This page has affiliate links. Here is what that means.
1. Write for the ear
Short sentences. One idea each. Read the script aloud before generating anything: if you run out of breath, so will the listener. Put the claim that stops the scroll in the first line.
2. Choose one voice and keep it
Pick a voice that matches who the ad is speaking to, then use it across the campaign so variants differ only in what you are testing.
3. Generate in sections
Generate the hook, the body and the call to action as separate clips. When one line sounds wrong you regenerate that clip, not the whole read, and you use fewer credits.
A text-to-speech tool such as ElevenLabs meters usage in credits, so a short regeneration costs less than a full one.
4. Mix it under captions
Most feed video is watched muted at first. Burn in captions, keep music well below the voice, and make sure the visual hook works with the sound off.
5. Check the licence and the label
Confirm the plan you are on grants commercial rights for the audio. ElevenLabs includes a commercial licence from its Starter plan; the free plan does not list one. Then label the ad: TikTok’s ad policy requires AI-generated content to carry its AI label or a clear disclaimer, and Meta applies its own “AI info” label to ads made with AI tools.
What it costs
Work it out per finished ad, not per month: credits used including retakes, divided by ads shipped.
Pickups
Do I own the audio commercially?
Check which plan you are on. ElevenLabs includes a commercial licence from its Starter plan; the free plan does not list one. That is the difference between a test and something you can put behind ad spend.
Do I have to disclose an AI voiceover in an ad?
TikTok's ad policy requires AI-generated content to carry its AI label or a clear disclaimer of your own. Meta applies its own "AI info" label to ads made or significantly edited with AI tools. Check both before the ad goes live.
How do I keep the cost of retakes down?
Usage is metered in credits, so regenerating one line costs a fraction of regenerating the script. Lock the script before you start generating, not after.
Should the voice match the on-screen person?
If a face is on screen and the voice does not match it, viewers notice before they can say why. Either keep the original voice or cut to footage where no one is speaking.
Source log
Read Nov 15, 2026. These pages change without notice; the source page wins over this one.
