Changing a live campaign on a hunch is risky, and Google Ads experiments remove most of that risk. An experiment runs a proposed change on part of your traffic while the original campaign keeps running, so you can compare the two before you commit. This guide covers what you can test, how to set an experiment up and how to judge the result.
In this guide: What a Google Ads Experiment Is | What You Can Test | How to Set One Up | How Long to Run It | Reading the Results | Common Mistakes | A Worked Example | Document What You Learn | When Not to Run an Experiment | Frequently Asked Questions | Final Thoughts
What a Google Ads Experiment Is
You create a test version of a campaign on the Experiments page in your account and split budget or traffic between the original and the test. Google’s help page on the Experiments page (formerly called drafts and experiments) lists the types available, including ad variations, custom experiments and Performance Max experiments.
The logic is the same as any fair test. Change one thing, keep everything else the same and compare the two sides over the same period.
What You Can Test
Custom experiments on Search, Display and App campaigns can test Smart Bidding, match types, landing pages, audiences or ad groups. Ad variations test text ads or responsive search ads. Good first tests include:
- A bidding change. Moving from manual bidding to a target-based strategy is a classic test. Read our guide to Smart Bidding first.
- A match type change. See how broad match behaves against your current setup, using what you learned in our guide to broad, phrase and exact match.
- New ad copy. Test a different angle with responsive search ads.
- A new landing page. Send the test half of the traffic to the new page.
How to Set One Up
- Write down the hypothesis. For example: “A target cost per action strategy will lower cost per lead without cutting lead volume.”
- Pick one change. If you change two things, you will not know which one caused the result.
- Choose the split. A 50/50 split gives the cleanest comparison and the fastest data. A smaller test share lowers risk but takes longer to show a result.
- Set the success metric before you start. Pick conversions or cost per conversion, not clicks.
- Set start and end dates. Leave enough time for the test to gather data.
How Long to Run It
Google advises that if results still show as in progress, undecided or unavailable, you should run the experiment for at least four to six weeks, and you may need a larger budget or a longer run to collect enough data. Bidding changes also need time to settle after the switch, so early numbers can mislead.
Leave both versions alone while the test runs. Changing budgets, bids or ads on either side during the test makes the comparison unreliable.
Reading the Results
Compare the original and the experiment on the metric you chose in advance. Look at conversion volume as well as cost, because a lower cost per conversion on far fewer conversions is not a win.
If the experiment performs better, Google lets you apply it to the original campaign or replace the original. For custom experiments, you can apply the settings to the base campaign or keep the trial as its own campaign.
Common Mistakes
- Testing too many changes at once. One variable per test keeps results clear.
- Stopping early. A few good days are not a result.
- Testing during an unusual period. A promotion, a seasonal peak or a budget change on the original can distort the comparison.
- Judging by clicks alone. A better click-through rate does not always mean better conversions.
A Worked Example
Suppose a lead generation account runs a steady budget on a campaign with manual bidding, and the owner wants to try a target cost per action strategy. The hypothesis is that the new strategy lowers cost per lead without cutting lead volume. You create a custom experiment, split traffic 50/50, choose cost per conversion as the main metric and let it run for six weeks.
At the end, compare conversions, cost per conversion and lead quality on both sides. If the test side shows a lower cost with similar volume, apply it. If it shows lower volume, the experiment has saved the base campaign from a costly mistake.
Document What You Learn
Keep a simple log with the hypothesis, the change, the dates, the split, the metric, the result and the decision. After a few tests you will see which kinds of changes move your account and which do not, and anyone who joins the team can see what has already been tried. Review old results before you repeat a test, because a change that failed under one setup can work after other improvements.
When Not to Run an Experiment
Some changes are too small or too urgent to test. Fixing a broken tracking tag, pausing a clearly wasteful keyword or correcting a typo in an ad does not need a split test. Experiments suit decisions where the outcome is uncertain and being wrong would cost real money.
Accounts with very few conversions may also find that a test never reaches a clear answer. In that case a careful before-and-after review of one change can be the practical choice.
Frequently Asked Questions
Does running an experiment cost extra?
The test spends from the budget you allocate to it, so decide the split before you start and make sure both sides can still reach enough conversions.
Can small accounts run experiments?
Yes, but they collect conversions slowly. Test bigger changes, run the test longer and avoid changing several things together.
Final Thoughts
Experiments turn opinions into evidence. Test one change at a time, measure the metric you picked in advance and give the test enough time before you apply it.
Want expert help with your account? See our Google Ads management service.