How to Set Up a Google Ads A/B Test With Ad Variations

Google Ads Ad Variations let you test a change across multiple ads without manually duplicating and editing every ad. They are useful when you want to compare messaging, calls to action or landing pages while keeping campaign targeting and bidding broadly consistent.

A good Google Ads A/B test starts before you click “Create.” You need a clear hypothesis, one primary metric and enough traffic to make the comparison useful.

When to use the Ad Variation tool

Ad Variations are designed for testing changes to ads, particularly in Search campaigns. Instead of building a separate campaign, you select existing ads and tell Google Ads what to change in the variation.

Typical tests include:

• Replacing a generic benefit with a specific benefit
• Testing “Get a Quote” against “Book a Free Consultation”
• Changing promotional language across several ads
• Sending traffic to a different PPC landing page
• Testing a problem-focused message against an outcome-focused message
• Applying a consistent wording change across several campaigns

For example, a software company could replace “Easy Project Management” with “Finish Projects on Time” across its non-brand Search ads.

When a custom experiment may be better

Ad Variations are not the right tool for every Google Ads test. Consider a custom experiment if you need to test broader campaign settings, such as:

• A different bidding strategy
• New keyword targeting
• A different audience setup
• Major structural changes
• Several connected changes across the campaign

Ad Variations are best when the experimental difference is mainly in the ad or its destination URL. The exact options available can depend on campaign type and Google Ads interface updates.

Plan the Google Ads A/B test first

Write a clear hypothesis

A hypothesis states what you are changing, what you expect to happen and why.

A useful example is:

Replacing our feature-focused headline with an outcome-focused headline will increase conversion rate because it communicates the customer benefit more clearly.

This is better than “Let’s try a new headline” because it gives the test a purpose and makes the result easier to learn from.

Choose one primary metric

Pick the metric that will determine the winner before launching the test. Common choices are:

• Click-through rate: Useful when testing whether the ad earns more clicks
• Conversion rate: Useful when testing lead or purchase efficiency after the click
• Cost per conversion: Useful when the main goal is acquiring conversions economically
• Conversion value per cost: Useful for ecommerce or accounts with meaningful conversion values

Do not switch the primary metric after seeing the data. If the variation improves click-through rate but lowers conversion rate, you need the original goal to decide whether that trade-off is acceptable.

Also confirm that conversion tracking is working and that both versions use the same conversion definitions.

Estimate how much traffic you need

A short test with a handful of conversions can produce a dramatic but unreliable difference. Before launch, enter your baseline conversion rate and the smallest improvement worth detecting into the Sample Size Calculator.

The smallest worthwhile improvement is often called the minimum detectable effect. In plain English, it is the smallest change that would be valuable enough for you to act on.

Use the estimate to judge whether the selected campaigns can produce enough traffic within a practical test period. A narrow campaign receiving a few clicks per day may not be suitable for a standalone test.

How to create an Ad Variation in Google Ads

Google occasionally changes menu names and layout, but the workflow generally follows these steps.

1. Open Ad Variations

Sign in to Google Ads and open the Experiments area. Look for Ad variations, then choose the option to create a new variation.

In some account layouts, Experiments appears within the Campaigns navigation. You will need permission to edit the campaigns involved.

2. Select campaigns and ads

Choose the campaigns to include. Keep the group focused enough that the same hypothesis makes sense across every selected campaign.

You may also be able to filter the ads using conditions such as existing text, ad type or other attributes. Review the selection carefully. A wording change suitable for non-brand ads may make little sense in a brand campaign.

3. Create the variation

Choose the available variation action and make the intended change. Depending on the ads and current Google Ads options, this may include finding and replacing text, updating ad text or changing a URL.

Make one meaningful change where possible. If you rewrite every headline, change the call to action and send users to a new landing page simultaneously, you will not know which change caused the result.

Responsive search ads create an extra complication because Google can combine and serve assets in different ways. If you replace one asset, its exposure may vary between auctions. Review asset combinations and avoid assuming every impression displayed the changed headline in the same position.

Pinning an asset forces it toward a particular position, but pinning also changes how the responsive ad can serve. Do not add new pinning solely to make the test look cleaner unless that is intentionally part of the hypothesis.

4. Set the experiment split

Choose what percentage of traffic should receive the variation. A 50/50 split is usually the simplest option because it gathers data for both versions at a similar rate.

A smaller variation share limits exposure to the new version, but it also means the test will take longer. That can be reasonable for a high-risk claim or landing-page change, but it should be a deliberate trade-off.

5. Choose start and end dates

Give the test enough time to cover normal changes in demand. Avoid judging it after one unusually strong day or during a promotion that affects only part of the test.

Your schedule should account for:

• Expected traffic and conversion volume
• Weekday and weekend behaviour
• Sales events and promotional periods
• Conversion delay between an ad click and a recorded lead or sale
• Material budget, targeting or website changes

Avoid making unrelated edits to the selected campaigns while the test runs. If you must make a major change, document it and consider restarting the experiment.

6. Name and launch the variation

Use a name that records the hypothesis, changed element and launch period. For example:

Outcome headline vs feature headline — non-brand — May

Review the campaign selection, variation, traffic split and schedule, then create the experiment.

Worked example: testing an outcome-focused headline

Suppose a B2B advertiser tests these messages:

• Original: “Simple Invoice Software”
• Variation: “Get Invoices Paid Faster”

The hypothesis is that the outcome-focused headline will attract more qualified visitors. The primary metric is landing-page conversion rate, not click-through rate.

After allowing time for delayed conversions, the report shows:

Version	Impressions	Clicks	Conversions
Original	20,000	1,100	88
Variation	20,000	1,230	111

The original click-to-conversion rate is 88 ÷ 1,100, or 8%. The variation's rate is 111 ÷ 1,230, or about 9.0%.

Because landing-page conversion rate is the primary metric, use clicks as visitors and conversions as successes in the A/B Test Results Calculator:

• Original visitors: 1,100
• Original conversions: 88
• Variation visitors: 1,230
• Variation conversions: 111

If the primary metric were click-through rate, you would instead enter impressions as visitors and clicks as conversions. Never combine impressions for one version with post-click conversions for the other calculation; the numerator and denominator must describe the same funnel step.

The calculator can estimate the probability that one version is better and the likely conversion-rate range. That does not replace commercial judgment. Before rolling out the variation, check cost per conversion, lead quality and whether the observed improvement is large enough to matter.

How to read the results responsibly

Wait for conversion delay

Someone may click today and convert several days later. If you read the report immediately after ending the variation, recent traffic can appear worse simply because its conversions have not arrived yet.

Wait for a period that reflects your normal conversion delay before making the final decision. Use the same attribution and conversion settings for both versions.

Separate evidence from business value

A variation can probably be better but still deliver an improvement too small to justify implementation. It can also increase conversion volume while producing lower-quality leads.

Review:

• Probability that the variation is better
• Expected size of the improvement
• Cost per conversion
• Revenue or lead quality, where available
• Performance across strategically important segments

Treat device, location and audience breakdowns as supporting analysis unless you planned those comparisons in advance. Searching dozens of segments for a winner makes random patterns easier to mistake for real effects.

Common Google Ads A/B testing mistakes

Avoid these frequent problems:

• Testing too many changes at once: You cannot identify what caused the result.
• Stopping after an early lead: Random variation is strongest when data is limited.
• Running overlapping experiments: The same traffic can be influenced by multiple changes.
• Changing budgets or targeting mid-test: This makes the comparison harder to interpret.
• Ignoring responsive ad behaviour: A changed asset may not appear in every impression.
• Choosing a winner on clicks alone: More clicks are not automatically more customers.
• Ignoring conversion lag: Late conversions can change the apparent result.
• Testing a trivial difference: A result can be measurable without being commercially worthwhile.

The goal is not to declare as many winners as possible. It is to learn which message improves a business outcome under a fair comparison.