ScaleyGoogle Ads Experiments: How to Test Ads and Settings
The built-in way to A/B test ad copy, bidding and settings in Google Ads, without guessing and without breaking your live campaign.
Written and reviewed by Christopher Krassnig
Published Last checked
Google Ads experiments let you test a change on part of a campaign's traffic before you roll it out. A custom experiment copies a campaign, changes one thing and splits traffic, with 50% the split Google recommends. Ad variations test one ad text change, like an RSA headline, across many campaigns at once.
Also called: ad testing, A/B tests, ad variations, custom experiments, RSA tests
The Experiment Types in Google Ads
Google lists them on its Experiments page. The ones a store uses most:
| Type | Best for | Campaign types |
|---|---|---|
| Ad variations | One text change across many ads, like a new call to action in every RSA. | Search |
| Custom experiments | Bidding, match types, landing pages, audiences and ad groups. | Search, Display, Video and Hotel (not App or Shopping) |
| Performance Max experiments | Measuring the uplift of Performance Max, or testing its features. | Performance Max |
| AI Max experiments | A one-click test of AI Max features, split 50/50 inside the campaign. | Search |
| Demand Gen experiments | Which images or videos work best. | Demand Gen |
Note the gap for stores: Google says you "won't be able to create custom experiments for App or Shopping campaigns." To test Shopping, most stores split products or budgets across campaigns and compare, which is a weaker test.
How to Run an RSA Test
If you searched "RSA test" for Google Ads, this is the part you want. Google's tool for testing ad text is ad variations. You pick the scope (the whole account, one campaign or a custom set), pick responsive search ads, and set the change, like finding "Local Hotels" and replacing it with "Nearby Hotels". Then you set the end date and the share of traffic that sees the new version.
Google says ad variations are best when you "test one change across multiple campaigns or your entire account". Custom experiments are better when you want to test "multiple simultaneous changes on a smaller scale".
Inside one RSA, Google is already testing your headlines against each other. An experiment is for testing a decision, like a new offer line in every ad, against what you run today.
The Rules That Trip People Up
- You can schedule up to 5 experiments for a campaign, but only one runs at a time.
- The experiment shares the original campaign's traffic and budget. Google recommends a 50% split for the best comparison.
- Cookie-based splits keep each person in one version. Search-based splits can be faster but one person may see both.
- With audience lists and a cookie split, Google says you should have at least 10,000 users in the list, or results may be less accurate.
- Changes you make to the original campaign during the test are not copied into the experiment, and they muddy the result.
- If the base campaign is paused or ends early, the experiment stops too.
Reading the Result
Pick up to 2 success metrics when you set it up, like conversions and cost per conversion. Let it run long enough to cover a full buying cycle, and do not touch either side while it runs.
When one side wins, you can apply the change to the original campaign or turn the experiment into a new campaign. Google keeps the performance data either way. A tiny difference over a short test is noise. Run it longer before you act on it.
Example: A Supplement Store Tests Smart Bidding
Made-up numbers, to show the method. A supplement store runs its Search campaign on manual bids and wonders if Target ROAS would do better.
It creates a custom experiment from the campaign, switches the experiment arm to Target ROAS, and sets a 50% cookie-based split for 6 weeks. Success metrics: conversion value and cost.
After 6 weeks the experiment arm shows more conversion value at the same spend, and the gap held steady for the last 4 weeks. The store applies the experiment to the original campaign. Our Smart Bidding answer explains the strategy it switched to.
Common Mistakes
Testing five things at once
If the experiment changes bids, keywords and ads together, you will not know which change did it. Test one decision.
Ending it after a few days
Short tests catch noise. Cover at least one full buying cycle.
Editing the original mid-test
Google says changes to the original are not copied into the experiment, so your comparison breaks.
Looking for custom experiments on Shopping
Google does not offer custom experiments for Shopping campaigns. Plan a different test there.
Questions People Ask
Responsive search ad. It is the standard Search ad, built from up to 15 headlines and 4 descriptions you write.
Google's Own Pages We Quoted
Every rule, limit and size on this page comes from these pages, read on 8 October 2026. Google changes them, so check the page itself before you act.
Where to Go Next
- What Is Smart Bidding?The most common thing stores test in an experiment.
- Google AI Max for SearchAI Max has its own one-click experiment.
- Scaling Rules That Actually WorkWhat to do with a winner once the test is done.
- PPC Strategy: How to Build a Google Ads Strategy That PaysWhere testing fits in a full PPC strategy.
Ask Where Your Ad Money Went
Connect read-only in under five minutes. Ask which search terms and products lost money last month. Not ready to connect? The two-minute quiz scores you with no Google Ads login.
3-day free trial. Card required. Cancel anytime