Google Ads Experiments: How to Test Campaign Changes Safely

Google Ads Experiments: How to Test Campaign Changes Safely

Google Ads experiments let you test changes to your campaigns -- such as new bidding strategies, ad copy, keywords, or landing pages -- against your current setup with a controlled portion of traffic and budget. They replace guesswork with data so you make evidence-based decisions instead of risking live campaign performance on untested ideas.

TL;DR: Google Ads offers three types of experiments: Campaign Experiments (split tests using drafts), Ad Variations, and Custom Experiments. Each lets you test a different layer of your account -- campaign settings, ad creative, or broader measurement goals like brand lift. Run experiments for a minimum of two weeks, use statistically significant data to declare winners, and never test too many variables at once.

What Is a Google Ads Experiment?

A Google Ads experiment is a controlled test that splits traffic and budget between your original campaign and a modified version so you can measure which performs better against a defined goal. Google serves each variation to a separate audience segment -- they never overlap within a single experiment -- and reports results side by side so you can compare metrics like conversions, cost per conversion, click-through rate, and return on ad spend.

Experiments give you a safe sandbox. Instead of changing a live campaign and hoping for the best, you run the change on a defined percentage of traffic (typically 50%) while the original keeps running unchanged. If the test variant wins, you apply the change. If it loses, you end the experiment and nothing breaks. This is especially valuable for B2B SaaS and startup advertisers with limited budgets who cannot afford to gamble on untested ideas -- see our Google Ads guide for startups for broader paid search foundations.

Not every change needs an experiment. Small tweaks like adding a negative keyword or adjusting an ad extension are safe to apply directly. But structural changes, bidding strategy overhauls, new ad copy angles, or landing page swaps all justify the rigor of a controlled experiment.

How Do Drafts and Experiments Work?

Campaign experiments in Google Ads are built on a two-step system: you first create a draft, which is a snapshot of your campaign with proposed changes, then you convert that draft into an experiment that runs live against the original.

A draft is a non-public workspace. When you create a draft, Google copies your campaign as it exists at that moment. You can then modify anything inside the draft -- bids, budgets, keywords, ad copy, audiences, device targeting -- without affecting the live campaign. The original campaign continues running normally while you prepare your changes.

Once you are satisfied with the draft, you convert it to an experiment. At that point you choose the experiment split (what percentage of budget and traffic goes to the experiment versus the original), set a start and end date, and launch. Google then serves the original and experiment variants simultaneously, using cookies to ensure each user consistently sees the same version. During the run, both sides accrue their own performance data independently.

Key distinction: drafts are prep work; experiments are the actual live test. You cannot run an experiment without a draft, but you can save a draft without ever launching it. This is useful when you want to prepare several test ideas and launch them one at a time.

How Do You Create a Campaign Experiment (Split Test)?

Here is a step-by-step walkthrough for creating a campaign-level experiment in Google Ads:

1. Navigate to the Campaigns page and check the box next to the campaign you want to test.
2. Click the "Experiments" link in the left-hand navigation panel, or find "Drafts and experiments" in the campaign-level menu.
3. Click the plus button to create a new draft. Google copies the selected campaign into a draft. Give the draft a descriptive name that includes the variable being tested -- for example, "Search -- Test Target CPA $50 vs $35".
4. Open the draft and make your changes. You can adjust bids, add or remove keywords, swap ad copy, change audience targeting, modify device bid adjustments, or alter the campaign's campaign structure itself.
5. When the draft is ready, click "Run an experiment" or "Convert to experiment."
6. Name the experiment and set the split. A 50/50 split is the most common and statistically clean choice. You can allocate as little as 1% to the experiment, but smaller splits need much longer to reach statistical significance.
7. Set a start date and, optionally, an end date. If you skip the end date, the experiment runs until you manually stop it.
8. Choose a goal metric the experiment will optimize toward during the test -- usually the same goal as the original campaign.
9. Launch the experiment. Both the original and the experiment will begin serving side by side within hours.

During the experiment, Google surfaces a side-by-side comparison table showing metrics for both variants and highlighting statistically significant differences. You can watch this dashboard to see how the experiment is trending before deciding to roll out or end.

What Is an Ad Variation?

An Ad Variation is a simpler, more focused testing tool built directly into Google Ads. Unlike campaign experiments -- which test changes at the campaign or ad group level -- ad variations test ad creative only: headlines, descriptions, final URLs, and display paths.

You create an ad variation by selecting one or more campaigns, choosing the type of change (for example, "Find and replace" text in headlines, "Update text" across all ads, or "Swap" two existing headlines), and setting the split. Google then serves the modified ads alongside the originals and reports which version drives better performance.

Ad variations are ideal when you want to test a single hypothesis about ad copy without the overhead of building a full campaign draft and experiment. For instance, you might test whether including a number in headline two improves CTR, or whether a shorter description line one lifts conversion rate. Because ad variations work across multiple campaigns simultaneously, they are also useful when you have a consistent copy change you want to validate at scale.

For a deeper dive into systematic ad testing, see our guide on building an ad creative testing framework.

Can You Test Smart Bidding with Experiments?

Yes, and smart bidding changes are among the most common and highest-impact experiments you can run. A typical test compares two automated bid strategies -- for example, your current Target CPA bidding versus Maximize Conversions, or Target ROAS versus Maximize Conversion Value.

Since smart bidding strategies rely on historical conversion data to optimize, an experiment is the only reliable way to answer "will this new strategy actually improve my cost per conversion?" without exposing your entire budget to an unproven approach. The campaign experiment framework ensures the new strategy receives only the allocated split of traffic and budget, and you can monitor whether the algorithm finds its footing within the test period.

One caution: smart bidding experiments need sufficient conversion volume to reach significance. If your campaign generates fewer than 30 conversions per month, a 50/50 split experiment may take two months or longer to yield a clear signal. In those cases, either extend the test window or consider whether the experiment is worth running at all. For more on bid strategy selection, read our guide to Smart Bidding in Google Ads.

How Long Should a Google Ads Experiment Run?

The shortest defensible experiment window is two weeks, but the right answer depends on your campaign volume, conversion cycle, and the size of the change being tested. Google recommends a minimum of two weeks and ideally three to four to account for day-of-week effects and conversion lag.

Three factors determine the ideal duration:

• Conversion volume: The lower your volume, the longer you need. Aim for at least 100 conversions per variant before drawing conclusions. At 10 conversions per week per side, that means a minimum of 10 weeks.
• Conversion delay: If your product has a free trial that converts after 14 days, a two-week experiment will undercount conversions that started during the test but completed after it ended. Extend the window or factor in the delay when reading results.
• Learning phase: Smart bidding strategies go through a learning period when launched or significantly changed. Budget for one to two weeks of learning that should not be counted when evaluating performance, or run the experiment long enough that the learning period is a small fraction of total data.

Most advertisers should default to four weeks for a clean read, and never stop an experiment early just because early results look promising or disappointing. Early data is noisy; let statistical significance -- indicated by confidence badges in the experiment report -- be your guide, not gut feel on day three.

How Do You Read Results and Roll Out?

Google Ads experiments display a comparison table with key metrics for the original and experiment variants side by side, plus a delta column showing the difference. A green upward arrow means the experiment variant performed better on that metric; a red downward arrow means it performed worse. Google also flags differences that reach statistical significance with a notation.

When evaluating results, prioritize the metric that matches your test objective:

• If you were testing a bid strategy change, look at cost per conversion or ROAS.
• If you tested ad copy, look at click-through rate and conversion rate.
• If you tested landing pages, look at conversion rate and bounce rate if available.

Do not cherry-pick metrics. It is common for one variant to win on CTR but lose on conversion rate. The metric tied to your business goal is the one that matters.

To roll out a winning experiment, click "Apply" on the experiment page. Google gives you two choices: apply all changes to the original campaign (which replaces the original with the experiment's settings), or create a new campaign with the experiment's settings. Rolling out keeps the experiment's accumulated performance history in the original campaign's data.

If the experiment is inconclusive, extend it. If the experiment variant clearly loses, end it and keep the original. Not every test produces a winner, and that is fine -- eliminating losing ideas is as valuable as finding winning ones.

Common Mistakes to Avoid

Testing too many variables at once. If you change the bid strategy, add keywords, and rewrite all the ads in a single experiment, you will not know which change caused the performance difference. Isolate one variable per experiment.

Ending experiments too early. Statistically significant results need time. Pulling the plug after a week because the experiment looks like a winner or a loser guarantees you are acting on noise. Commit to a minimum duration before launching and stick to it unless something has clearly broken.

Running overlapping experiments on the same campaign. Google does not allow multiple simultaneous campaign experiments on the same campaign. If you need to test more than one thing, run them sequentially.

Ignoring external factors. Seasonality, competitor promotions, and changes to your own website or funnel can all influence experiment results. If you ran a LinkedIn push or a product launch during the experiment window, the results may be contaminated. Note external events in your experiment log and factor them into your analysis.

Chasing marginal improvements. A 2% improvement in CTR that is statistically significant but practically meaningless is not worth the cost of running and analyzing an experiment. Prioritize tests with material potential impact -- bid strategy changes, structural reorganizations, and major creative overhauls.

Actionable Checklist Before Launching an Experiment

• Define exactly one independent variable to test.
• Choose a primary success metric tied to your business goal.
• Set a minimum duration (at least two weeks) and minimum conversion threshold.
• Confirm the campaign has enough volume to reach significance in that timeframe.
• Document the hypothesis, methodology, and expected outcome before launching.
• Check that no overlapping experiment is running on the same campaign.
• Set a calendar reminder to review results at the planned end date -- do not peek and act early.

Frequently Asked Questions

What Is the Difference Between a Draft and an Experiment in Google Ads?

A draft is a private, non-public copy of your campaign where you prepare and save changes. An experiment is the live, traffic-split test that you launch from a draft. Drafts let you safely build and review changes before anything goes live; experiments put those changes in front of real users alongside the original campaign so you can measure performance differences.

How Long Should a Google Ads Experiment Run?

A minimum of two weeks is recommended to account for day-of-week patterns and bid strategy learning. Three to four weeks is safer for most campaigns. Low-volume accounts may need significantly longer -- aim for at least 100 conversions per variant. Smart bidding experiments should factor in an additional one-to-two-week learning period that should not be counted when evaluating results.

Can You Run an Experiment on a Smart Bidding Strategy?

Yes. Campaign experiments fully support testing one automated bid strategy against another, such as Target CPA versus Maximize Conversions. This is one of the most common and impactful uses of experiments. Just ensure the campaign has enough conversion volume for the new strategy to complete its learning phase and for the results to reach statistical significance within a reasonable window.

What Is an Ad Variation in Google Ads?

An Ad variation is a lightweight testing tool that lets you modify ad copy -- headlines, descriptions, final URLs, and display paths -- across one or more campaigns and split-test the modified ads against the originals. Unlike a full campaign experiment, ad variations do not require a draft and test only creative elements, making them the fastest way to validate ad copy hypotheses at scale.

How Do You Decide Whether to Roll Out an Experiment?

Roll out when the experiment variant shows a statistically significant improvement on the primary success metric you defined before launch -- typically cost per conversion, ROAS, or conversion rate. Avoid rolling out based on secondary metrics alone. If results are inconclusive, extend the experiment. If the experiment clearly underperforms, end it and keep the original campaign unchanged.