The confession
Open by admitting the thing the buyer already does and would rather not say out loud. It earns attention because it sounds like a person instead of a claim.
What actually goes into a creator style video ad, why the first two seconds decide the rest, and how to brief five versions of it in one instruction.
product: magnesium sleep tea, 30 sachets
audience: women 28-45 who wake at 3am
format: testimonial to camera
creator: female, mid thirties, kitchen, morning light
hook: "I stopped blaming my mattress."
body: two lines on the 3am waking, one on the tea
close: "Link in the bio, first box ships today"
aspect: 9:16
length: 20s
variants: 5 hooks, same body, same closeA whole ad brief is text, so it can live in a file, a sheet row or a line of chat.
A creator style video ad where the person on camera was generated instead of booked.
UGC ads are the video ads that look like a customer made them. Handheld, vertical, one person talking straight to the lens, no studio light and no logo sting on the front. They work because the shape of the thing signals a person rather than a brand, and people scroll past brands. Everything else about them is ordinary paid social: a hook, a claim, a close, a landing page.
An AI UGC ad keeps that shape and drops the shoot. The face is generated, the voice is generated, and the script is whatever you wrote. What comes back is a vertical file you upload to an ad account like any other asset. If you want to see how close the footage gets today, and where it still gives itself away, that is written up on AI UGC videos.
The trade is speed against control. A real creator brings timing, judgement and a face some people already trust. A generated one brings a turnaround measured in minutes and no booking calendar. When you need eight angles before you know which one lands, the second trade is usually right. For a hero brand film it is not, and nothing on this site will tell you otherwise.
The part people underestimate is that the render is not the work. The writing is. A good AI UGC ad and a bad one come out of the same model on the same day. The difference is the first line.
Almost every one of these that works has the same three parts in the same order, and the first part does most of the job.
One line that names the problem, the result or the mistake. It has to land with the sound off, so the claim goes in the on screen text as well as in the mouth. If the hook is weak, nothing behind it gets watched.
Two or three specific sentences. What it is, what changed, and the question a sceptic asks next. One idea per sentence, so the edit has somewhere to cut.
One instruction, spoken and shown. Ask for the click rather than for consideration. A soft close on cold traffic reads like an ad that did not want anything.
0:00 hook, straight to camera
"I stopped blaming my mattress."
0:02 cut to the mug on the counter
"I was waking up at three every night for
about a year."
0:07 back to camera, holding the sachet
"One of these an hour before bed. That is
the whole change I made."
0:14 close, straight to camera
"Two weeks in I sleep through it.
Link is in the bio."
0:20 endFour beats, twenty seconds. The hook line doubles as the on screen text.
Six openings that keep showing up in ads that survive past the first week of spend. Pick one per variant and hold the rest of the script still.
Open by admitting the thing the buyer already does and would rather not say out loud. It earns attention because it sounds like a person instead of a claim.
Name what most people get wrong about the category, then correct it. Useful when the product needs a reason to exist before it needs a feature list.
Lead with the outcome and make the viewer wait for the method. Strongest when the result is something the camera can actually show.
Say who the ad is for in the first line. It sheds the wrong audience early and lifts the hold rate on the right one.
Start with the reason someone would not buy, then answer it. This one earns its keep on retargeting, where the doubt is already in the room.
Hold what they use now against what you sell. Keep it about the job being done, not about a rival brand by name.
One ad is a guess. A set is a test, and a test is what the ad account is actually asking you for.
That is a lot of clicking if every variant means choosing an avatar and pasting a script into a form. It is one instruction if the generator can be called by something that can count.
Making one of these in a browser tab is fine. The cost only shows up on the twentieth, when the same form has already been filled nineteen times.
Wireflow is our own tool and this is the part we built it for. Its hosted MCP endpoint exposes the generation workflows as tools, so any MCP client can list them, run them and poll for the file. The same wedge, applied to the whole production line rather than one batch, is UGC automation.
Take the brief in briefs/sleep-tea.md and give me five hook variants. Same body, same close, 9:16.
Read the brief. Five hooks: the mattress line, the three in the morning line, the melatonin objection, a result first open and a callout to shift workers. Running them on the same creator so the only variable is the first sentence.
Five files, named by hook. The melatonin one came back nine seconds longer because that script carries an extra sentence. Want it trimmed to match the others before you upload?
list_workflows, run_workflow and get_execution are the tool names the server really exposes. Only run_workflow spends credits.
The three questions that come next once the format makes sense.
It is a creator style video ad where the person on camera was generated rather than filmed. The format is the same as the handheld vertical ads that fill TikTok and Reels: one person, one take, a spoken claim and a close. The difference is only in how it was produced, and the file that comes out is uploaded to an ad account like any other asset.
It depends on the offer and the placement, and anyone who gives you a single number is guessing. What is reliably true is that the writing carries more weight than the render. A generated creator delivering a sharp hook beats a real one delivering a vague script, and the fastest way to find the sharp hook is to test more of them than a shoot budget allows.
Most direct response creative lands between fifteen and thirty seconds. TikTok in-feed ads accept five to sixty seconds and Instagram Reels ads go up to ninety, but the useful constraint is not the platform limit. It is that one idea, said once, with a hook and a close, rarely needs more than twenty five seconds.
Five hooks against a fixed body and close is a sensible first batch. That gives the ad account enough to find a direction without splitting the budget across a dozen half tested ideas. Once one hook is clearly ahead, hold it and vary the creator next.
Assume yes. The major ad platforms have AI content labels and expect synthetic people to be marked, and advertising rules in several markets treat an undisclosed synthetic testimonial as a misleading claim. Check the policy for the platform you are buying on before the ad goes live, not after.
Yes. The generation workflows are exposed over MCP, so an agent such as Claude, Cursor or an n8n flow can list them, run them and poll for the finished file. You describe the batch in the chat you are already in, and the video URLs come back in the same reply.
Connect the MCP endpoint once, paste the brief, and ask for five hooks. Listing the workflows costs nothing, so you can see the inputs before you spend a credit.