Cookbook › TikTok UGC ad
Video·Practical·25–40 min·ChatGPT · Veo · ElevenLabs
How to create a TikTok UGC ad with AI
UGC (user-generated content) is the format that converts hardest on TikTok because it feels like a friend's recommendation, not an ad. This recipe shows you how to make one from scratch — no creator hired, no face on camera.
What this recipe solves
You want to run UGC creatives but: (a) you don't have creators in your network, (b) you don't want to be on camera, (c) the budget won't cover an agency. The typical solo result is an ad that looks like an ad — robotic voice, locked-off shot, stock background. It doesn't convert.
When to use: validating an offer, testing 5 angles in a single day, or replacing a tired creative.
Tools you'll need
- ChatGPT (GPT-4o or higher) — for hook and script
- Veo 3 or Runway Gen-3 — B-roll generation
- ElevenLabs (Conversational v2 voice) — narration
- CapCut or DaVinci Resolve — edit and captions
Step-by-step
Define the hook (3 to 5 seconds)
The hook isn't "Look at this cool product." It's a problem. Something that makes the viewer stop scrolling because they feel this is about them.
- "Bought this 3 weeks ago and I can't go back"
- "Thought it was marketing — turns out it actually works"
- "Nobody warns you about this before you buy X"
Write the script (15 to 30 seconds)
Structure: hook → tension (1 sentence on the problem) → reveal (the product) → proof (concrete result) → soft CTA ("link in bio," no shouting).
Paste this prompt into ChatGPT:
Generate visuals (B-roll with Veo)
Don't try to generate a person speaking to camera — it looks plastic. Generate B-roll instead: the product in use, hands interacting, details. Voiceover on top creates the UGC feel.
For each beat of the script (about 4-5 clips), write a Veo prompt like:
Generate the voice (ElevenLabs)
Pick a Conversational voice — not Narration. Stability 35-45%, Style 30%. Add commas and "..." to sound human.
Want it even more natural? Record yourself doing "mm," "like," "you know," and mix into the final cut.
Assemble the final cut
In CapCut: 9:16 timeline, voice on track 1, B-roll on track 2, ambient music at -22dB on track 3. Cut on the beat — each clip 2-4 seconds. Add big yellow captions (auto-captions, then proof).
Export 1080×1920, 30fps, MP4.
Full prompt (paste into ChatGPT)
Prompt variations
- Instead of generic UGC: "customer testimonial style" (more serious)
- For Gen-Z: "ironic tone, with self-deprecation"
- For B2B: "tone of a peer giving a tip"
Example output
Common mistakes to avoid
- Trying to generate a talking person. Doesn't work in 2026. Use B-roll + voiceover.
- Music at 100%. If music drowns the voice, the ad is dead. -22dB in the background, always.
- Narrator-tone voice. "Documentary" style kills UGC. Use Conversational or record yourself.
- Hook that sells. A hook that sells doesn't convert. A hook that names a problem does.
- Forgetting captions. 80% of viewers watch muted. No captions, no audience.
Related recipes
Jump to Module 2 of the course: Anatomy of a prompt that works.
Open Module 2 →