Google Ads Experiments let you run controlled split tests on campaign settings — bidding strategies, audience targeting, ad copy, landing pages, and more — by diverting a percentage of traffic to a variant while keeping the original campaign running unchanged. This eliminates guesswork from campaign optimisation and provides statistically significant data before you commit to changes. For Indian advertisers spending ₹50,000–₹10,00,000/month on Google Ads, experiments can prevent costly mistakes and reveal the highest-performing configurations.
Last updated October 2026
What Are Google Ads Experiments?
Google Ads Experiments (formerly known as Campaign Drafts and Experiments, now integrated as “Experiments” in the Google Ads interface) allow you to create a copy of an existing Search or Performance Max campaign, modify specific settings, and run both versions simultaneously. Google splits eligible auction traffic between the original and the experiment based on a percentage you define.
You can test:
- Bidding strategies: Manual CPC vs Target CPA vs Maximise Conversions vs Target ROAS
- Target CPA/ROAS values: test ₹200 target CPA vs ₹150 target CPA
- Ad copy variations: different headlines, descriptions, calls to action
- Landing page changes: new page design or layout against the original
- Audience segments: adding or removing audience targeting
- Keyword match types: broad match vs phrase match performance
- Ad extensions/assets: testing different sitelinks, callouts, or structured snippets
How to Set Up a Google Ads Experiment: Step-by-Step
Step 1: Navigate to Experiments
In Google Ads, click “Experiments” in the left navigation (under “Campaigns”). Click the blue “+” button to create a new experiment. Select “Custom experiment” for Search campaigns or “Performance Max experiment” for PMax campaigns.
Step 2: Choose the Base Campaign
Select the campaign you want to test. Google will create a copy with identical settings. Important: the campaign must have enough historical data (at least 100 conversions in the past 30 days is recommended for reliable results).
Step 3: Make Your Changes
Modify only the variable you want to test. Testing multiple changes simultaneously makes it impossible to attribute results to any single factor. Examples:
- Change bidding from Manual CPC to Target CPA at ₹300
- Replace all responsive search ad headlines with new copy
- Swap final URLs to a new landing page
- Add an in-market audience as observation or targeting
Step 4: Set Traffic Split
Choose what percentage of traffic goes to the experiment. Common splits:
- 50/50: fastest results, equal exposure, but full impact if the experiment performs poorly
- 70/30 (original/experiment): conservative approach, limits risk while still gathering data
- 80/20: very conservative, useful for high-spend campaigns where you cannot afford significant budget on a losing variant
For most Indian campaigns spending ₹1,00,000–₹5,00,000/month, a 50/50 split provides the fastest, most reliable results. Below ₹50,000/month, you may need to run the experiment longer to reach statistical significance.
Step 5: Set Duration and Goals
Google recommends running experiments for at least 2–4 weeks. Set a start and end date. Define your primary success metric — conversions, CPA, ROAS, click-through rate, or conversion rate. Google will calculate statistical confidence for you.
Step 6: Launch and Monitor
Once launched, monitor the experiment dashboard. Google shows performance comparisons with confidence intervals. Do not make changes to either the original or experiment campaign during the test period — this invalidates results.
What Can You Test with Google Ads Experiments?
Bidding Strategy Tests
The most impactful experiment type for most advertisers. Common tests for Indian campaigns:
- Manual CPC → Target CPA: let Google’s AI set bids based on your conversion target
- Target CPA → Maximise Conversions: test whether removing the CPA cap increases total conversions within budget
- Target CPA at ₹250 vs ₹350: find the optimal target that balances volume and efficiency
- Target ROAS at 400% vs 600%: for e-commerce, find the right aggressiveness level
Ad Copy Tests
Test different messaging approaches:
- Price-led headlines (“Starting at ₹999/month”) vs benefit-led (“Grow Revenue 3x Faster”)
- Urgency language (“Limited offer”) vs trust language (“Trusted by 500+ businesses”)
- Different call-to-action text (“Get Free Quote” vs “Start Free Trial” vs “Book a Demo”)
Landing Page Tests
Send experiment traffic to a different landing page while keeping ads identical. Test:
- Long-form vs short-form page
- Video hero vs static image hero
- Single CTA vs multiple CTAs
- With pricing displayed vs hidden (requesting a quote)
Reading Experiment Results
The experiment dashboard shows:
- Performance comparison table: clicks, impressions, CTR, conversions, CPA, cost, ROAS for original vs experiment
- Statistical significance: shown as a confidence level (e.g., “95% confident the experiment performs better”). Google recommends 95% confidence before making decisions
- Direction arrows: green up arrow means the experiment outperforms the original for that metric; red down arrow means it underperforms
Important nuances:
- Do not end an experiment early just because one variant looks better in the first few days — wait for statistical significance
- If results are inconclusive after 4 weeks, the change likely has minimal impact — keep the original
- Look at secondary metrics too: a lower CPA but much lower conversion volume might not be a net win
Applying or Discarding Results
Once the experiment concludes with clear results:
- Apply the experiment: if the experiment wins, click “Apply” to push the experiment’s settings to the original campaign. Google will show which specific settings will change
- Discard: if the original wins, end the experiment. No changes are made to the original campaign
- Run a follow-up test: if results are close, test a different variation or run a larger-scale experiment
Budget Considerations for Indian Advertisers
During an experiment, your daily budget is shared between the original and experiment campaigns based on the traffic split. If your daily budget is ₹5,000 and you set a 50/50 split, each version gets approximately ₹2,500 in traffic.
Minimum recommended budgets for reliable experiment results:
- Lead generation campaigns: at least ₹1,00,000 total experiment spend (both variants combined) over the test period
- E-commerce campaigns: at least ₹2,00,000 total experiment spend for ROAS-based testing
- Low-volume campaigns (under ₹30,000/month): experiments may take 6–8 weeks to reach significance — consider testing at the ad group or ad level instead
Frequently Asked Questions
Can I run experiments on Performance Max campaigns?
Yes, Google introduced Performance Max experiments in 2023. You can test different asset groups, audience signals, bidding strategies, and final URL expansion settings. However, PMax experiments have fewer configurable variables than Search campaign experiments.
How long should I run a Google Ads experiment?
A minimum of 2 weeks, ideally 4 weeks. The exact duration depends on your traffic volume and conversion rate. Google will show a confidence level on the experiment dashboard. Wait until you see at least 95% confidence before making decisions. For Indian campaigns with lower budgets, 4–6 weeks is common.
Can I run multiple experiments on the same campaign simultaneously?
No. Google allows only one active experiment per campaign at a time. If you need to test multiple variables, run them sequentially. Alternatively, test variables at the ad group level (e.g., ad copy A/B tests within a single campaign) while running a campaign-level experiment on bidding.
Do experiments affect my Quality Score?
The experiment campaign inherits the original campaign’s Quality Score history. During the experiment, each variant’s Quality Score may diverge based on its own CTR and landing page performance. Applying an experiment does not reset Quality Score.
What happens to my data if I discard an experiment?
The experiment data remains in your Google Ads account for reporting purposes, but no changes are applied to the original campaign. You can reference historical experiment data for future planning. The original campaign continues as if the experiment never happened.
