PPC Course
Google Ads Experiments: How to A/B Test Ads and Landing Pages
By Daniel | PPC Performance Marketing Specialist
How to run Google Ads experiments and landing page A/B tests: pick the right test, set up splits, choose a metric, decide duration and read results correctly.
Updated 31 October 2026 · Part of the PPC guide series
Short answer: A PPC A/B test compares a control (your current setup) with one change (the treatment) on a split of the same traffic, then keeps whichever performs better on one agreed metric. In Google Ads, use the Experiments page: custom experiments for bidding, keyword, audience or landing page changes, and ad variations for text changes across many ads. Test one variable at a time, run long enough to collect meaningful data, and avoid editing the original campaign during the test.
Why test instead of just changing things
Performance moves for many reasons: seasonality, competitors, auctions, budgets. If you change a bid strategy and conversions rise, you cannot know if the change caused it. A split test runs both versions at the same time on comparable traffic, so the difference is much more likely to come from your change.
Pick the right type of test
| You want to test | Use |
|---|---|
| Bid strategy, keywords, match types, audiences, final URL (landing page) | Custom experiment (Search, Display, Video, Hotel) |
| Headline or description wording across many ads | Ad variations / text optimization experiments |
| Adding Performance Max, or PMax vs Standard Shopping | Performance Max experiments |
| Video creative | Video experiments |
| Landing page layout or copy (not URL-level) | A/B testing tool on your site, or two URLs via a custom experiment |
| Several ads in one ad group | Rotate ads evenly (informal; less controlled than an experiment) |
Step 1: Write a hypothesis
A useful hypothesis names the change, the expected effect and the reason:
"If we switch from Maximize Clicks to Target CPA, cost per lead will fall because bids will focus on searches likely to convert."
Choose one success metric before you start (for example cost per conversion or conversion value). Google's best-practice guide recommends letting that metric decide the winner.
Step 2: Set up a custom experiment
- Open Campaigns → Experiments and create a custom experiment.
- Select the original (base) campaign.
- Make the single change in the treatment.
- Set the split, often 50/50, so both arms get comparable traffic and budget.
- For Search, choose the split type:
- Search-based split: each search is randomly assigned, so one person may see both versions.
- Cookie-based split: each user sees only one version. Better for tests where repeated exposure matters (for example audiences). If you use audience lists with a cookie split, Google suggests at least 10,000 users in the list, or results may be less accurate.
- Choose how to judge results: directional, or require statistical significance.
- Set start date and duration, then schedule.
Display experiments always use cookie-based splits.
Step 3: Leave it alone
- Do not change the original campaign during the test unless sync is turned on, or you will not know which change caused the result.
- Avoid launching other big changes (budgets, new campaigns) in the same period.
- Expect a learning period when testing Smart Bidding changes; judge results after it settles.
How long to run a test
There is no universal number. A test needs enough conversions in each arm to tell a real difference from random noise. In practice:
- Low-volume accounts need longer tests or bigger changes to detect.
- Run for at least one full weekly cycle, and ideally several, so weekdays and weekends are both included.
- Account for conversion delay: if customers take days to convert after clicking, include that lag before judging.
Use the significance indicator in Google Ads, or an A/B test calculator, rather than stopping as soon as one arm looks ahead.
Worked Example: Why Small Differences Can Mislead
A landing page test sends 1,000 clicks to each version (illustrative numbers).
| Clicks | Conversions | Conversion rate | |
|---|---|---|---|
| Control | 1,000 | 40 | 4.0% |
| Treatment | 1,000 | 50 | 5.0% |
The treatment looks 25% better, but a gap of 10 conversions on samples this small can easily happen by chance. A significance calculator would not call this a reliable win, and the right move is to keep the test running until each arm has more conversions. If the same 4.0% vs 5.0% gap held at 10,000 clicks per arm (400 vs 500 conversions), the evidence would be far stronger. Always decide how you will judge a test before you see the numbers.
Step 4: Read and apply results
- Look at the success metric first, then check side effects (volume, impression share, conversion value).
- If the treatment wins with confidence, apply it. Applying to the original campaign keeps its history; converting to a new campaign suits structural tests.
- If results are inconclusive, that is still a learning: the change did not matter enough to detect. Record it and move on.
Testing ads
- Test one message idea at a time (price vs speed, benefit vs proof).
- Use ad variations to change a phrase across many ads at once.
- Inside a responsive search ad, Google combines assets automatically, so a classic one-headline A/B test is limited; compare messages at ad or campaign level instead. See our guide to responsive search ads and Ad Strength.
Testing landing pages
Two ways:
- URL-level test in Google Ads: a custom experiment where the treatment uses a different final URL. Good for completely different pages.
- On-page A/B tool: splits visitors on the same URL and swaps elements. Good for headline, form or layout tests, and works for all traffic sources.
Ideas worth testing: headline matching the ad more closely, shorter forms, adding proof near the call to action, different offers. For what makes a strong page in the first place, see PPC landing page optimization.
Check the Page Before You Test It
A test of a slow or broken page only compares 2 problems. Run both versions through a free SEO audit first to compare load time, mobile friendliness and basic on-page issues, and make sure your conversion tag fires on both versions before sending traffic.
What about testing more than two versions?
Google Ads experiments compare a control with a treatment. To test several ideas, run them in sequence (winner vs next challenger), or use separate campaigns with care, knowing they will not get a clean random split.
Keep a test log
| Field | Example |
|---|---|
| Date and duration | 3 to 31 March, 4 weeks |
| Hypothesis | Target CPA will lower cost per lead |
| Change | Max Clicks → Target CPA (£40) |
| Split and type | 50/50, cookie-based |
| Success metric | Cost per lead |
| Result | (Fill in: winner, difference, significance) |
| Decision | Applied / not applied |
| Next test |
(Example values for illustration.)
A log stops you from repeating tests and shows stakeholders what the team has learned.
Common testing mistakes
- Changing more than one thing at a time.
- Stopping early because one arm looks better.
- Editing the base campaign mid-test.
- Testing tiny changes in low-traffic campaigns.
- Not tracking conversions properly before testing.
- Ignoring seasonality or promotions during the test.
FAQs
What are Google Ads experiments?
A feature that splits a campaign's traffic and budget between the original (control) and a modified version (treatment), so you can measure whether a change such as a new bid strategy, keywords or landing page improves results.
What is the difference between cookie-based and search-based splits?
A search-based split randomly assigns each search, so the same person may see both versions. A cookie-based split assigns each user to one version only. Display experiments always use cookie-based splits.
How long should a Google Ads experiment run?
Until each arm has enough conversions to show a reliable difference, covering at least one full weekly cycle and any conversion delay. Low-volume campaigns usually need several weeks.
What Should I Test First?
Start with the change that could matter most and is easiest to measure: the offer, the headline or the form length on the page with the most traffic. Tiny changes, such as a button color, usually need huge traffic to show an effect.
What If My Account Is Too Small to Test?
Test bigger changes, run tests for longer or test on the highest-volume campaign. If there are too few conversions for a reliable result, use before-and-after comparisons with care, and record the limits in your test log.
Can I A/B test landing pages in Google Ads?
Yes. Create a custom experiment where the treatment uses a different final URL, or use an on-page testing tool that splits visitors on the same URL.