Ad Creative Testing: 7 Days Minimum, 30 Days Max

Seven days minimum, thirty days maximum. Both TikTok and Meta publish that same window for a creative split test, on two separate help pages, for two separate products. Under seven days you are reading noise. Past thirty the platform ends the test for you. The harder question is what has to be true before day one, because that is what decides whether the answer is usable at all.

Below: what each platform actually publishes with its page and date, the settings that have to be locked before the test starts, how far apart two creatives need to be, and a short checklist for deciding after the fact whether a finished test told you anything.

What is an ad creative test?

An ad creative test is a comparison where creative is the only thing that changes. Audience, placement, budget, optimisation event, landing page and schedule are held identical, and the platform splits traffic into groups that do not overlap. If any second variable moves, the result describes the pair of changes and cannot be attributed to the creative.

That is the whole discipline. Everything below is about protecting the one-variable condition against the ordinary pressure to fiddle with a campaign while it runs.

How long should an ad creative test run?

TikTok's split test guidance, last updated January 2026, recommends a minimum of seven days and caps a split test at thirty. It says explicitly that testing under seven days may not be long enough to determine a winner, and that running long risks a budget shortfall. That is on TikTok's split test best practices page.

Meta publishes the same two boundaries for its creative A/B testing tool. Its FAQ, updated 16 December 2024, states that tests run for a maximum of thirty days and that statistical significance can be reached in seven days at the earliest. Worth being precise about scope: that page documents the creative A/B testing tool for Meta Horizon Store product detail pages, not Ads Manager. It is a Meta-owned, dated statement of how Meta runs a creative test, which is why it is quoted here, and it is not a statement about your ad account.

What each platform publishes, checked 16 September 2026

• TikTok minimum duration: 7 days. Source: split test best practices, last updated January 2026.
• TikTok maximum duration: 30 days. Same page.
• TikTok target statistical power: at least 80 percent, shown as Estimated Testing Power while you set the budget. Same page.
• Meta Horizon creative A/B test maximum duration: 30 days. Source: creative A/B testing FAQ, updated 16 December 2024.
• Meta Horizon earliest significance: 7 days. Same page.
• Meta Horizon significance bar: error rate under 5 percent, a 95 percent confidence interval. Same page.
• Meta Horizon audience split options: 10, 25 or 50 percent, with 50/50 described as reaching significance faster. Same page.

The convergence is the useful part. Two platforms, two different testing products, same seven and thirty. Seven days is not a superstition about weekly seasonality, it is roughly the floor at which either system will call anything.

What has to be set before the test starts?

A test fails at setup far more often than it fails at analysis. Six things to fix before anything goes live.

1. Write the hypothesis as a sentence with a predicted direction. Not "test hooks" but "the problem-state open will beat the product open on cost per purchase." A test with no prediction cannot be wrong, which means it cannot be informative.
2. Pick one metric and commit to it in writing. Cost per purchase and click through rate frequently disagree, and choosing after the fact is how a losing test gets reported as a win.
3. Fund the window, not the day. Budget has to carry the full seven days at minimum. TikTok shows an Estimated Testing Power figure as you set budget and recommends a budget that gets you to at least 80 percent.
4. Check the platform minimums. TikTok's budget page, last updated July 2026, requires a daily budget above $50 at campaign level and above $20 at ad group level, and computes a lifetime ad group minimum as total days times the minimum daily figure.
5. Lock the naming before launch, not after. On both platforms the ad name is written into the tracking link at publish time, so a rename mid-test does not update what already shipped.
6. Decide in advance what you will do with each outcome. If the answer changes nothing you were going to do, the test is not worth the window it costs.

TikTok's stated budget minimums and the lifetime calculation are on its page about budgets in TikTok Ads Manager. The naming point, and the tokens that get baked into the URL at publish, are covered in this breakdown of Meta and TikTok ad naming conventions.

How different do the two creatives have to be?

Far more different than most teams make them. TikTok's split test guidance says to set up test groups with large differences in variables, and that there should be obvious differences so the two ad groups do not produce similar results and the system can pick a winner. Its creative best practices page, June 2025, repeats it: it is always better to use creatives with big differences, especially when testing.

Two variants that differ by an adjective will return two results inside each other's margin, and the test spends its full window to report that nothing happened. A change of opening premise, a change of format, or a change of proof type will separate. A change of word will not.

The practical way to guarantee separation is to test openings built on different objections rather than different phrasings. Twenty scripts organised that way are in this set of UGC ad hook scripts by product category.

Why do most creative tests produce an unusable answer?

• The test was edited while running. TikTok's guidance is explicit that changing an ad group after a split test starts may affect results or send the ad group back into review.
• Two variables moved. Meta's FAQ notes that when you test more than one variable you will not know which asset produced the win. That applies whether the tool enforces it or not.
• The budget could not carry the window, so the test was stopped early and read anyway.
• The variants were too similar to separate, which costs a full window to learn nothing.
• The metric was chosen after the numbers came in.
• Two tests ran at once on overlapping audiences. Meta's creative testing tool permits only one test at a time per app precisely to stop user groups overlapping.

What about the confidence percentages you see quoted?

A 65 percent confidence threshold for Ads Manager A/B tests circulates widely in ad-industry writing, usually stated without a link. It is worth being careful with it. On the Meta-owned pages reachable in September 2026, the published significance standard for a Meta creative A/B test is the one quoted above: an error rate under 5 percent, a 95 percent confidence interval, and that is the Horizon Store tool rather than Ads Manager.

The honest position is that Ads Manager displays a confidence figure inside the test results view, and the number in front of you in your own account is the one to act on. Repeating a threshold from a blog post that does not cite Meta is how a team ends up shipping on a result the platform never called.

This matters more than it sounds. The gap between a 65 percent bar and a 95 percent bar is the difference between acting on roughly a one in three chance of being wrong and roughly one in twenty.

Did this test produce a usable answer?

Run this after the test closes and before anyone reports it. Any no means the result is directional at best.

• Did the test run at least seven full days without being paused or edited?
• Was creative the only thing that differed between groups?
• Was the winning metric the one written down before launch?
• Did the platform itself call a winner, rather than a human reading a dashboard?
• Were the variants different enough that a difference was detectable at all?
• Did the budget actually deliver across the whole window in both groups?
• Would the opposite result have changed what you do next?

If a test fails the checklist, the right move is to rerun it properly rather than to argue about the numbers. A rerun costs one window. A creative strategy built on a broken test costs every window after it.

How does testing relate to refreshing creative?

They are different jobs and the second one runs continuously. TikTok's creative guidance recommends refreshing ad group creatives when delivery results show a consistently declining trend, and specifically advises adding new creatives to an existing ad group rather than building a new one, so the ad group keeps its history. That is on the same creative best practices page. How many variants that appetite actually consumes per month is worked through in this look at ad creative fatigue and how many Meta ads you need.

Frequently asked questions

How long should a Meta or TikTok creative test run?

At least seven days, and no more than thirty. TikTok publishes both boundaries on its split test guidance, last updated January 2026. Meta's creative A/B testing FAQ, updated December 2024, gives thirty days as the maximum and seven days as the earliest point significance can be reached.

Can I stop a creative test early if one side is clearly winning?

You can, and the result stops being a test result. Early leads reverse routinely inside the first week because the delivery system is still distributing spend unevenly. If you stop early, treat the outcome as a hunch worth retesting rather than an answer worth scaling.

How much budget does a creative test need?

Enough to fund the minimum window in both groups at once. TikTok's published minimums are above $50 daily at campaign level and above $20 at ad group level, and it shows an Estimated Testing Power reading as you set budget, recommending a figure that reaches at least 80 percent.

Can I test more than one thing at a time?

Only if you accept not knowing which change won. Meta's FAQ states directly that testing multiple variables leaves you unable to tell which asset produced the result. That is sometimes acceptable, for example when a campaign requires several assets to change together.

What if the platform never calls a winner?

That is itself a finding: the two variants performed within each other's margin. Usually the cause is variants that were too similar to separate, or a budget too small for the window. Rebuild the test with a genuinely different premise rather than extending the same one.

Key takeaways

• Seven days minimum and thirty days maximum, published independently by TikTok in January 2026 and by Meta in December 2024.
• Write the hypothesis and the single winning metric down before launch, not after the numbers arrive.
• TikTok recommends an Estimated Testing Power of at least 80 percent, and shows the figure while you set the budget.
• Make the variants obviously different. Both TikTok pages say so, and near identical variants waste a full window.
• Do not edit an ad group mid-test. TikTok warns it may affect results or trigger a fresh review.
• Treat the widely quoted 65 percent confidence figure as unverified, and act on the number your own Ads Manager displays.
• A test that would not have changed your next move was not worth the window.

Where the constraint usually sits

Most teams do not have a testing problem, they have a supply problem: seven days per test with genuinely distinct variants is a production rate, not an analysis skill. That is the bottleneck most DTC and ecommerce brands hit first. Specs for the assets themselves, where the published numbers do not always agree, are checked in this review of Meta's ad specs.

EchoPulse Media produces that volume as a standing pipeline rather than a project. The 14 day paid Pilot is $299, you keep everything produced, and nothing rolls into a retainer unless you ask it to.