How to write a UGC ad script

Watch six UGC advertisements in a row and what stands out is how similar the structure is and how different the first two seconds are. The structure is nearly a formula. The opening line is where the performance lives.

What a UGC ad is and is not

A UGC advertisement is a person talking to their own phone, in a room, with the product in shot. It is not a testimonial video, which is a filmed interview with a real customer, and it is not influencer content, which runs on somebody else’s account and borrows their audience.

That distinction sets the whole brief. Influencer content opens to people who already follow that person. A UGC advertisement opens to a stranger who did not ask for it, so nothing can be assumed and nothing can be a slow build.

It also means the visual quality should be worse than your other work: a handheld frame, a real room, uneven light. Polish it and it reads as an advertisement pretending to be a person.

The four beats

A UGC advertisement is built from four kinds of scene and no others. The table at the foot of this post gives each one its clock and its word budget.

The talking scenes are the spine, and there are only ever two or three of them: the hook, the experience, and the recommendation. Each runs five to seven seconds of speech, which at a normal pace is thirteen to eighteen words. The whole video is capped under thirty seconds, so three talking scenes plus three cutaways is already the ceiling.

The proof and credibility scenes are the cutaways between them, held to two and a half to three seconds. They are short because they are evidence, not argument. A three second shot of the thing working says more than fifteen words about it.

The close is always the last scene, always one line, and always contains the name. If a viewer remembers one thing, it should be the name attached to the one reason.

A cut straight from one talking head to another reads as a slideshow, which is why the format wants motion between them.

Writing the hook: ten patterns

The first two seconds are most of the result. Here are ten patterns that work, each filled in for a fictional product, an overnight face serum from an invented brand called Tidewell.

1. The flat problem. “I have had the same three spots on my chin since March.”
2. The unexpected result. “I did not expect a nine pound serum to fix the thing my dermatologist shrugged at.”
3. The correction. “Everyone told me to use more retinol. That was the problem.”
4. The count. “Night eleven. Here is the difference.”
5. The confession. “I bought this because of an advert, which I never do.”
6. The comparison. “This is the forty pound one. This is the nine pound one.”
7. The direct address. “If your skin stings every time you put anything on it, this bit is for you.”
8. The negative claim. “This will not clear acne. Let me tell you what it does do.”
9. The demonstration. “Watch what happens when I do this.”
10. The specific stakes. “I have a wedding in nine days and I panicked.”

Three things in common. None names the brand first. None is a question. And each could only be said by a person in a specific situation, which is the trick.

Write six, from at least three patterns, and keep everything after the hook identical while you test. That is the only way to learn anything about the hook.

The proof beat: show, do not claim

The proof scene fails when it is a second opinion. “It really works” is not proof, it is the same claim in a different voice.

Real proof is something the viewer can assess: the texture on a hand, a before and after in the same light, a screen with a number on it, the product on a shelf in a shop they recognise. Two and a half seconds is enough for exactly one of those, so pick one.

Credibility is the neighbouring beat and a different job: a rating, a customer count, a stockist, a guarantee. It answers “why believe this person”, where proof answers “does the thing work”.

The order that reads best is hook, proof, experience, credibility, close. The proof arrives while the viewer is still deciding whether to keep watching, and the credibility lands just before the ask.

Disclosure and claims

Three sets of rules, and none of them cares whether the person on screen is real.

16 CFR 255.5 requires that a connection between endorser and seller which the audience would not expect, and which might affect the weight they give the endorsement, is disclosed clearly and conspicuously. Section 255.0(f) says what that means for a video: where the claim is visual the disclosure belongs in the visual part, where it is audible it belongs in the audible part, and where it is both it belongs in both. The FTC’s influencer guidance is blunter. Put it in the video, not in the description.

16 CFR 255.2 is the one people trip over. An endorsement is read as representing what a customer will generally get, so an unusual before and after has to carry the generally expected result alongside it, and the FTC’s own testing found a “results not typical” line insufficient on its own.

16 CFR Part 465, in force since October 2024, bans writing, selling or disseminating a testimonial that misrepresents that the person giving it exists, that they used the product, or what their experience was. The FTC said on announcing it that AI generated reviews are covered.

In Australia the ACCC is equally direct about fake and incentivised reviews, and adds one line aimed straight at this format: scripts should never require a creator to misrepresent their experience or views on a product.

The implication is clear enough. A synthetic presenter explaining a product is fine. A synthetic presenter describing a customer experience nobody had is a fabricated endorsement, and no line at the bottom of the frame repairs it.

Length and pace

Under thirty seconds total, which is a hard cap in this format rather than advice. Cutaways at two and a half to three seconds. Talking scenes as long as their line and no longer.

TikTok’s own creative guidance asks for the content proposition inside the first three seconds and the hook prioritised within six. That is tighter than the beat structure implies, and it is the right instinct: the first talking scene has to carry the whole proposition, not set it up. Length across placements is covered in more detail in how long a promo video should be.

Briefing a creator or generating one

If you are briefing a person, give them the beats, the word count, the hook, and the one claim they may not exceed. Do not give a word-for-word script, because the format works by sounding unrehearsed.

If you are generating one, you write one description of the persona instead: gender, age, look, clothing, setting. One portrait is made from it and one voice speaks every line, so the person cannot drift between scenes, and three talking clips is the maximum. There is more on making a video with no camera in how to make a video without filming.

The example script

Tidewell Overnight Serum. Twenty seven seconds, five scenes, entirely fictional.

Hook, talking, 6s, 16 words. “I have had the same three spots on my chin since March. Night eleven of this.”

Proof, 3s, 7 words. Close on the same patch of skin, the same lamp, day one beside tonight. “Same light, same corner of the bathroom.”

Experience, talking, 7s, 18 words. “It does not sting, which is the whole reason I stuck with it. Two drops, and I sleep.”

Credibility, 3s, 8 words. A shelf shot. “Nine pounds. Sold in about four hundred pharmacies.”

Close, talking, 5s, 12 words. “If your skin hates everything, try Tidewell. That is genuinely all I have.”

Notice what is not in it. No adjective doing work a picture could do. No brand name until the last line. And no claim that the serum cleared anything, which is also the only version the rules would let you run.