Google Ads Experiments: How to Test Bidding, AI Max and Landing Pages

#01What should you test?

#02How do you set one up?

1. Go to Campaigns → Experiments and create a custom experiment.
2. Choose the original campaign and make the change in the treatment.
3. Set the split. 50% is fastest to read.
4. For Search, choose cookie-based (each user sees one version) or search-based (faster, but users may see both).
5. Optionally turn on auto-apply if the treatment wins, choosing your success metric and whether significance is required.
6. Schedule it and don't edit the original campaign while it runs.

#03How long should you run it?

Use the bidding rule from lesson 11: allow learning, then read over at least two conversion cycles or two weeks, whichever is longer. Google's experiment reports show whether a difference is statistically significant; don't call a winner before it is.

Our view: one well-run experiment a month beats ten changes nobody can attribute. For changes that experiments can't cover, such as a Performance Max launch, use geo or time-based tests (see GEO experiment design).

#04Which split should you use?

Audience list tests need enough users; Google mentions 10,000 users for audience lists.

#05How do you run an experiment, step by step?

1. Write the hypothesis and one primary metric (for example cost per sales-qualified opportunity).
2. Write the decision rule before launch: what result means apply, end or extend.
3. Create the experiment from the Experiments page, choose the split and percentage.
4. Leave it alone: no other big changes to either arm.
5. Run it for at least a full conversion cycle, longer for lead generation.
6. Read the result with its confidence interval.
7. Apply, end or extend according to your rule. Google can auto-apply a winning trial if you set it.

#06Troubleshooting

#07Practitioner hacks for testing

• Test big ideas. Small changes need far more data to show a difference.
• Keep a test log with hypothesis, dates, result and decision.
• Don't test during unusual periods such as major holidays, unless that's the point.
• Agree the decision rule with stakeholders before launch, so nobody argues about the result afterwards.
• Re-check after applying. A winner can fade.

#08What should you test first?

#09Questions to ask before launching a test

1. What's the hypothesis?
2. What's the primary metric?
3. What's the decision rule?
4. How long will it run, and is there enough data?
5. What changes are frozen during the test?

#10What edge cases come up?

• Shared budgets can make arms compete; check budgets before starting.
• Low conversion volume means tests take a long time; consider a higher-volume metric as a guide, while keeping the business outcome as the decision metric.
• Overlapping tests on the same campaign confuse results; run one at a time.

Related: GEO experiment design for testing changes covers testing on the organic side.

#11A 30-minute audit for this lesson

1. List past and current experiments.
2. Check each had a written decision rule.
3. Check for overlapping tests.
4. Check conversion volume for a planned test.
5. Write the next hypothesis.

#12Key terms

#13Common scenarios and fixes

Experiment shows no clear winner after four weeks. Diagnosis: Too little data or too small a change. Fix: Extend or test a bigger change.

Someone changed the control mid-test. Diagnosis: No freeze agreed. Fix: Restart and agree a change freeze.

The winner faded after applying. Diagnosis: Novelty or seasonality. Fix: Monitor for several weeks and keep the test log.

#14How long should a test run?

Long enough to see a reliable difference in the metric you care about, and at least one full conversion cycle. For lead generation, where a lead may take days to turn into a qualified opportunity, that usually means several weeks. Google's experiment reports show whether a difference is statistically significant; use that rather than eyeballing daily charts.

If your account has low conversion volume, a test on the final business outcome may take months. In that case you can watch a higher-volume metric, such as leads, as an early guide, while still making the final decision on the business outcome. Write that down in the decision rule.

Avoid stopping a test early because it looks good after a week. Early results swing, and the first days include learning. Equally, don't keep a test running forever: if after the planned time there is no clear difference, the change probably doesn't matter much, and you can choose on cost or simplicity.

#15How do you write a decision rule?

A good decision rule names the metric, the threshold and the time. For example: "After six weeks, apply the trial if cost per sales-qualified opportunity is lower and the result is statistically significant; end it if it is higher; otherwise keep the control because it is simpler." Write it before launch, share it with stakeholders, and store it in the test log. It removes most of the argument when the result arrives, because everyone agreed the rule in advance.

#16Worked example: Ledgerline's bidding test

Ledgerline is a fictional accounts-payable software company. Details and numbers are illustrative.

Ledgerline wanted to move from Target CPA to value-based bidding after starting to upload pipeline values. It ran a 50% search-based custom experiment for six weeks on its biggest non-brand campaign, with the rule "apply if conversion value is significantly higher at similar cost". The treatment won and was applied.

Walkthrough: Ledgerline's landing page test

1. Hypothesis: a page focused on mobile approvals will lift demo requests.
2. Created an experiment with a cookie-based 50/50 split and a final URL change in the trial.
3. Decision rule: apply only if cost per demo is lower with a confident result after 6 weeks.
4. Reviewed weekly without changing either arm, then applied the result at week 6.

#17Try it yourself

1. Write one hypothesis, a primary metric and a decision rule.
2. Choose the split type and say why.
3. Estimate whether you have enough conversions in 6 weeks.

#18Quick quiz

1. Which split suits ad copy tests?

Answer: Cookie-based.

2. What should you write before launch?

Answer: The hypothesis, primary metric and decision rule.

3. Can Google apply a winning trial automatically?

Answer: Yes, if you set it to auto-apply.

#19What should you check before the next lesson?

• Pick one change for a custom experiment this month.
• Choose a split and success metric, and write them down.
• Leave the original campaign unchanged during the test.
• Apply or end the test based on the written rule.

Next: Google Ads scripts for marketers. Or return to the Google Ads for Marketers track.

#20Common questions

What split should I use for a Search experiment? Search experiments can use a cookie-based or search-based split. Cookie-based keeps each user in one arm; search-based splits each search and reaches significance faster.

How big do audience lists need to be for experiments? Google says that with a cookie-based split and audience lists, you should have at least 10,000 users in the list.

What happens if the experiment wins? You can apply the changes to the original campaign, or convert the experiment into a new campaign. Auto-apply can do this for you.