vs. extend
End early only if results reach 95% confidence and the direction is unmistakable. Extend if the power score is low or if results are still directional but not conclusive. For conversion-based bidding in Demand Gen experiments, you need a minimum of 50 conversions per arm before the results are meaningful.
---
How to Set Up a Google Ads Experiment
The setup takes less than 10 minutes once you know the flow.
Choose your campaign and create the experiment
Go to Campaigns, then Experiments. Select the base campaign you want to test. Google automatically blocks experiments on paused or inactive campaigns, so make sure your base campaign is running.
Select your traffic split and metrics
Set your split to 50/50. Choose the metric that matters most to your goal. conversions, CPC, or ROAS. This becomes your primary success metric for the experiment.
Label your test clearly
Name your experiment so anyone can tell what changed. "Creative Test. Static vs. Video Banner - Feb 2025" beats "Test 2". Good labels cut reporting time later and make your testing history readable.
Schedule and launch
Set start and end dates within the 2-12 week window. Check for policy flags. Per Google Ads documentation, campaigns with policy-violating keywords can't copy those keywords into the experiment. Fix violations in the base campaign first. Then launch.
---
Understanding Your Experiment Results
The results table shows which version won and how confident you should be in that result.
Confidence levels (70%, 80%, 95%)
Google shows results at three confidence levels. 70% is directional. the trend is real but not conclusive. 80% is a reasonable balance for most advertisers. 95% is conclusive. Per Google Ads Help Center guidance, pick the level that matches your risk tolerance before acting on the winner.
Reading the results table
The results table compares control vs. experiment across your key metrics. Green means the experiment version is ahead. Red means it's behind. Gray means no measurable difference yet. Read the trend alongside the confidence level, not in isolation.
When results are inconclusive
A gray result after the experiment window closes is still useful data. It tells you the variable didn't produce a measurable effect. That rules out a hypothesis and saves you from chasing a dead end in future tests.
Applying the winner to your campaign
If the experiment wins, you have three options. Apply the winning settings directly to your original campaign. Convert the experiment into a new standalone campaign. Or end the experiment without action. Most advertisers apply the winner and move to the next test.
---
How Coinis Speeds Up Creative Testing
Google's experiment framework is solid. The bottleneck is usually building the creative variants.
Generate variations faster with Variate
The slowest step in A/B testing is producing the assets. Coinis Revise includes a Variate capability that generates multiple creative variations from a single base image using cutting-edge AI models. You get ready-to-test assets without waiting on a designer. Export them, load them into your Google Ads experiment, and run.
Test copy variations with AI Copywriting
Strong Google Ads headlines and descriptions take time to write well. Coinis AI Copywriting generates multiple headline, body copy, and CTA variations, all informed by your Brand Profile. Enter each experiment with more tested hypotheses. More hypotheses per quarter means more learning compounded over time.
Funnel to cross-platform testing
Coinis doesn't publish directly to Google Ads today. Direct Google Ads integration is on the roadmap. But the creative and copy engine works for any channel. Generate your Google Ads assets in Coinis, export them, and load them into your experiment. The same variations work across Meta, TikTok, and every other channel you run.
---
Or let Coinis do it.
From a product URL to a live Meta campaign. AI-generated creatives. On-brand copy. Direct publish to Facebook and Instagram. Real performance reporting. All in one platform.
Start free. Upgrade when you're ready.
15 AI tokens a month. No credit card.
Frequently Asked Questions
What is the recommended traffic split for Google Ads experiments?
Google recommends a 50/50 split between your control and experiment campaigns. Equal traffic gives you the most reliable comparison and helps you reach statistical significance faster than an uneven split.
Can I A/B test Performance Max campaigns in Google Ads?
Yes, but not with custom experiments. Performance Max has its own experiment types: uplift experiments (comparing PMax against non-PMax campaigns) and optimization experiments (testing variations within a single PMax campaign). Standard custom experiments don't apply.
How do I know when my Google Ads experiment has enough data?
Check the Experiment Power score in Campaign Guidance. Results at 95% confidence are conclusive. Results at 70-80% are directional but not final. For Demand Gen experiments with conversion-based bidding, you need at least 50 conversions per arm before the data is meaningful.
How many variables can I test at once in a Google Ads experiment?
One. Google Ads best practice is to test a single variable per experiment. Multiple changes in one experiment make it impossible to identify which change caused the performance difference. Run sequential experiments to build a clear picture.