AI UGC
What Makes a UGC Video Ad Perform
Four things decide whether a UGC video ad works, and they are not equally weighted. The opening two seconds decide whether anything else is seen. The performance decides whether the claim is believed. The claim decides whether the click is worth anything. And the number of variations you can afford to test decides whether you ever find the combination that does all three.
Talk through your creative testing planWhat this rests on
600,000+ videos analysed
The production model is trained on more than 600,000 high performing ecommerce videos from TikTok, Instagram and YouTube Shorts.
Real human performance
Every video is built on a real filmed human performance rather than a synthetic character generated from a text prompt.
Built for paid social
Videos are delivered in the formats TikTok, Meta, YouTube Shorts and Instagram expect.
In order of weight
The four parts, and why the order is not arbitrary
The opening two seconds
Nothing else on this list matters until this works, because everything else is only seen by people the opening kept. Treat the hook as a separate artefact from the video: write ten, produce all ten against the same body, and let the retention curve tell you which one earned the rest of the ad.
The performance
The performance decides whether the claim is believed. This is the part that degrades most obviously when the person on screen is assembled rather than filmed, and it degrades in a way viewers register as distrust rather than as a technical flaw. They do not think the video looks synthetic. They think the claim sounds false.
The claim
One specific claim beats three vague ones. A specific claim can be disbelieved, which is precisely why it works: it is falsifiable, so it reads as something a person would actually say. Copy that could describe any product in the category persuades nobody and is the most common failure in otherwise well produced ads.
The number of variations behind it
Most creative fails. That is the normal state, not a sign of a broken process. The advantage belongs to whoever can afford enough attempts that a failure is unremarkable, because they kill losers quickly instead of defending them. Volume is not a substitute for judgement, it is what makes judgement affordable.
How to read the evidence
Be careful with every statistic, including ours
Performance figures in this category are almost always drawn from the accounts of whoever is publishing them, on their own product mix, at a moment of their choosing. That does not make them false. It does mean a lift measured on another advertiser account is a hypothesis about yours rather than a finding.
Our own view is shaped by the corpus the production model was trained on, which skews heavily toward ecommerce, short form and paid placement. It reflects what performs in that context. It says much less about long form, about organic reach, or about considered purchases with a long sales cycle.
The only figure that settles an argument about your advertising is one from your own account. Everything on this page is worth treating as a prior to test against, and worth abandoning quickly when your own data disagrees.
Practical
A testing rhythm that survives contact with a real week
Pick one variable per batch. If a batch changes the hook, the performer and the offer at once, a winner teaches you nothing transferable and you will rebuild the same guess next month.
Decide the kill criteria before the batch runs, not while watching it. Creative gets defended for reasons that have nothing to do with performance, and a threshold agreed in advance is the cheapest protection against that.
Keep the losers. A record of what failed and why is more valuable after six months than a folder of winners, because the winners are already running and the failures are what stop the same idea being tried a third time.
Questions
Common questions
How many variations should I test?
Enough that a losing creative is unremarkable rather than expensive. Brands testing a handful per month tend to protect their losers; brands testing a large batch tend to kill them quickly, which is the actual advantage.
Does the same video work on every platform?
The performance travels. The framing, the caption safe area and the pacing do not, which is why the same take is delivered cut for each destination rather than uploaded once.
Is polished production a disadvantage?
In feed, usually yes. The format the viewer is scrolling through is informal, and an ad that announces itself as an ad in the first frame is skipped before the claim is ever heard.
Keep reading
Related reading
What Is AI UGC?
Understand what AI UGC is and how it differs from creator filmed UGC.
How to Choose an AI UGC Platform
Decide which AI UGC provider fits, using criteria that can be checked before buying.
The Complete Guide to UGC Video Formats for Paid Social
Talking head UGC ads deliver 31% higher CTR than product-only videos, while testimonial formats drive 2.4x higher conversion rates on retargeting campaigns. This guide breaks down every major UGC video format, with platform-specific performance data for TikTok, Meta, and YouTube Shorts.
ROAS Optimization: How Creative Volume Impacts Ad Performance
Campaigns with 10 to 15+ creative variations significantly outperform those with fewer. Advertisers using Advantage+ creative features see 22% higher ROAS, and the brands testing 50+ variations monthly are finding winners faster while their competitors burn budget on fatigued ads.
Want this for your product?
Tell us what you are advertising and what you have tried. We will come back within one business day.
Talk through your creative testing plan