KloveraBeta Log in Get started
Blog

How do you test video ads properly?

Adam Ginsberg, Founder and CEO of Klovera.ai, Super Intelligence Software for Everyone.

18 November 2026

To test a video ad properly, run two or more versions at the same time, change only one thing between them, and split the audience so each person sees only one version. Then wait until the platform says the result is conclusive, which on Google can take weeks, not days.

Why isn't one video a test?

One video has nothing to be compared with. If it does badly, you can't tell whether the hook, the presenter, the offer, the audience or the week was to blame. If it does well, you can't tell which part earned it, so you can't repeat it on purpose.

A test needs a comparison, and the comparison has to be fair. That's how the ad platforms build their own testing tools, and their help pages spell it out.

What the platforms mean by a test

Google's help page Create a video experiment starts before any video is made: "Start with a hypothesis to determine your testing strategy." A hypothesis is simply the question your test will answer.

For a video seller, it can be as plain as: a question hook beats a product-first hook for this product. Or: this presenter holds attention longer than that one. Write it down before you post, so the result answers a question you actually asked.

TikTok describes its tool the same way. Its page About Split Testing (last updated January 2026) says: "Split testing in TikTok Ads Manager allows you to test two different versions of your ads". The variables it lets you test include targeting, placement, bidding and optimization, budget strategy and creative.

How many things should you change at once?

One. Google's page says: "We recommend one variable (such as audiences) to keep the experiment clean."

In practice, keep the script, the product, the length and the audience the same, and change only the hook. Or keep everything and change only the presenter. If you change the hook and the presenter together and one version wins, you won't know which change did it.

This doesn't limit you to two videos. Google lets you add more versions to one experiment: "Select campaigns for your treatment arms (up to 10 arms total)." Each version still differs from the original in the one thing you're testing.

Split the audience, don't take turns

Posting version A this week and version B next week isn't a fair test. The audience, the season and the competition all change between the two weeks.

Both platforms run the versions side by side instead. Google says: "When an experiment is running, traffic is split evenly between the campaigns." TikTok goes further and keeps the groups apart, so each audience group "will exclusively see one ad group". That way the two versions aren't competing for the same people.

How long should a video ad test run?

Longer than most people expect. Google's help page on the Experiments page says that while a video experiment's results are still pending, "we recommend allowing it to run for at least 4 to 6 weeks to collect enough data to evaluate the results." If there's still no clear answer, it adds that "you may need to adjust the budget or allow it to run for a longer duration."

The bar for a result is set in advance, too. On Google: "If you choose to wait for conclusive results, those will be at a 95% confidence interval." TikTok's page says its tool "is scientifically designed to perform an accurate A/B test with a 90% confidence rate". Google allows earlier, directional readings, but its advice is plain: "We recommend waiting until the experiment finishes for conclusive results".

A version that looks ahead after two days is a hint, not an answer.

What if neither version wins?

That's a result too. Google's page Monitor the performance of a video experiment says: "Inconclusive results can also be insightful". When a version shows similar performance, Google means that "there is no statistical evidence that the arm is performing better or worse" than the original.

Google's own example is useful for anyone paying for content: you may have two videos that do equally well, "but one of the creatives may be cheaper to produce than the other." In that case, the cheaper one is the one to keep making.

When a version does win clearly, Google suggests "pausing your other experiment arms and shifting all the budget to the higher performing experiment arm". TikTok offers the same: "At the end of a split test, you can simply choose to continue running the winning campaign or ad group". Then you start the next test from the winner.

A simple plan for your next test

  1. Write the question. One sentence: what are you trying to find out?
  2. Pick one variable. Hook, presenter, opening shot, offer or length. Only one.
  3. Make the versions. Same product, same script outside the part you're testing.
  4. Use the platform's own testing tool, so the audience is split and each person sees one version.
  5. Wait for a conclusive result before you call it.
  6. Keep the winner, or the cheaper one if they tie, and test the next variable against it.

For ideas on what to test first, our article on one product and twenty hooks shows how many different openings a single product can carry.

Where Klovera fits

The hard part of testing is making several versions that differ in one thing only. Klovera starts from a product link or an idea: it writes the script, casts a realistic presenter and makes the video. Keep the script and change the presenter, or keep the presenter and change the hook, and you have clean versions to put into Google's or TikTok's own testing tools. The test itself still runs on the ad platform.

Frequently asked questions

Can I test video ads by posting them one after another?

Not fairly. The audience and the conditions change from week to week, so you can't tell what caused a difference. Google splits traffic evenly between versions running at the same time, and TikTok makes sure each audience group sees only one version.

How many versions can I test at once?

On Google, an experiment can have "up to 10 arms total", counting the original. Each version should still differ from the original in only one thing, so you know what made the difference.

How confident should I be before picking a winner?

Google's conclusive results are at a 95% confidence interval, and TikTok's split tests are designed around a 90% confidence rate. Wait for the platform to report a conclusive result rather than judging after a day or two.

Sources: Google, "Create a video experiment - Google Ads Help", support.google.com, no date on the page, read 18 November 2026. Google, "Monitor the performance of a video experiment - Google Ads Help", support.google.com, no date on the page, read 18 November 2026. Google, "Google Ads Help" (the article on the Experiments page, formerly drafts and experiments), support.google.com, no date on the page, read 18 November 2026. TikTok For Business, "About Split Testing | TikTok Ads Manager", ads.tiktok.com, last updated January 2026, read 18 November 2026. Platform tools change; check each help page before you set up a test.

Get started