Most Google Ads accounts running today are making decisions based on vibes.
Someone changes a bid strategy, CPA drops for two weeks, and everyone high-fives. Then it climbs back up and the team blames seasonality. Nobody actually knows what worked. Nobody ran a controlled test. And the same guessing game repeats every quarter.
Google Ads Experiments exists to fix exactly this. It’s a native, built-in A/B testing framework that splits your traffic between a control and a variant — cleanly, without contamination — so you can measure what actually changed performance versus what just happened to coincide with it. It’s one of the most powerful features in the platform, and surveys consistently show fewer than one in ten advertisers use it regularly.
That’s your competitive edge, sitting right there in the interface.
- Google Ads Experiments (formerly Drafts & Experiments) runs controlled A/B tests within your live campaigns — same budget period, same auction, split traffic.
- The three highest-ROI test categories are bidding strategy changes, landing page variants, and ad copy — in that order.
- Statistical confidence matters: don’t call a winner before 95% confidence and at least two full weeks of data.
- Experiments protect your baseline — if the variant tanks, you can kill it without having destroyed your original campaign’s learning history.
- Running one well-structured experiment per month compounds into a significant performance advantage over 12 months — most of your competitors aren’t running any.
What Google Ads Experiments Actually Does (And How It’s Different From Just Changing a Campaign)
Here’s the mistake people make constantly: they change something in a live campaign, watch for two weeks, and call it a test. That’s not a test. That’s an observation. You’ve introduced confounds — different time period, different competitive auction dynamics, different seasonality — and you’re comparing apples to memories of apples.
Google Ads Experiments works differently. When you create an experiment, Google splits your campaign’s eligible traffic between the original (the control) and your modified version (the variant) simultaneously. Same auctions. Same time window. Same external conditions. The only variable is whatever you changed.
You control the traffic split. A 50/50 split reaches statistical significance fastest but means half your budget runs on an unproven variant. A 70/30 split (control/variant) is more conservative — better for high-spend accounts where a bad variant is genuinely costly.
This used to live under “Drafts & Experiments” in the old interface. Google rebranded and simplified it into just “Experiments” — you’ll find it in the left-nav sidebar. The mechanics are the same, the naming is cleaner.
What to Test First: The Priority Stack That Actually Moves the Needle
Not all experiments are created equal. Testing button color on your landing page when your bid strategy is wrong is rearranging deck chairs. Here’s the sequence we run for clients, ordered by expected impact:
1. Bidding Strategy Changes
This is the single most valuable use of Experiments, and it’s exactly what the feature was built for. Switching a campaign from Manual CPC or Enhanced CPC to Target CPA or Target ROAS is a major change — one that can take 4–6 weeks to stabilize during the algorithm’s learning period. If it goes badly, you’ve torched your campaign history and your CPA for a quarter.
Run it as an experiment instead. Set your control to your current bidding strategy, set your variant to the new smart bidding strategy, split 50/50, and let it run for at least 4 weeks. If the variant hits your CPA target and shows statistical significance, apply it and move on. If it doesn’t, you’ve lost nothing — your control kept running the whole time.
We’ve moved clients from Max Clicks to Target CPA this way and seen CPA drop 30%+ within the experiment window. We’ve also seen it go the other direction. The point is you find out with data, not with a prayer. For more on how smart bidding actually behaves under the hood, this breakdown on how Google Ads Smart Bidding actually works is worth reading before you set up your first bidding experiment.
2. Landing Page Variants
Landing page testing is where experiments get genuinely exciting — and where most accounts are leaving serious money on the table. Your ad click gets someone to your page. What happens next is entirely outside Google’s control, but it’s 100% inside yours.
To A/B test landing pages with Experiments, you create a variant of your campaign where the final URLs point to a different landing page. Everything else stays identical. Same ads. Same keywords. Same bids. Different destination.
What’s worth testing? Lead form placement (above the fold vs. scrolled to). Headline framing (feature-led vs. outcome-led). Social proof placement. Page length. CTA copy. Run one change at a time — the moment you test two things simultaneously, you lose the ability to know which one drove the result.
A client in professional services ran an experiment testing a long-form landing page against a stripped-down page with just a headline, three bullets, and a form. The stripped-down page converted 41% better. We never would have known without the experiment — the original page had always “felt” right to the team. For deeper context on what makes landing pages convert in a Google Ads context, see our guide to Google Ads landing page best practices that actually lift conversion rates.
3. Ad Copy and RSA Headline Strategies
The native RSA pinning and asset performance data inside Google gives you some signal on copy — but it’s messy and blended across the whole campaign. Experiments let you test more aggressively: a campaign with one RSA against a campaign with a different RSA, where you’ve swapped out the entire creative direction.
Good copy experiments to run: benefit-led headlines vs. feature-led headlines, urgency-based CTAs vs. value-based CTAs, and brand name in headline 1 vs. no brand in headline 1. For a framework on writing ad copy that converts before you run the test, this RSA copy guide will set you up properly.
4. Match Type Strategies
Broad match versus exact match is one of the most debated topics in PPC right now. Stop debating and test it. Set up an experiment where the control runs phrase/exact and the variant introduces broad match keywords with the same negatives. Let the data decide — because the right answer is genuinely account-specific, and anyone who tells you otherwise is oversimplifying.
How to Set Up a Google Ads Experiment: Step by Step
This is the actual workflow, not the watered-down version:
Step 1: Navigate to Experiments in the Google Ads sidebar. Click the “+” to create a new experiment.
Step 2: Select the campaign you want to test. Google will ask you what type of experiment — for custom tests, choose “Custom Experiment.” (Google also offers pre-built experiment templates for specific tests like Performance Max asset testing — those are fine but limited.)
Step 3: Set your traffic split. 50/50 reaches significance fastest. If your campaign spends $10K+/month and a bad variant would genuinely hurt you, go 70/30 control-heavy.
Step 4: Define your experiment duration before you start. Minimum two weeks. Four weeks if you’re testing a bidding strategy (it needs time to exit the learning phase). Set a hard end date so you’re not tempted to call it early.
Step 5: Make your single change in the variant. One variable. Only one.
Step 6: Launch and resist the urge to check it daily. Set a calendar reminder to review at the two-week mark.
Reading Your Experiment Results Without Fooling Yourself
This is where experiments go wrong most often. People see the variant performing better after five days and immediately apply it. That’s not how statistics work.
Wait for 95% statistical confidence. Google shows this directly in the Experiments dashboard — it’s labeled as the confidence interval next to your primary metric. Don’t call a winner before that threshold. A result that looks like a 15% CPA improvement on day seven can normalize to zero difference by day twenty-one as variance irons out.
Pick one primary metric before you start. If your goal is lead volume, your primary metric is conversions. If it’s efficiency, it’s CPA or ROAS. Decide this before launching. If you wait until after to pick the metric where the variant looks best, you’ve run a bad test — you’re just fishing for a result that validates what you already wanted to do.
Look for directional consistency, not just the headline number. If the variant wins on CPA but loses on conversion rate, something weird is happening. Maybe it’s getting lower-quality traffic. Maybe the bid strategy is being too conservative. A result that doesn’t make logical sense is worth interrogating before you apply it.
Document everything. Date, hypothesis, result, confidence level, what you applied or didn’t and why. We use a simple spreadsheet. This compounds over time — after a year, you have an institutional knowledge base that tells you exactly what works in your specific account.
If your campaigns are already showing performance issues that make it hard to get clean experiment data, it’s worth diagnosing the underlying problems first. Our Google Ads stop working diagnosis framework can help you rule out structural issues before you start testing on top of them.
The Experiments Cadence That Actually Compounds
One experiment a quarter is table stakes — you’re not learning fast enough. One experiment a month is where the real advantage starts to show.
Here’s the cadence we run for managed accounts:
- Month 1: Baseline audit, identify the single highest-impact variable to test. Usually bidding strategy.
- Month 2: Landing page experiment on the highest-traffic campaign.
- Month 3: Ad copy direction test based on what the landing page data revealed about messaging.
- Ongoing: One experiment running at all times, rotating through bidding, landing pages, and copy in a logical sequence.
After 12 months of this, you’ve run 10–12 controlled tests. You know definitively what your audience responds to in terms of messaging, what bidding strategy fits your conversion volume, and what landing page structure drives the lowest CPA. Your competitors who are running on gut feel have learned nothing they can prove.
This pairs particularly well with a well-structured account — experiments produce cleaner data when your campaign architecture is sound. If you haven’t reviewed your structure lately, our Google Ads account structure best practices guide is the place to start.
Common Experiment Mistakes That Waste Your Time and Budget
Testing too many variables at once. If you change the bid strategy AND the landing page AND the ad copy in the variant, and it wins, you know nothing. You don’t know which change drove the result, so you can’t replicate it or build on it. One variable. Always.
Running experiments on campaigns with too little volume. If your campaign gets 10 conversions a month, you’ll wait six months for statistical significance. Experiments work best on campaigns generating at least 30–50 conversions per month. Below that, focus on improving conversion volume first.
Calling it early.** The p-value is a probability, not a certainty. Every time you peek at results and consider calling it, you inflate your false-positive risk. Set the duration upfront. Look at the two-week mark and the end date. Not daily.
Not defining what “winning” means in advance. If your target CPA is $80 and the variant delivers $78, that’s not necessarily a meaningful win if the confidence level is 72%. Define your success criteria before launch: metric, threshold, and minimum confidence level. Then honor it.
FAQ: Google Ads Experiments
How is Google Ads Experiments different from just editing a campaign?
When you edit a live campaign, you’re comparing before-and-after data across different time periods — which means seasonality, auction changes, and competitor behavior all contaminate your results. Experiments runs control and variant simultaneously, splitting the same traffic, so the only variable is what you changed. That’s what makes it a real test.
How long should I run a Google Ads experiment?
At minimum, two full weeks. If you’re testing a Smart Bidding strategy change, run for at least four weeks — Smart Bidding needs time to exit the learning phase before you can fairly evaluate performance. Always set an end date before you launch and don’t call it early based on early results.
What traffic split should I use for my experiment?
50/50 reaches statistical significance fastest and is fine for most accounts. If you’re testing something risky (a completely new bidding strategy on a high-spend campaign), go 70/30 control-heavy so a bad variant does less damage. Don’t go below 20% for the variant — you won’t get enough data.
Can I run experiments on Performance Max campaigns?
Google offers specific experiment types for Performance Max, including asset testing. However, the custom experiment options are more limited than with standard Search campaigns. For Search campaigns, you have full control over what you test.
What’s the most valuable thing to A/B test in Google Ads?
Bidding strategy experiments have the highest potential impact because they affect every single auction your campaign enters. After that, landing page tests — because conversion rate improvements reduce your effective CPA without touching your bids at all. Ad copy comes third; the impact is real but usually smaller than the first two.
How many experiments should I run at the same time?
One per campaign at a time. Running two experiments in the same campaign simultaneously makes it impossible to attribute results accurately. Across your account, you can run multiple experiments simultaneously as long as they’re in different campaigns.
What if my experiment shows no statistically significant difference?
That’s a valid and useful result. It tells you the variant is not meaningfully better or worse than the control — which means you shouldn’t apply the change (stick with what’s working) and should move on to testing something with a higher expected impact. A null result is not a failed experiment; it’s information.
If your current Google Ads agency can’t point to a documented experiment they’ve run in your account in the last 90 days — with a hypothesis, a result, and a confidence level — they’re managing your campaigns on intuition. That might be fine intuition, but it’s not compounding into anything.
A good agency runs experiments constantly, documents the results, and uses those results to build a testing roadmap that makes the account measurably better every quarter. If that’s not happening in your account, it’s worth a second opinion.
Our Google Ads audit framework walks you through exactly what to look for — including whether your account has any testing infrastructure at all.