Most sites need roughly 1,000 conversions per variation per month to run a trustworthy A/B test. On a page converting at 2 percent, that works out to about 50,000 visitors per variation, or 100,000 visitors a month for a simple two-version test. If your traffic is well below that, an A/B test will not give you a reliable answer, and the honest move is to fix known conversion problems directly instead of waiting on a test that cannot reach significance.
That rule of thumb is deliberately blunt, and the real number depends on your baseline conversion rate and how big a change you are trying to detect. Here is how to work out your own number, why most published test results are wrong, and what to do when the math says you do not have the traffic. Last updated July 2026.
How much traffic do I need for an A/B test?
Three things drive sample size: your current conversion rate, the size of the improvement you want to detect, and how confident you want to be that the result is real.
The counterintuitive part is the second one. Detecting a big lift is easy and needs little traffic. Detecting a small lift is brutally expensive. Because the effect you are looking for enters the math squared, halving the lift you want to detect roughly quadruples the traffic you need. This is the single most misunderstood fact in A/B testing, and it is why so many test programs quietly fail.
| Baseline conversion rate | Lift you want to detect | Approx. visitors needed per variation |
|---|---|---|
| 2% | 50% (2% to 3%) | About 2,300 |
| 2% | 20% (2% to 2.4%) | About 13,000 |
| 2% | 10% (2% to 2.2%) | About 51,000 |
| 2% | 5% (2% to 2.1%) | About 200,000 |
| 5% | 10% (5% to 5.5%) | About 19,000 |
| 10% | 10% (10% to 11%) | About 8,500 |
These are approximations at 95 percent confidence and 80 percent statistical power, the standard settings in most calculators. Treat them as the right order of magnitude rather than exact figures, and run your own numbers in a sample size calculator before committing to a test.
Read the middle rows carefully, because that is where reality bites. A realistic 10 percent lift on a 2 percent page needs about 51,000 visitors per variation. Two variations means 102,000 visitors. If you get 10,000 visitors a month, that test takes ten months, by which point your traffic mix, your pricing, and probably your product have changed enough to invalidate the comparison.
Why higher-converting pages are cheaper to test
Notice that a page converting at 10 percent needs about 8,500 visitors per variation to detect the same relative lift that costs a 2 percent page around 51,000. Rare events need more observations to measure. This has a practical consequence people miss: test deeper in the funnel where rates are higher. Testing a checkout step that converts at 40 percent is far more tractable than testing a top-of-funnel landing page at 1.5 percent, even though the landing page gets more traffic.
How long should an A/B test run?
At minimum one full week, and preferably two, regardless of what the sample size math says. Traffic behaves differently on weekdays than weekends, and B2B traffic in particular collapses on Saturday and Sunday. A test that runs Monday to Thursday is measuring a slice of your audience, not your audience. Full weeks also protect you from campaign spikes and payday effects.
The other rule is to decide the duration before you start and then leave it alone. Which brings us to the most common way tests lie.
Peeking is why your last test was wrong
If you check a running test daily and stop it the moment it shows significance, you have broken the statistics. Classic significance testing assumes you fixed the sample size in advance and looked once at the end. Every extra look is another chance for random noise to cross the threshold, and with enough peeking a test between two identical pages will eventually show a "winner." Some studies put the false positive rate from aggressive peeking near 30 percent, against the 5 percent you thought you were accepting.
This is the quiet reason so many wins fail to replicate. The test was called early, on a random fluctuation, by someone excited that the new version was up 18 percent on day three. Calculate the sample size, run it to completion, then look. If you genuinely need to monitor continuously, use a tool with sequential testing or Bayesian methods designed for it, rather than pretending a fixed-horizon test tolerates peeking.
What if I do not have the traffic?
This is where most people actually are, and there is no clever workaround that makes an underpowered test valid. You have four honest options.
Test bigger changes. If you can only detect a 50 percent lift, then stop testing button colors and test genuinely different propositions: a new headline and offer, not a new shade of blue. Small changes produce small effects, and small effects are the expensive ones to measure. Radical redesigns are the only tests that make sense at low volume.
Test higher in the rate, not higher in the funnel. As above, move the test to a step that converts at 20 or 40 percent. You will need a fraction of the traffic for the same reliability.
Skip testing and fix what is knowably broken. This is the option people resist because it feels less rigorous, but it is usually correct at low volume. A lot of conversion problems are not close calls that need an experiment to settle. A headline that describes your company instead of the visitor's outcome, three competing CTAs above the fold, a form asking eleven questions before offering any value, a five-second mobile load: none of these need a test to adjudicate. They need fixing. Testing is for deciding between two plausible options. It is not for discovering that your page is unclear.
Get more traffic. If the honest diagnosis is that you have 3,000 visitors a month, no testing tool solves that. The constraint is upstream, and the durable fix is compounding organic traffic rather than a bigger ad budget, which is why so many small teams end up publishing search-focused articles on a schedule long before they ever need an experimentation platform. Come back to testing when the volume is there.
Do I need an A/B testing tool at all?
Probably not yet, if you are under the thresholds above. Testing platforms are priced for teams that have the traffic to use them: Convert Experiences starts around $299 a month, and VWO, AB Tasty, and Optimizely are sales-quoted annual contracts commonly running into five figures. Paying that to run tests that cannot reach significance is the most common waste in the CRO budget.
The sequence that actually works for a small site is unglamorous. Get free behavior data from Microsoft Clarity to see where people drop off. Fix the obvious clarity, CTA, and speed problems directly, without testing them. Grow the traffic. Then, when you have enough volume that a test can resolve in two weeks, buy the testing tool and start settling genuine close calls. Our A/B testing tools comparison covers the field, and the best CRO tools roundup maps which tool does which job.
For the middle step, fixing what is knowably broken, Converto audits a live URL and generates the improved copy, CTAs, and layout for you, from $19 a month with nothing to install. No sample size required, because it is not a test: it reads the page against known conversion patterns and hands you a specific change to ship. Suggestions are data-informed, not guarantees. If you want the numbers behind the tooling decision, we break down how much CRO costs across tools, agencies, and consultants.
The short version
Work out your own number rather than trusting a rule of thumb, but as a starting point: about 1,000 conversions per variation per month, or roughly 50,000 visitors per variation on a 2 percent page, for a realistic 10 percent lift at 95 percent confidence. Run full weeks. Never peek and stop early. And if the math says you do not have the traffic, believe it: fix the page directly, grow the volume, and buy the testing platform when it can actually earn its price.
You can paste your page into Converto and see what the audit flags in about a minute.
Paste your page and see what is leaking
Drop in any live URL and Converto audits the copy, layout, CTAs, speed, and AI-readability of the page you already have, then hands you ready-to-ship variants to lift signups and sales. Suggestions are data-informed, not guarantees.