Google Ads Experiments: How Drafts and Experiments Work

Google Ads experiments let you test changes to a campaign on a share of real traffic before you roll them out everywhere. Our team reaches for this feature before switching bidding strategy, rewriting ad copy, or expanding targeting. A direct edit is hard to undo once performance drops, so testing first lowers the risk. This guide walks through drafts, experiments, traffic splitting, and how to read results with confidence.

What Are Google Ads Experiments?

Google Ads experiments let you create an editable copy of a campaign, called a draft. You then run that draft against a percentage of real traffic alongside the original campaign. The original keeps serving ads unchanged, while the experiment arm competes for the same audience pool.

This setup lets you see whether a new setting actually works before you commit your whole budget to it. For example, once the experiment ends, you can apply the change to the original campaign. Alternatively, you can graduate it into its own independent campaign, or end it entirely. Our team treats this step as standard practice before any major account change.

You manage experiments from the campaign management section of the interface, usually under a drafts and experiments tab. Exact menu names shift with interface updates, so confirm the current path inside your own account. The underlying mechanics, including control and treatment arms, are documented in Google's official developer documentation.

What Is the Difference Between a Draft and an Experiment?

A draft is an editable copy of a campaign that never serves ads. Instead, it only lets you prepare changes. An experiment, on the other hand, is the version you launch from that draft, and it does serve real ads to a defined share of traffic.

Suppose you want to apply a change without testing its impact first. In that case, you apply the draft directly to the original campaign and skip the experiment step entirely. Alternatively, if you want to measure the impact, you launch an experiment from that same draft and split traffic between the two versions.

In short, a draft is a preparation step. An experiment, by contrast, is the stage where that preparation gets tested under real conditions. Keeping the two terms separate makes it easier to track which change is live and when.

Some teams skip this distinction. They build a draft and never convert it into an experiment, which means they never collect any real performance data about the change. For that reason, treat the move from draft to experiment as a deliberate, separate decision rather than an automatic next click.

Which Campaign Types Support Google Ads Experiments?

Drafts and experiments work primarily on Search campaigns and Display Network campaigns. Campaigns that rely on a shared budget may have limited draft support. For that reason, check your budget structure before you start.

• Search campaigns: the most common use case, for bidding strategy, ad copy, and landing page tests.
• Display Network campaigns: well suited to targeting and creative tests.
• Each original campaign can run only one active experiment at a time.

Coverage for other campaign types can change as Google rolls out updates. So, confirm the current scope in the official Google Ads Help Center before you rely on it. Video and Shopping campaigns may offer similar tools under different names. In addition, their scope may be narrower, so check the relevant section in your account first.

When Should You Run a Google Ads Experiment?

An experiment makes sense whenever you cannot predict the outcome of a change. It also helps when a wrong guess would be costly. Switching bidding strategy, rewriting a large share of your ad copy, or expanding targeting all fall into that category.

For small, easily reversible edits, such as adding a single keyword, running a full experiment is usually unnecessary overhead. Instead, our team reserves experiments for decisions with uncertain impact and a meaningful effect on budget.

We also recommend testing seasonal changes through an experiment. After all, seasonal swings can make a change look more or less effective than it really is. For example, starting an experiment before a peak season helps you avoid crediting a traffic surge to a change you only just made.

How Does Traffic Splitting Work?

When you create an experiment, you set a traffic split percentage. That percentage determines how user traffic divides between the original campaign and the experiment campaign. An even split typically uses fifty percent for each arm.

The split usually happens at random, so both arms reach a similar audience profile across similar time periods. As a result, this randomness helps ensure that any measured difference reflects a genuine performance gap rather than a sampling quirk.

If you set the traffic share too low, the experiment arm collects less data. Consequently, it takes longer to produce a reliable signal. For that reason, we favor a balanced split whenever your budget supports it, since it gets you a trustworthy answer faster. On smaller budgets, we often give the experiment arm a smaller share instead. That way, it doesn't overshadow the main campaign while it collects data.

What Are the Control Arm and Treatment Arm?

The control arm is the original campaign, left unchanged. It serves as the baseline for comparison. The treatment arm carries the changes from your draft, and its performance gets compared against the control.

Both arms draw from the same traffic pool during the same period. As a result, external factors such as seasonality, competition, and news events affect both arms in a similar way. That means most of the observed difference comes directly from the change you made.

The control and treatment concept comes straight from scientific experiment design. Specifically, Google Ads adapts that logic to advertising, so decisions rely on data instead of guesswork. This matters most on large budgets, where a wrong assumption carries a real cost.

Never edit the control arm while an experiment runs. Otherwise, if you tweak the original campaign mid-test, you lose the ability to tell which arm's change actually caused the difference you observe.

Testing a Bidding Strategy Change With an Experiment

Switching to an automated strategy like Target CPA or Target ROAS changes both spend and conversion volume directly. Testing that switch through an experiment shows clearly whether the new strategy genuinely outperforms the old one.

Our team keeps bidding strategy experiments running for at least a few weeks. After all, automated bidding systems need a learning period to stabilize. Ending the test too early misrepresents the strategy's real potential, and that can lead you to the wrong call.

Watch cost per conversion and conversion volume together during a bidding test. Otherwise, a strategy that improves one can quietly hurt the other. So, we always check both metrics rather than relying on a single number.

Testing Ad Copy Changes With an Experiment

A new headline or description set can shift click-through rate and conversion rate in either direction. Instead of deleting live ads and publishing new ones outright, testing the new copy in the treatment arm keeps the original version safe.

If the new copy underperforms, the control arm keeps running without interruption. When results look positive, you apply the change to the original campaign. Otherwise, you end the experiment and keep the existing copy.

Ad copy tests usually produce a readable signal faster than bidding strategy tests. That's because clicks and impressions accumulate faster than conversions. In practice, that makes copy testing a good habit to repeat often.

Testing a Targeting Change With an Experiment

Adding a new location, device, or audience setting can meaningfully shift the type of user your campaign reaches. Testing that expansion through an experiment helps you catch a drop in conversion quality early.

This matters most for geographic expansion, or for adding a new audience segment. In those cases, an experiment keeps your budget from flowing into lower quality traffic unchecked. When results are positive, therefore, you make the expansion permanent.

Track both quality and volume in a targeting test. Specifically, a new segment can lower conversion rate while still raising total conversions. Weighing both metrics together clarifies whether the expansion is actually worth keeping.

How Do You Create a Google Ads Experiment?

Creating an experiment starts with preparing a draft. Next, you launch the experiment from that same draft. The general sequence looks like this.

1. Create a draft from the original campaign you want to test.
2. Make your bidding strategy, ad copy, or targeting changes inside the draft.
3. Launch an experiment from the draft, and set the traffic split percentage and the experiment's duration.
4. Then, monitor performance metrics regularly while the experiment runs.
5. At the end of the period, decide whether to apply, graduate, or end the experiment based on the results.

Exact button and tab names inside the interface change over time. So, check the official Google Ads Help Center for the current steps before you rely on this outline. Our team always launches a first experiment on a new account with a small traffic share. Then, we increase it once we trust the setup.

How Long Should a Google Ads Experiment Run?

Experiment duration depends on the campaign's daily traffic volume. It also depends on how large the tested change is. Lower traffic campaigns need more time before a consistent difference becomes visible.

Automated bidding tests need time for the system to finish its learning phase. Otherwise, ending them early can produce a misleading read. Ad copy tests, on the other hand, typically accumulate meaningful data faster.

We wait until both arms gather enough conversions before drawing a conclusion. Otherwise, random day-to-day noise is easy to mistake for a real effect. A quick weekly check-in on early signals keeps the process predictable, without forcing an early call.

How Do You Read Google Ads Experiment Results?

When you read results, look beyond total conversion count. Instead, check how conversion rate, cost per conversion, and spend move together. An improvement in one metric can mask a decline in another.

A small sample size can produce a difference that isn't statistically meaningful. So, make sure enough data has accumulated before you act on it. The interface lets you compare both arms side by side.

When the result still looks unclear, extending the experiment a bit longer beats making an early, shaky call. Our team often adds one more week when we're on the fence. That extra time rarely costs much against the risk of acting too soon.

Drafts vs. Experiments: A Quick Comparison

The table below summarizes the core differences between a draft and an experiment.

This table gives you a quick reference for deciding which tool fits a given situation. In addition, we share a version of it with clients during onboarding to make both concepts concrete.

Applying, Graduating, or Ending an Experiment: What's the Difference?

At the end of an experiment, you have three options. First, you can apply the changes to the original campaign. Second, you can turn the experiment into its own independent campaign. Finally, you can end it entirely.

• Apply: carries the changes into the original campaign, and the experiment campaign disappears.
• Graduate: turns the experiment into a separate, permanent campaign with its own budget.
• End: closes the experiment irreversibly without applying any change.

Your choice depends on how strong the results are. It also depends on whether you want the experiment campaign to run independently. Applying covers most cases; graduating, on the other hand, usually suits situations where you want a different budget or reporting structure going forward.

What Mistakes Do Teams Make With Google Ads Experiments?

The most common mistake is ending an experiment before enough data has accumulated. A close second is testing several major changes inside a single experiment. That mix makes it impossible to tell which change actually drove the result.

• Closing an experiment too soon and mistaking random noise for a real effect.
• Testing more than one variable inside a single experiment.
• Setting a traffic split that ignores the campaign's actual volume.
• Applying results without accounting for seasonal effects.

Avoiding these mistakes makes it far easier for an experiment to give you a sound, reliable answer. Another frequent error is launching the treatment arm with a different starting budget or bid cap than the control. As a result, the two arms were never competing on equal footing to begin with.

Should You Apply a Draft Directly or Run an Experiment?

If you're confident about a change's impact and the risk is low, applying the draft directly saves time. However, if the impact is uncertain, or the change affects a large share of your budget, running an experiment is the safer path.

Our general rule for clients is simple. Apply small, reversible changes directly, and test large, costly-to-reverse changes through an experiment first. This rule clarifies the right path in most decision moments, even under time pressure. After all, a rushed direct apply can turn into a spend loss that's hard to undo.

Are Google Ads Experiments the Same as A/B Testing?

Google Ads experiments are conceptually an A/B test. Specifically, they compare a control arm and a treatment arm drawn from the same traffic pool. The difference is that this test runs inside the Google Ads interface itself, at the campaign level, under platform-specific rules.

The general term A/B testing can also cover website, email, or app tests. Google Ads experiments, by contrast, apply specifically to ad campaigns. Knowing that difference clarifies which tool fits which scenario. Likewise, it means you don't need a separate testing platform, since the campaign-level infrastructure is already built into the interface.

How Should You Share Experiment Results With Your Team?

Sharing results on a regular rhythm keeps a decision grounded in a shared view. Otherwise, a decision risks resting on one person's read of the data alone. Our team writes a short weekly summary for every active experiment and shares it with everyone involved.

Specifically, that summary covers the traffic split, elapsed time, accumulated conversions, and the metric gap worth watching. As a result, everyone looks at the same numbers when the decision point arrives, instead of relying on a single person's memory.

Folding this summary into a monthly performance review turns experimentation into a lasting habit rather than a one-off activity. It also keeps the experiment's status from slipping through the cracks. Clients tend to trust a recommendation more once they see the same weekly pattern repeated over several tests.

What Should You Check Before Launching an Experiment?

Before you launch an experiment, write down the exact question you're trying to answer in one sentence. A vague goal makes it hard to say, once the test ends, which outcome actually counts as success.

Our team answers three questions before every experiment. First, which variable are we testing? Second, which metric will decide the outcome? Third, how long will we wait? Agreeing on these upfront prevents disagreement once the data comes in.

First, note the campaign's recent performance baseline before starting. That baseline gives later comparisons real context. Without it, the difference the experiment reveals is harder to interpret correctly. A missing baseline is a common reason teams argue about a result that otherwise looks clear.

What Should You Monitor After an Experiment Ends?

The process doesn't end once you make a decision and close the experiment. Instead, the effect of a change you apply can play out somewhat differently once it covers the full campaign rather than a limited traffic share.

Our team watches the first few weeks after applying a change closely. After all, an effect that looked clear on a small traffic share can shift once it spans the whole budget. Any unexpected swing during that window is worth a second look.

Keeping a short record of each applied change and its outcome gives you a ready reference the next time a similar decision comes up. In this way, scattered test history turns into a shared, lasting resource for the whole team, rather than staying locked in one person's memory.

Which Metrics Should Guide Your Google Ads Experiment Decisions?

The metrics worth checking most are conversion count, conversion rate, cost per conversion, and overall spend. Depending on your goal, one of these becomes the primary reference, while the others confirm the result.

As an example scenario, a bidding strategy experiment usually prioritizes cost per conversion. An ad copy experiment, by contrast, weighs click-through rate and conversion rate together. Our team presents these metrics side by side in reports, so clients can decide with confidence. Targeting experiments often come down to a trade-off between conversion quality and total conversion volume.

Looking at several metrics together, rather than one in isolation, gives you a sturdier read on what an experiment actually proved. This habit helps separate a short-term blip from a lasting improvement. Likewise, reading a few weeks of data together before you apply a change strengthens the final call. In practice, patience at this stage saves more budget than any single clever setting.

If you'd like hands-on support running these tests, our Google Ads management service covers bidding strategy rollouts and experiment setup as part of ongoing account work.

Which Tools and Guides Help With Google Ads Experiments?

A few calculators speed up the decision process once an experiment produces results. For instance, the A/B test calculator helps you check whether the gap between two arms is statistically meaningful.

For budget planning ahead of a test, use the Google Ads budget calculator. To track a shifting conversion rate, use the conversion rate calculator. For Target ROAS tests, meanwhile, the ROAS calculator is a handy reference. The Google Ads audit tool is also worth a look to check the account's overall health.

Before testing a bidding strategy switch through an experiment, our guide on updating Target CPA and Target ROAS bidding covers the concept in more depth. In addition, our Quality Score guide explains how score shifts can color experiment results.

For a healthy approach to using conversion data inside experiment decisions, our piece on offline conversion import setup is a useful companion read. After all, feeding incomplete offline data into a test risks pointing you toward the wrong call. On the Meta side, our guide on setting a Meta ad account spending limit covers a similar budget-safety topic worth comparing.

Core guide: What Is Digital Marketing? Its types, examples and how it is done, on one page with an interactive film that travels through the brain.