On this page
- The steps
- Real skincare video ads
- Why two videos in one ad set is not an A/B test
- Pick one variable, and make it the opening
- What long-running skincare openers show you
- Build the test in Ads Manager
- Pick the winner metric first, then read results
- Iterate, label and scale what you learn
- Prompts to copy
- Example A/B test plan for an AI-generated opener
- Mistakes to avoid
- Questions
Real skincare video ads to model yours on
6,616 video ads from 759 advertisers in the Meta Ad Library, read October 5, 2026. The median one ran 18 days, and an ad still paid for after months is the closest public sign that it works. Live ads first.
- Ran 1 day
- Ran 6 days
- Ran 1 day
- Ran 8 days
- Ran 1 day
- Ran 8 days
- Ran 2 days
+6,604 more skincare video ads
Hooks, landing pages and run length for each
- Ran 10 days
- Ran 3 days
- Ran 5 days
- Ran 3 days
Every creative, with its copy, format and run length
The openers that ran longest
The Best in Men's Skincare & Grooming
Brickell · running 1,146 daysUltra-Potent Natural Skincare
Frøya Organics · running 643 days- Limited Time Only- Take 50% OFF LifeCell + Free Skin Mask
LifeCell · running 632 days𝐒𝐭𝐫𝐮𝐠𝐠𝐥𝐢𝐧𝐠 𝐰𝐢𝐭𝐡 𝐒𝐤𝐢𝐧 𝐨𝐫 𝐇𝐚𝐢𝐫 𝐢𝐬𝐬𝐮𝐞𝐬?
Deyga Organics · ran 592 days
Why two videos in one ad set is not an A/B test
The most common shortcut is dropping two videos into the same ad set and checking which one has the lower cost per result. It feels like a test. It is not one. Meta's delivery system decides how much spend each ad gets, and it often favors one early. The ad that gets more delivery is not necessarily better, it just got the chance to prove itself.
A real A/B test splits your audience into separate groups so each version is shown to a different set of people. That is what the A/B Test tool in Ads Manager and the Experiments area do. Each group sees only one version, with budget split evenly, so the comparison is fair.
The in-ad-set approach still has a place. It is good for optimization, letting Meta push spend toward whatever it predicts will perform. Use it when you want to scale. Use the A/B Test tool when you want to learn something you can reuse.
Pick one variable, and make it the opening
AI video makes it cheap to produce many different ads, which tempts you to change everything at once: new script, new voice, new visuals, new pacing. If B wins, you learn only that B is better as a whole package. You cannot say why, and you cannot repeat it.
The cleanest first test is the opening. Keep the same 15-second product demo in both versions. Swap only the first three seconds: a filmed opener for A, an AI-generated opener showing the same product benefit for B. The ad body, voiceover, captions and end card stay identical.
Be honest about what the result means. If B wins, you have learned which opening works better for this ad, with this audience. You have not proven that AI video beats filmed video in general. Write the test question in one sentence and put it in your ad naming so the finding survives a team change.
- Good variables: opening shot, first on-screen line, product reveal style, hook length
- Bad variables: anything bundled with the opener, like a new offer or a new CTA
- One test, one change, one metric
Want skincare video ads like these for your product?
The skincare video ads that kept running, read from the ad libraries. Every change waits for your approval.
What long-running skincare openers show you
The Sprites ads library shows 6,616 skincare video ads on Meta from 741 advertisers, with a median run of 18 days. The ads that survive far past that tend to open with one plain, concrete promise. Brickell opened with "The Best in Men's Skincare & Grooming" and ran 1,146 days with a Shop now button. Glov Beauty opened with "See the Difference After just 1 Treatment" and ran 378 days.
Two more show the same shape. Based Supplies ran "Restore Your Skin in 3 Weeks!" for 365 days. DermExcel by Dermal Health Science ran "Stop Guessing Your Skincare" for 236 days. Each line names a result, a time frame or a clear audience in the first breath. None of them tease or build slowly.
That tells you what to test. Your A and B openers should make the same promise visually, but in different ways: a product reveal against a texture close-up, or a claim on screen against a claim in the voiceover. Long-running ads are likely survivors of repeated testing, so borrow their clarity, not their exact wording. For more category patterns, see skincare video ads.
Build the test in Ads Manager
Start from your control ad. Duplicate it, then swap in the new video for the challenger. Keep the audience settings, placements, optimization event, offer, ad copy, CTA and landing page exactly the same. If automatic creative enhancements could change the element you are testing, turn them off for both versions so Meta does not alter your opening behind your back.
Upload your AI-generated video as a finished file in the right format. Vertical 9:16 is the standard for Reels and Stories placements, and the Meta ad specs cover the details. Match the two videos on length, audio loudness, caption style and end card, so the only visible difference is the opening.
Then launch through A/B Test or Experiments. Select the two versions, split the budget evenly, and run them at the same time. Set the budget and end date before you launch. Give the test at least a full week so weekday and weekend behavior both show up, and do not edit either version once it is live.
- Same audience, placements and optimization event
- Same copy, CTA, offer and landing page
- Even budget split, simultaneous start
- Budget and end date fixed before launch
Pick the winner metric first, then read results
Decide your success metric before launch, because it is too easy to pick the flattering number afterward. For sales, use cost per purchase. For lead generation, use cost per qualified lead if you can track lead quality, or cost per lead if you cannot. This is the number that decides the test.
Video views, hook rate and click-through rate are diagnostic. Use them to understand why an ad performed differently. If B has a higher click-through rate but a worse cost per purchase, the opening attracted clicks without buying intent. That is a finding, not a win.
If purchases are sparse or the gap is small and uncertain, call the test inconclusive. Do not crown whichever ad is slightly ahead. An inconclusive result is still useful: it says this opener change does not move the needle, so test something bolder next time, such as a different promise rather than a different visual style.
Iterate, label and scale what you learn
Treat the winner as your new control and test the next single variable against it. A sensible order is opening shot, then opening line, then offer framing, then end card. Each round teaches you one thing. Keep a simple log: test question, versions, metric, result, and next test.
Check Meta's current rules on labeling and disclosing AI-generated content for ads before you publish, and follow them. Keep your branding and claims consistent across versions, so a policy review does not pause one version mid-test and break the comparison.
If producing variants is your bottleneck, a repeatable pipeline helps. Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human. Whatever tool you use, the rule holds: one change per test, finished files, same everything else. The video ads hub has more formats to test.
Prompts to copy
Paste each one as written, then change the product, the setting and the light to match your own footage.
Variant B hook: product reveal
Vertical 9:16 video, 3 seconds. A frosted glass serum bottle with a white dropper sits on a pale stone surface in soft morning window light. The camera pushes in slowly as a single droplet forms and falls from the dropper. Clean, minimal, premium skincare look, shallow depth of field, no text, no people, no brand logos, no packaging labels.
Variant B hook: texture macro
Vertical 9:16 macro video, 3 seconds. Extreme close-up of a creamy moisturizer being swirled with a flat spatula on a glossy white surface, soft highlights, visible smooth texture. Slow, satisfying motion, bright neutral lighting, shallow focus. No people, no faces, no text overlays, no brand logos, no readable labels.
Variant B hook: bathroom shelf routine
Vertical 9:16 video, 3 seconds. A tidy bathroom shelf with three unbranded skincare bottles in a row. The camera tracks left to right and stops on the middle bottle as warm light brightens around it. Calm, realistic, clean aesthetic, gentle motion. No people, no readable text, no brand logos.
Shared body: application close-up
Vertical 9:16 video, 5 seconds. Close-up of a smooth cream being pressed onto the back of a hand model-style mannequin hand, absorbing into a soft, even finish. Natural daylight, neutral background, steady camera with subtle drift. Realistic skin texture, no real people, no faces, no text, no brand logos.
Shared end card background
Vertical 9:16 video, 4 seconds. A calm, minimal tabletop scene with a single unbranded skincare jar centered on a light linen surface, a small green leaf beside it, soft diffused light. The camera is nearly still with a slight slow zoom. Leave the top and bottom thirds empty for captions and a button. No people, no text, no logos.
Example A/B test plan for an AI-generated opener
| Element | Version A (control) | Version B (challenger) | What the result tells you |
|---|---|---|---|
| Opening 3 seconds | Filmed product shot | AI-generated product reveal | Which opener works better for this ad |
| Ad body and voiceover | Same 15-second demo | Same 15-second demo | Nothing, it is held constant |
| Copy, CTA, landing page | Identical | Identical | Nothing, it is held constant |
| Audience and placements | Same settings | Same settings | Keeps the comparison fair |
| Winner metric | Cost per purchase | Cost per purchase | Decides the test |
| Diagnostic metrics | Hook rate, click-through rate | Hook rate, click-through rate | Explains why one version won |
Mistakes to avoid
- Putting both videos in one ad set and calling the higher-delivery ad the winner.
- Changing the script, voice and visuals together, so you cannot tell what caused the difference.
- Leaving automatic creative enhancements on when they could alter the element you are testing.
- Picking the winner metric after you see the results, or crowning a winner on click-through rate alone.
- Editing budgets, copy or creative while the test is running.
- Declaring a winner on a slight, uncertain lead instead of calling the test inconclusive.
Rather have the videos made for you?
Director by Sprites is a full creative team for your startup. Every Monday you get finished, on-brand video ads, reviewed by a human. You never write a prompt.
Apply to Director by Sprites →Questions
Can I A/B test AI video ads against my regular filmed ads on Meta?
Yes. Keep the filmed ad as version A and an AI version as B, and change as little else as possible. The cleanest version swaps only the first three seconds. A win tells you which opening works for that ad, not whether AI video beats filmed video in general.
What is the difference between an A/B test and putting two ads in one ad set?
An A/B test splits your audience into separate groups and gives each version an even share of budget. Two ads in one ad set compete for delivery, and Meta can favor one early. That is good for optimization but not a clean comparison.
What metric should decide the winner of a Meta video ad test?
Pick it before launch. Use cost per purchase for sales. For lead generation, use cost per qualified lead if you can track quality, or cost per lead if you cannot. Treat views and click-through rate as clues about why, not as the deciding number.
What should I change when testing an AI-generated video?
Change one thing, ideally the opening three seconds. Keep the body, voiceover, captions, copy, CTA and landing page identical. Single-variable tests produce findings you can reuse in the next ad.
What if the A/B test result is unclear?
Call it inconclusive. If purchases are sparse or the gap between versions is small and uncertain, do not pick the ad that is slightly ahead. Run a bolder test next, such as a different promise in the opener instead of a different visual style.
Should I turn off Meta's automatic creative enhancements during a test?
If an enhancement could change the element you are testing, turn it off for both versions. Otherwise Meta may alter the creative differently in each ad, and you will not know whether your change or the enhancement drove the result.