User-generated-content ads, the casual person-to-camera spots that look like a real customer made them, have been the dominant paid-social format for years. The catch was always supply. You needed creators, products to seed, briefs, and a queue of revisions. In 2026, AI video closed most of that gap. You can now produce UGC-style video at a volume that used to be impossible, which means the bottleneck moved from production back to ideas.
This guide covers what actually works: the six formats worth making, the tools to make them, how to write a script the model can perform, the five angles to test first, and the one rule that decides whether any of it gets watched.
The six UGC formats that work
Not all UGC is the same. These six cover the formats that consistently earn spend.
1. Talking head testimonial
A person speaks to camera about the product, what problem it solved, and why they recommend it. This is the backbone format. It works because a face and a voice read as a real endorsement, and it adapts to almost any product.
2. Unboxing and first impression
The creator opens the package, reacts, and walks through what is inside. It works because it puts the viewer in the moment of receiving the product and shows the packaging and contents without a hard pitch.
3. Problem and solution demo
Open on the frustration, then show the product solving it on camera. It works because leading with the pain self-selects the right viewer, and seeing the fix beats hearing about it.
4. Before and after
Show the state before the product and the result after. It works for the same reason the static version does: a visible transformation is the most persuasive proof there is. Skincare, fitness, cleaning, and home products live here.
5. Listicle or "reasons I love it"
The creator counts through three to five reasons in quick cuts. It works because the structure keeps pace high and lets you pack several selling points into one watch.
6. Street interview or vox pop style
A casual, on-the-street feel where someone reacts to or is asked about the product. It works because the format reads as unscripted and current, which lowers the guard.
The tools that make AI UGC
The space moves fast, but a few tools define the categories in 2026.
For talking-avatar UGC, Arcads is the name most operators start with. You write a script, pick an AI actor, and it generates the person-to-camera performance with voice and lip sync. It is built specifically for the testimonial and listicle formats above, which makes it the most direct path to volume UGC.
For raw AI video generation, the model layer matters. Seedance, Kling, and Veo are the generation engines doing the heavy lifting for b-roll, product motion, and scene generation. Kling and Veo have pushed realism and motion coherence to the point where short product clips hold up in the feed. Seedance competes on speed and cost per clip. You will often combine these with an avatar tool, using the avatar for the talking and a generation model for the supporting shots.
The practical stack is usually a talking-avatar tool for the performance, a generation model for b-roll, and an editor to cut them together with captions and a hook card. Captions are not optional. Most of the feed watches on mute.
How to write a script the model can perform
A weak script makes even a good avatar look fake. A few rules keep it natural.
Write the way people talk, not the way brands write. Short sentences. Contractions. One idea per line. If you would not say it out loud to a friend, cut it. Models perform conversational copy far better than polished marketing copy, and conversational copy converts better anyway.
Open with the hook in the first line, not the third. The model will deliver whatever you give it, so the first sentence has to earn the next five seconds. We will come back to this.
Keep it to one core message. A 20 to 30 second UGC spot has room for one main idea and maybe one supporting point. Trying to say everything says nothing. Pick the single most compelling reason and build the whole script around it.
End with a soft, specific call to action. "Go check it out" is weak. "I linked it, the travel size is what I started with" is specific and feels like a real recommendation.
If you have ever written a static ad brief, this will feel familiar. The discipline of naming the real product, the real price, and the real proof, and never inventing a claim, is the same discipline that makes static briefs work. It is the backbone of how we write briefs for static concepts too, which you can see in the brief preview tool.
The five angles to test first
Do not test six formats times ten scripts on day one. Start with five distinct angles, because angle matters more than format.
First, the discovery angle: "I finally found a product that does X." Second, the skeptic-converted angle: "I did not think this would work, then it did." Third, the comparison angle: "I tried the cheap version and the expensive version, here is the difference." Fourth, the problem-first angle: "If you struggle with X, watch this." Fifth, the social-proof angle: "Everyone kept talking about this, so I tried it."
Each angle is a different reason to care, aimed at a different buyer. Run one script per angle, see which angles pull, then make more variations of the winners. This is the same logic that makes a concept library beat a single static ad. Breadth of angle finds the pocket that converts.
The first three seconds rule
This is the rule that overrides everything else. If the first three seconds do not stop the scroll, nothing after them matters. Most ad spend is wasted in the opening frames, because that is where viewers leave.
Three things make the first three seconds work. Motion or a visual pattern interrupt, so the frame does not look like every other frame in the feed. A spoken or on-screen hook that names a problem or makes a claim the right viewer cannot ignore. And the product or the payoff teased early, so the viewer knows there is a reason to stay.
Practical version: open on the creator already mid-sentence with the hook, not on a logo or a slow pan. Put a bold caption card on the first frame. Show the product within the first two seconds. Front-load everything, because there is no back half if nobody watches it.
Test your hooks independently of your scripts. The same body with five different opening lines is five different ads, and the opening line is usually what decides the winner. You can browse proven hook structures and see them filled in for your category with the free hook generator.
How static fits with video
Video gets the attention, but static still does a lot of the work, especially for cold prospecting, retargeting, and any placement where a still image loads faster and reads cleaner. The smart play is to run both and let them cover different jobs.
The screenshot static formats are a bridge between the two worlds. An iMessage thread, a Reddit post, or a TikTok comment section rendered as a still carries the same native, made-by-a-real-person feel as UGC video, at a fraction of the production cost. They are some of the best cold-traffic creative you can run, and they pair naturally with a UGC video strategy.
That is where Paintgym fits. It is static-first, built to generate a wall of ad concepts, including the native screenshot formats, from one product URL, with briefs written by Claude and a review step on every render. Video features live alongside the static engine, so you can build your testing layer in one place. Walk through the whole flow to see how it comes together.
Putting it together
A realistic AI UGC program in 2026 looks like this. Pick two or three of the six formats that fit your product. Write five scripts, one per angle, each with a different first-line hook. Generate them with an avatar tool plus a generation model for b-roll, add captions, and make sure every spot front-loads its hook into the first three seconds. Launch, read the data after a few days, double down on the angles that pull, and refresh before fatigue.
Then back the whole thing with static. Run your hero concepts, your proof concepts, and the native screenshot formats so cold and warm traffic both have creative built for them.
The production constraint that defined UGC for a decade is mostly gone. What is left is the part that always mattered: knowing your buyer, picking the right angle, and earning the first three seconds. Get those right and the tools will keep up with you.
Ready to build the static half of your testing layer. Start free with Paintgym, or read the 49-concept guide to plan your first batch.