Google Ads Experiments feature that lets you test campaign changes (bids, targeting, creatives) against a control to measure performance impact before full rollout.
What is A/B Testing (Experiments) in Google Ads? A/B Testing (Experiments): test campaign changes (bids, targeting, creatives) with 50/50 traffic split between control and treatment. Run 2-4 weeks until 95% statistical significance. Available for Search, Display, Video. Test one variable at a time. Scientifically validates changes before full rollout, preventing costly mistakes and uncovering unexpected winners.
A/B Testing in Google Ads (officially called "Experiments") allows advertisers to scientifically test campaign modifications by splitting traffic between a control (original campaign) and treatment (modified version), then measuring statistical differences in performance. Create experiments from original campaigns to test one variable at a time: bidding strategies (Manual CPC vs Target CPA), targeting changes (adding new audiences, geo expansion), creative variations (new ad copy, different CTAs), landing page tests (different URLs). Google splits traffic 50/50 (customizable) using cookie-based assignment (user sees same version consistently) or search-based assignment (each search randomly assigned). Experiments run until statistical significance is achieved (typically 2-4 weeks minimum, depending on traffic volume). Available for Search, Display, Video, Hotel, and Demand Gen campaigns (not App or Shopping). Can schedule up to 5 experiments per campaign but run only 1 at a time.
Official Source: Definition verified from Google Ads Help Center (Last verified: January 2026)
"Custom experiments let you test changes to your campaign's settings by creating a duplicate experiment campaign and running them both at the same time to compare results and understand which performs better."
A B2B software company wants to test whether switching from Manual CPC to Target CPA bidding improves efficiency. Current Manual CPC campaign: $6.50 CPC, 4.2% CVR, $155 CPA, 120 conversions/month. Wants scientific validation before changing bidding strategy.
Experiment Setup: - Original campaign (Control): Manual CPC, $6.50 avg CPC - Experiment campaign (Treatment): Target CPA bidding, $130 target (15% below current $155) - Traffic split: 50/50 - Duration: 30 days (need 60 conversions per arm for significance) - Hypothesis: Target CPA will reduce CPA by 15% while maintaining conversion volume Results after 30 days: Control (Manual CPC): - Clicks: 1,450 - CPC: $6.50 - Cost: $9,425 - Conversions: 62 - CVR: 4.3% - CPA: $152 Treatment (Target CPA $130): - Clicks: 1,680 (+16% more clicks) - CPC: $5.20 (-20% lower CPC) - Cost: $8,736 (-7% lower cost) - Conversions: 67 (+8% more conversions) - CVR: 4.0% (-0.3pp, statistically insignificant) - CPA: $130 (-14% lower CPA, statistically significant) Statistical Analysis: Google Ads Experiments Dashboard shows: - CPA difference: -$22 (-14.5%) - 95% Confidence interval: -$18 to -$26 - Statistical significance: YES ✓ - Recommended action: Apply to campaign Decision: Applied Target CPA $130 to full campaign. Post-experiment results (next 30 days, full campaign on Target CPA): - Conversions: 130 (+8% vs Manual CPC baseline) - CPA: $133 (-14% vs Manual CPC baseline, close to $130 target) - Total cost: $17,290 (saved ~$1,350/month vs continuing Manual CPC) - Annual savings: $16,200 Key insight: Experiment prevented blind rollout uncertainty. Statistical proof that Target CPA would improve CPA without harming conversion volume gave confidence to make permanent change. Without experiment, might have avoided switch due to fear of algorithm underperformance.
Experiments eliminate guesswork from optimization decisions by providing statistical proof that changes improve (or harm) performance. Instead of changing Target CPA from $50 to $40 across entire account and hoping for the best, experiments let you test $40 Target CPA on 50% of traffic while maintaining $50 on other 50%, then measure actual CPA difference with 95% confidence. This prevents costly mistakes: testing reveals that lowering Target CPA to $40 actually increases CPA to $55 due to reduced auction competitiveness, saving you from rolling out a harmful change. Experiments also uncover unexpected winners: adding Affinity Segments increases CPA by 8% but increases conversion volume by 45%, making it worthwhile despite higher cost. Without experiments, you'd roll back the change thinking it failed. Proper A/B testing compounds—running 1 experiment monthly identifying 10% improvements yields 12.7% compounding annual improvement (1.1^12).
Testing multiple variables simultaneously—changing both bidding strategy AND audience targeting in one experiment makes it impossible to know which change caused performance differences. Test one variable at a time for clear attribution.
Ending experiments too early before statistical significance—running experiment for only 3 days with "treatment looks better!" results in false conclusions. Wait for Google's significance indicator or minimum 95% confidence, typically 2-4 weeks depending on traffic.
Not monitoring for external factors—running experiment during Black Friday vs normal period, or while competitor launches major promotion, contaminates results with external variables. Control timing for clean tests.
Test one variable at a time for clear causality—Experiment 1: Test bidding strategy only (keep targeting, creatives, budgets identical). Experiment 2: After Experiment 1 concludes, test new audience layer. Sequential single-variable tests provide actionable insights.
Run experiments for minimum 2 full weeks and wait for statistical significance—Google provides confidence intervals in Experiments dashboard. Don't end experiment until: (1) minimum 2 weeks elapsed, (2) 95%+ confidence achieved, (3) meaningful difference observed (>5% improvement).
Use 50/50 traffic split for fastest statistical significance—uneven splits (80/20) take 4x longer to reach confidence. Use 50/50 unless you're risk-averse and want to minimize potential downside from treatment.
Document and apply learnings systematically—maintain experiment log: date, hypothesis, variable tested, result, decision (applied/rejected/iterate). Successful experiments become best practices applied account-wide.
An automated bidding strategy that sets bids to get as many conversions as possible at your target cost per acquisition.
An automated bidding strategy that sets bids to maximize conversion value while targeting a specific return on ad spend percentage.
A subset of automated bid strategies that use machine learning to optimize bids for conversions or conversion value in every auction.
Primary Search ad format with 15 headlines and 4 descriptions that Google automatically tests and optimizes combinations for best performance.
The percentage of ad clicks that result in a conversion (purchase, lead, signup, etc.), calculated as conversions divided by clicks.
Run experiments until you achieve both minimum duration AND statistical significance: Minimum duration: 2 full weeks (14 days) to account for weekly seasonality patterns (weekday vs weekend performance). Campaigns with day-of-week performance variation (B2B peaking Tuesday-Thursday) need minimum 2 weeks to capture full weekly cycle. Statistical significance: Wait for 95% confidence interval in Google Ads Experiments dashboard before concluding. This typically requires 50-100 conversions per experiment arm (control + treatment combined need 100-200 conversions). Formula: Time needed = (Conversions required) ÷ (Current conversion rate × Daily traffic ÷ 2). Example: Campaign gets 100 clicks/day, 5% CVR = 5 conversions/day. Need 100 conversions (50 per arm) = 20 days. High-traffic campaigns (1,000+ clicks/day, 50+ conversions/day): 2-3 weeks sufficient. Medium-traffic campaigns (100-500 clicks/day, 5-25 conversions/day): 3-4 weeks needed. Low-traffic campaigns (<100 clicks/day, <5 conversions/day): 6-8 weeks or more, consider testing higher-traffic campaigns first. When to end early: If treatment shows statistically significant HARM after 2 weeks (CPA increased 30% with 95% confidence), end experiment early and reject treatment to minimize damage. Never end early because treatment "looks promising"—wait for significance. External events exception: If major external event occurs mid-experiment (surprise competitor sale, industry news, holiday), pause or restart experiment after event concludes to avoid contaminated results. Best practice: Set experiment end date 30 days out initially. After 2 weeks, check significance daily. End when significance achieved or 30 days elapsed, whichever comes first. If 30 days pass without significance, experiment is inconclusive (likely no meaningful difference, or insufficient traffic).
Get a complete audit of your Google Ads account and see exactly where you stand on A/B Testing (Experiments) and 46 other critical factors.
Results in under 3 minutes. No account access required.