How Long to Run a Google Ads A/B Test (and When to Call It)

Most Google Ads A/B tests need at least four weeks, but the calendar is a stand-in for something else: enough clicks and conversions, collected across full weekly cycles, to tell a real difference from noise. A test on a campaign with 200 conversions a week can be readable much faster than one with 10. So how long to run a Google Ads A/B test depends on your volume, your bidding strategy's learning period, and what you're changing.

Below is how to set a test length before you launch, what Google's own experiment tools do and don't tell you, and the signals that say keep waiting or stop.

Why Enough Data Beats a Fixed Number of Days

A test is valid when it has collected enough data to separate a real effect from random variation. Days only matter because data accumulates over them. Fourteen days with 8 conversions per variant proves very little. Fourteen days with 150 conversions per variant might prove plenty.

For most accounts, a starting range of about 4 to 6 weeks works well. Treat that as a practitioner rule of thumb, not a Google requirement. As of 2026, Google Ads Help describes experiments and their reporting, but it does not set one mandatory duration for every test. Check the current experiments documentation in Google Ads Help before you lock in a plan.

Two terms matter here:

• Statistical significance means the gap between your variants is unlikely to be pure chance. It is expressed through a confidence level. At 95% confidence, you accept roughly a 1 in 20 risk that a difference this large would show up even if the variants performed the same.
• Confidence level is the threshold you choose for that risk. Higher confidence needs more data.

It helps to separate a test that is done from one that is merely early. A done test has hit the minimum run time and conversion count you set in advance, and its result is stable. An early test just has a leader. Leaders flip, especially in the first week, so a lead alone is not a result.

Also note that significance is not the same as importance. A statistically solid 1% lift in conversion rate may not justify the effort of changing anything. Decide up front what size of difference would matter to the business.

What Sets Your Test Length: Volume, Conversion Rate, and Learning Periods

Three things drive how long you'll wait: how many conversions you get, how much bidding needs to settle, and how your customers behave across the week.

Conversion volume

Here is an illustration, not a benchmark. Suppose your control and your variant each get 10 conversions a week. After four weeks you'd have about 40 per arm, and a gap like 40 versus 48 is well within normal swings. Now suppose each arm gets 100 conversions a week. After one week, 100 versus 120 is a much more meaningful gap. Same difference in relative terms, very different reliability. Low-volume accounts either wait much longer or test a metric that occurs more often, such as clicks or CTR, accepting that it's a weaker stand-in for revenue.

Learning periods

Smart Bidding strategies like Target CPA and Target ROAS go through a learning period after you create them or change them significantly. Performance can swing during that time as the system calibrates. If your test changes the bid strategy, targets, or conversion actions, that settling time counts against your test length. Google Ads Help documents learning periods for Smart Bidding, and the guidance has been revised over time, so read the current page rather than relying on a number you remember. A safe habit: don't begin judging results until the learning period is over for the experiment arm.

Weekly cycles and conversion lag

Most accounts behave differently on Tuesday than on Saturday. B2B often peaks midweek, and local services or retail might peak on weekends. Start and end tests on the same weekday and run whole weeks so each variant sees every day equally.

Conversion lag is the delay between a click and the conversion being recorded. If your sales cycle is two weeks, the last week of clicks in your test hasn't finished converting when the test ends, and recent data will undercount both variants. Check your conversion delay in your reports, and let the test run past the end of the window long enough for late conversions to arrive before you read results.

Which Type of Test You're Running Changes the Clock

"A/B test" covers several different things in Google Ads, and each one has its own bottleneck.

Ad copy tests

These are driven first by impressions and CTR, because clicks arrive far faster than conversions. You can often see a CTR trend sooner than a conversion trend. But CTR isn't revenue. A headline that attracts more clicks from the wrong people can raise CTR and lower conversion rate, so still wait for conversions before declaring a winner on anything that matters.

Responsive search ads complicate this. Google mixes headline and description assets into combinations, so you can't cleanly isolate one headline the way you could with a fixed expanded text ad. Asset-level reporting shows how assets perform, but the combination served varies per auction. Check the current Google Ads Help pages on RSA asset reporting and any ad-level experiment options to see what is measurable in your account. If you need a clean comparison, test two distinct ad concepts rather than a single swapped line.

Campaign experiments

Google Ads Experiments (the successor to the older drafts and experiments tool) lets you test changes to bids, settings, or landing pages against a base campaign, splitting traffic between them. Setup options, including the traffic split percentage and whether the split is by search or cookie, are defined in current Google Ads Help, so confirm them as of 2026 before you configure anything. A 50/50 split generally gives the cleanest and fastest comparison, since each arm collects data at the same pace. A lopsided split means the smaller arm takes much longer to reach a usable sample.

These tests usually need longer than ad copy tests because the metric you care about is conversions, cost per conversion, or return on ad spend, and that data is slower and noisier than clicks.

Landing page tests

These hinge on conversion rate, and conversion rate differences tend to be small. Detecting a small difference in a low-percentage event needs the most volume of any test type. If a page converts at a few percent, you may need hundreds of clicks per variant just to see a modest lift clearly. Plan for the longest runtime here, or test a bolder change that produces a bigger effect.

Mistakes That Make a Test Run Too Short or Too Long

Most bad test calls come from a handful of habits.

• Peeking and stopping on the first lead. Early results swing hard. Checking daily and ending the test when one variant pulls ahead is the fastest way to crown a false winner. Look at the data if you must, but don't act until your pre-set conditions are met.
• Editing mid-test. Changing budgets, bids, targeting, or the conversion actions in either arm means the two sides are no longer comparable. If a change is truly unavoidable, note the date and treat it as the start of a new test.
• Running forever with no decision rule. A test that never ends doesn't produce a decision. It also drifts: auction conditions, competitors, and seasonality change under it. If you reach your thresholds and the result is flat, that is a result.
• Ignoring outside events. A promotion, a holiday, a stock-out, a pricing change, or a competitor's sale can hit both arms unevenly or distort the whole window. Annotate your account when these happen and consider extending or restarting the test.

Then there's the misconception that Google picks the winner for you. Experiment reports in Google Ads show how each arm performed and may flag results that are statistically significant, but a flag is not a business decision. You still need to check that the primary metric moved in the direction you wanted, that cost and conversion quality held up, and that the result wasn't driven by one of the outside events above. Applying or ending an experiment is your call.

A Simple Rule for Deciding When to Stop

Before you launch, write down three things:

1. One primary metric. Cost per conversion, conversion rate, or ROAS, not all three. Others are watched, not decisive.
2. A minimum run time in full weeks. Four is a reasonable floor for most accounts, longer if you have a learning period or long conversion lag.
3. A minimum conversion count per variant. Pick a number you're comfortable defending, and don't read the result until both arms reach it.

A made-up example

Suppose you test a new landing page and commit to: primary metric of cost per lead, at least four full weeks, and at least 100 conversions per arm. At week three, the new page looks 25% cheaper, but only 60 conversions per arm have come in. Decision: continue. At week five, both arms have more than 100 conversions, and the new page is 12% cheaper with a consistent edge across weeks. Decision: stop and adopt it, after confirming lead quality didn't drop. If instead the gap were 2% and flipped between weeks, the call is inconclusive.

If it's inconclusive

Keep the control. Change one thing, ideally something bolder, and run it again. An inconclusive test still tells you that the change you made probably isn't a big lever.

Keep the data clean

Junk queries can skew both arms unevenly, especially in a test on ads or landing pages. Review the Search Terms Report before launch and throughout, add negative keywords for irrelevant search terms, and add the strong ones as positive keywords. Keywordme is a Chrome extension that lets you do this inside the Search Terms Report with one-click actions and match types, so you don't need a spreadsheet. It cleans up queries; it doesn't run tests or pick winners.

Set the Stopping Rule First, Then Let the Weeks Run

Write your stopping rule before you launch: one primary metric, full weeks, and a conversion minimum for each variant. Then leave the test alone until all three are met. That discipline matters more than any specific number of days.

Between tests, keep your search terms tidy so the next comparison starts from clean traffic.

Optimize Google Ads Campaigns 10X Faster. Without Leaving Your Account. Keywordme lets you remove junk search terms, build high-intent keyword lists, and apply match types instantly, right inside Google Ads. No spreadsheets, no switching tabs, just quick, seamless optimization. Start your free 7-day trial (then just $12/month) and take your Google Ads game to the next level.