At a glance
- At the Meta Ads median CTR of 2.43%, spotting a 20% lift needs 17,289 impressions per version. A 10% lift needs 66,098.
- At the Meta Ads median conversion rate of 1.32%, a 20% lift needs 32,228 clicks per version.
- Accounts vary widely: the middle half of Meta Ads accounts had a CTR between 1.43% and 3.97%, so use your own baseline when you have it.
Enter what each version received and what it got back. For a CTR test that is impressions and clicks. For a conversion rate test it is clicks and conversions.
Version A rate
-
Version B rate
-
Relative lift
-
z-score
-
p-value
-
Pick a baseline rate, then the smallest relative lift worth acting on. Add each version's daily volume to see how many days the test needs.
Sample per version
-
Both versions
-
Days to run
-
Two-sided pooled z-test. Significant at 95% when the p-value is under 0.05. Sample size uses 95% confidence and 80% power.
1. Baseline rates for ad tests, October 2026
Each row is a platform-level cell from GoMarble's October 2026 monthly benchmarks, across all spend levels. The calculator's presets use the medians. If you know roughly where your account sits, start from the nearer quartile instead.
| Platform | Metric | 25th percentile | Median | 75th percentile | Ad accounts |
|---|---|---|---|---|---|
| Meta Ads | CTR (all clicks) | 1.43% | 2.43% | 3.97% | 880 |
| Meta Ads | Conversion rate | 0.71% | 1.32% | 2.78% | 440 |
| Google Ads | CTR (all clicks) | 1.72% | 3.58% | 7.68% | 970 |
| Google Ads | Conversion rate | 2.31% | 5.05% | 11.24% | 800 |
CTR counts all clicks. Conversion rate uses only accounts with at least 10 purchases (Meta) or conversions (Google) in the month.
Baseline matters because the sample grows as the rate shrinks. The table below shows the sample each version needs at the medians, at 95% confidence and 80% power.
| Baseline | Lift to detect | Sample per version |
|---|---|---|
| Meta Ads CTR, 2.43% | 10% | 66,098 impressions |
| Meta Ads CTR, 2.43% | 20% | 17,289 impressions |
| Google Ads CTR, 3.58% | 15% | 20,143 impressions |
| Meta Ads conversion rate, 1.32% | 20% | 32,228 clicks |
| Google Ads conversion rate, 5.05% | 10% | 30,908 clicks |
Conversion rate tests take longer than CTR tests. A conversion only happens after a click, so the sample is counted in clicks, and clicks are scarcer than impressions. At 2,500 impressions a day per version, the 20% CTR test at the Meta median finishes in 7 days.
Download the four baseline cells as a CSV (ad-test-baselines-2026-10.csv).
2. Quick answers
How many impressions do I need for an ad A/B test?
It depends on your baseline CTR and the smallest lift you care about. At the Meta Ads median CTR of 2.43% in GoMarble's October 2026 benchmarks, detecting a 20% lift needs 17,289 impressions per version, at 95% confidence and 80% power. Detecting a 10% lift needs 66,098. Halving the lift roughly quadruples the sample.
How long should I run a Facebook ad test?
Until each version reaches its planned sample, and ideally a full week so weekday swings even out. At 2,500 impressions a day per version, a test sized for a 20% lift on the Meta Ads median CTR of 2.43% from GoMarble's benchmarks needs 7 days. Stopping early when one ad looks ahead creates false winners.
Is my A/B test result statistically significant?
Enter each version's sample and results in the first tab. The calculator runs a two-sided z-test and calls a result significant at 95% when the p-value is under 0.05. For example, 290 clicks against 243 from 10,000 impressions each is a +19.3% lift with a p-value of 0.0391, which is significant.
What conversion rate should I assume when planning a test?
Your own, if you have a month of data. Otherwise start from a benchmark. In GoMarble's October 2026 data, the median conversion rate was 1.32% across 440 Meta Ads accounts and 5.05% across 800 Google Ads accounts. At the Meta median, detecting a 20% lift needs 32,228 clicks per version.
3. How the calculator works
The significance tab compares two rates with a pooled two-proportion z-test. The sample size tab uses the standard formula for two proportions, at 95% confidence (two-sided) and 80% power.
Step by step, for a finished test:
- Rates. Divide each version's clicks by its impressions: 243 ÷ 10,000 = 2.43% and 290 ÷ 10,000 = 2.90%.
- Pooled rate. Add both versions' clicks and divide by both versions' impressions.
- Standard error. The square root of pooled rate × (1 − pooled rate) × (1 ÷ sample A + 1 ÷ sample B).
- z-score and p-value. The gap in rates divided by the standard error gives z = 2.063. The two-sided p-value is 0.0391, under 0.05, so the result is significant at 95%.
Relative lift is (rate B − rate A) ÷ rate A: here +19.3%. A small sample can show a big lift that is not significant. With 5,000 impressions each, 2.40% against 3.00% is a +25.0% lift, yet the p-value is 0.0642.
4. Running an ad test the calculator can trust
The maths assumes each person sees only one version, so the two samples are independent. Meta's own A/B test tool splits the audience to make that hold, while two ad sets in one campaign can overlap, so treat that result as a guide.
- Test one change at a time. Hook, image or offer, not all three, or the winner teaches you nothing reusable. The video ad hooks guide lists hook types worth testing.
- Size the test before launch. Pick the smallest lift worth acting on, then run until both versions reach the sample the second tab gives.
- Do not stop at the first significant reading. Checking daily and stopping early raises the chance of a false winner well above 5%.
- Match the metric to the decision. CTR tests finish fast but say nothing about sales. Conversion rate tests need clicks, so they run longer.
Know your own baseline first. The CTR calculator places your click rate against the same October 2026 sample. The CPA calculator does the same for conversion rate and cost per acquisition.
5. Methodology
The baseline rows on this page are four platform-level cells from GoMarble's monthly benchmark file for October 2026. The test maths is standard statistics, not GoMarble data.
- Sample: 1,900 ad accounts from 480 companies, with $6.3M of USD spend analysed. CTR cells cover 880 Meta Ads and 970 Google Ads accounts. Conversion rate cells cover 440 and 800 accounts with at least 10 purchases or conversions.
- Period: October 2026, 1 to 31 October. File generated on 12 November 2026.
- Method: Account-level metrics per month, deduplicated across GoMarble teams, from Meta and Google Ads campaign data. Quartiles are across ad accounts. CTR counts all clicks. Cost metrics (CPM, CPC, CPA) use USD-denominated accounts only. ROAS, conversion rate and CPA use only accounts with at least 10 purchases (Meta) or conversions (Google) in the month, valued as each platform reports them.
- Deduplication: metrics are account-level per month and deduplicated across GoMarble teams.
- Currency: cost metrics (CPM, CPC, CPA) use USD-denominated accounts only.
- Privacy thresholds: a cell is published only with at least 20 ad accounts from at least 10 companies, and no account above 25% of its spend. In this release 6 cells passed and 4 were suppressed, not estimated.
- Scope: no industry split exists in this release, and no customer, agency or brand is named.
Cite this
https://www.gomarble.ai/ai-tools/ad-ab-test-calculator/
The four baseline cells are free to reuse under CC BY 4.0, credited to GoMarble with a link to this calculator. License · Download the CSV
FAQ
What confidence level does the calculator use?
Should I count impressions or clicks as the sample?
Why does a small lift need such a big sample?
Does Meta's A/B test tool replace this?
Can I test more than two versions?
Are the baselines spend-weighted?
Read next
Find the ads worth testing next.
Connect Meta Ads and Google Ads, and GoMarble AI Agents analyse creative and diagnose performance by campaign, then propose changes for your approval.