Video ad variations for testing: a 5-step playbook

To make video ad variations for testing, keep one body and change one part at a time. Write three hooks and two endings for the same middle: that is six videos from one script. Test the three hooks first, in one ad group. Then test the two endings on the hook that leads. Keep the videos you did not launch for the day the leader wears out.

Why one video ad is not enough

A single ad gives you one result and nothing to compare it with. You learn whether it sold, but not why. A set of variations that differ in one part tells you which part did the work, and that lesson carries into the next script.

The second reason is fatigue: people get tired of seeing the same ad. Meta's analytics team studied repeated exposures to the same creative. A summary of that study quotes its finding that at four repeated exposures, the associated likelihood of a conversion drops by about 45%. The same summary reports that the study's guidance improved conversion rate by about 8% on average in high-fatigue cases, across roughly 26 thousand cases. Note the word "associated": the 45% is a pattern in the data, not a measured cause. The 8% comes from a split test.

Either way, the point for a small team is the same. You need the next version ready before the first one wears out, not after. Variations made in one batch give you that.

This playbook is for marketers and founders who make their own ads and have a modest budget. It assumes one product, one offer and one audience. If you change all three at once, no test can tell you which change mattered.

How many variations the platforms expect

The platforms give a range, and it is smaller than many people think.

• TikTok's creative best practices suggest 3 to 5 different creatives per ad group, and 3 to 5 diversified ad groups per campaign.
• Meta's A/B testing page says you can test up to five ad variants. It defines a test variable as one change within a campaign, ad set or ad.
• Google's help page on video experiments describes a two-armed A/B test of video assets for YouTube, and a custom experiment with up to 10 arms.
• An ad tool maker's guide on creative counts says that for a first batch, three to five distinct ideas may be manageable. It also warns that loading 20 or more ads into one ad set is usually counterproductive.

So plan for three to five videos in each test. More ads in one group split the budget thinner, and each ad gets less data. Fewer than three gives you a winner but not a pattern.

The 5-step playbook for video ad variations

The five steps below go from one script to a test that reads clearly, and then to the refresh. Each step changes one thing. The worked example after step 3 shows the arithmetic.

Step 1: Write one body and split it into parts

Start from one script and cut it into three parts:

1. The hook: the first seconds, before the viewer decides to stay.
2. The body: the product, the proof and the offer.
3. The ending: the call to action on the last screen.

The body stays the same in every version. It is the longest part to make, and it is the part you least need to test first. The hook and the ending are short, so new versions of them cost little.

TikTok's guidance shows why the hook comes first. The same best-practices page says to prioritize your hook in the first 6 seconds and to introduce your content proposition in the first 3 seconds. If the first seconds lose the viewer, the body never plays. Our post on openings that stop the scroll takes that first second apart frame by frame.

Step 2: Write three hooks that differ in idea

Three hooks that say the same thing in other words teach you nothing. Give each hook a different idea:

• Hook A, the problem: name the pain the product removes.
• Hook B, the result: show the end state first, then how to get there.
• Hook C, a question: ask the viewer something they answer in their head.

Put each hook's words on screen as well as in the voice, because many people watch with the sound off. TikTok's page recommends showing 5 to 10 words per second when you use text, so keep each hook line short. A hook of eight words fits in about one second at that pace.

Keep everything else in the hook the same: the length, the music, the first frame's layout. If hook B is also two seconds longer, a win for B can come from the length, not the idea.

Step 3: Write two endings

The ending asks for one action. Two endings that ask for different actions make a fair test: for example, "Try it" against "See the price". Keep the same last-screen layout, so that only the words change.

Test the ending after the hook, not with it. An ending only works on people who stayed to the end, and the hook decides how many people stay. In the Ritual Coffee launch ad, one note in the chat ("Button: order now.") changed the words on the last button, and nothing else in the film moved. That is the size of change an ending test needs.

A worked example: one script, six videos

Here is the arithmetic for one script.

• Hooks: 3 (A, B, C)
• Bodies: 1
• Endings: 2 (1, 2)
• Videos: 3 × 1 × 2 = 6

Each video then needs a size for each placement. With four sizes (9:16, 4:5, 1:1 and 16:9), the count is 6 × 4 = 24 files. That number is why variations feel slow when you edit them by hand. The script work is the same for one video or for six; the export work grows with every hook, every ending and every size.

You do not launch all six at once. You test in two rounds.

Round 1: hooks

Launch A1, B1 and C1 in one ad group. The body and the ending are the same, so a difference in results comes from the hook. Three ads sit inside TikTok's 3 to 5 range and inside Meta's limit of five variants.

Round 2: endings

Say hook A leads. Launch A1 against A2. Only the ending differs. A1 is already made and already running, so this round needs one new video, not two.

Count what you used: 3 videos in round 1, plus 1 new video in round 2, makes 4 of the 6. B2 and C2 stay on the shelf. When the leader's results start to fall, you already have videos with tested hooks and a second ending, ready to go.

Step 4: Set up the test so it reads clearly

• Change one thing in each round. Meta defines a test variable as one change, and that is the reason the two rounds above are separate.
• Keep the ads in one group. If the hooks run in different ad groups with different audiences, the audience changes as well as the hook.
• Use the platform's own test tool when you can. Meta's A/B testing, Google's video experiments and TikTok's ad groups each keep the comparison inside one system. Google's page says directional results show from 70% confidence and recommends waiting until the experiment finishes for conclusive results.
• Size every variation the same way. A hook that loses because its text sat under the app's buttons did not lose on its idea. Check the Reels safe zone before you export the 9:16 files.
• Name the files by their parts. "A1-9x16" tells you the hook, the ending and the size. You read the results in a table, and the names make that table readable.
• Write the rule before you launch. Decide which number picks the winner (cost per result, click-through rate, or watch time) and how long the round runs. The guide on creative counts says to set the observation period and the decision plan before launch, and not to declare a winner from the first reporting spike.

Step 5: Refresh before the ad wears out

TikTok's best-practices page says to refresh your ad group creatives when delivery results show a consistently declining trend, or when daily new users are low. The Meta study above shows why waiting costs money: each repeat view of the same creative goes with a lower chance of a conversion.

The fix is a shelf of videos you made but did not launch. In the example, that shelf holds B2 and C2 from the first script. For the next script, keep the hook idea that won and write a new body around it. Each new script starts from what the last test taught you, so each test is smaller and faster than the one before.

Common mistakes and their fixes

• Every version changes everything. New hook, new music, new offer and new ending in each video. The test picks a winner, but you cannot say why. Fix: one variable for each round.
• Hooks that differ only in words. "Save time on X" and "Get time back on X" are one idea. Fix: give each hook a different idea, as in step 2.
• Too many ads in one group. Twenty versions split a small budget into twenty thin slices, and none gets enough delivery. Fix: three to five for each test.
• A winner called on day one. An early spike is noise as often as a signal. Fix: write the rule and the length of the round before you launch, and keep to them.
• Different formats inside one test. A 9:16 video against a 1:1 video tests the format, not the hook. Fix: compare like with like, size by size.
• No shelf. The winner wears out and the next video is not made yet. Fix: make all six in the first batch, even if you launch only three.

How to make the variations faster

The slow part is not the writing. Three hook lines and two endings fit on one page. The slow part is turning six versions into 24 files, each with its own voice line, on-screen text and size.

This is the part FoxyLab does from a chat, for ads and for any other marketing film: a launch film, an explainer or a social clip. You paste a link to your product or write one line about it. The studio writes the film, voices it, animates it and renders it in 9:16, 4:5, 1:1 and 16:9. Then you write a note such as "make two more hooks: one that opens on the price, one that opens on a question", and the studio changes the film from that note. You keep the body and get the new versions. Our post on how to give feedback on a video shows how to write notes like that one.

Checklist before you launch

1. One body, the same in every version.
2. Three hooks, each a different idea, each the same length.
3. Hook words on screen as well as in the voice.
4. Two endings with the same layout and different words.
5. Six videos made, three launched in round 1.
6. Every file named by hook, ending and size, for example A1-9x16.
7. Text inside the safe zone for each size.
8. All round 1 ads in one ad group, or in one platform test.
9. The winning number and the length of the round written down before launch.
10. B2 and C2 on the shelf for the refresh.

Questions and answers

How many video ad variations should I test at once?

Three to five. TikTok suggests 3 to 5 creatives per ad group, and Meta's A/B testing takes up to five variants. More than that splits a small budget into slices too thin to read.

What should I test first in a video ad?

The hook. It decides whether anyone sees the rest. TikTok's guidance puts the hook in the first 6 seconds and the content proposition in the first 3. Test the ending after you know which hook leads.

Do I need a new video for each variation?

No. Keep one body and change only the hook or the ending. Three hooks and two endings on one body give six videos, and the body is made once.

When should I replace a winning ad?

When its results show a steady decline. TikTok's page names a consistently declining trend in delivery results as the time to refresh. Have the next variations made before that day, so the switch takes minutes.

Pick one product and write its one body today. Then write three hooks and two endings, and make the six videos in the FoxyLab studio, so that round 1 can go live this week.