AI Voice Over for YouTube Shorts: Script, Voice, Captions
An AI voice over for YouTube Shorts works when the script is written to be heard and the captions follow the voice word for word. Write short spoken lines and read them out loud once yourself. Pick one voice and keep it for the whole channel. Then cut the picture to the voice, not the other way round: the voice sets the pace, and everything else follows it.
Why do most AI voice overs sound flat?
The script was written to be read, not heard. A sentence that looks fine on a page (“The platform offers numerous features designed to improve productivity”) falls apart out loud. Spoken lines are shorter, use plain words and put the important word last.
Three habits fix most of it:
- One idea per sentence. If you need a comma to breathe, split the line.
- Numbers written the way they’re said. Write “twenty twenty-six” if you want it read that way, and “ten K subs” instead of “10k”.
- A turn every few lines. “So why does this work?” gives the voice somewhere to land.
How long should the script be for a Short?
Count words, not seconds. Many voices read somewhere around 150 words a minute at a normal pace. That puts a 30-second Short at roughly 70 to 80 words, and a 50-second one near 120. Read your draft out loud with a timer once; your number beats any rule of thumb.
| Short length | Script length to aim for | What fits |
|---|---|---|
| 15 s | 35 to 40 words | One claim and one proof |
| 30 s | 70 to 80 words | Hook, three beats, a closing line |
| 45 to 60 s | 110 to 140 words | A short story or a mini tutorial |
Cut every line that doesn’t move the viewer forward. The first draft is usually a third too long.
How do you write the first line?
Lead with the payoff or the tension, not with a greeting. “This mistake cost me my first 10,000 views” pulls harder than “Hey guys, today I want to talk about thumbnails.” Keep the first line under about ten words so the voice says it in two or three seconds, while the viewer is still deciding. Our guide to the first three seconds covers hook patterns in depth.
Which voice should you pick?
Pick by topic, then keep it. A calm, low voice suits history and explainers. A brighter, quicker voice suits gaming and fast tips. Viewers come back for a voice they recognize, so switching every video costs you more than a slightly imperfect match.
Two settings matter more than the voice itself:
- Speed. Slightly faster than a natural read keeps a Short moving. Push it too far and the captions can’t keep up.
- Stability. Lower stability sounds more expressive; higher sounds steady but can turn monotone. Try two takes of the hook and keep the better one.
Should the script be one voice file or one per line?
One per line, or per short paragraph, when you expect changes. If the whole script is one file, fixing a single word means remaking the whole voice. Separate lines also let you leave a breath between them, or hold a line under a reveal, without touching the others.
How do the captions stay in sync?
They should come from the voice’s own word timings, not from a guess. When the timing comes from the audio, each word lights up as it’s said and cuts can land on a word. Nothing drifts after the first few seconds. Our guide to word-by-word captions covers styles, and captions for reels covers size and position.
What should play under the voice?
Keep music well below the voice and let it dip further while the voice speaks. That keeps the energy without fighting the words. The guide to ducking music under a voice shows how. The whooshes and hits that mark each cut are in sound effects for YouTube videos.
How do you make one with Cut Express?
You write the script, or ask Claude to write it with you. Claude does the rest through Cut Express:
- The voice. Claude lists the voices and makes the voice over with the voice, speed and model you pick. Each clip comes back ready, with the timing of every word.
- The picture. Claude adds b-roll for each line: your own clips, or free stock it searches for you (free stock footage).
- The captions. Word-by-word captions in the style you ask for follow the voice automatically.
- The mix. Music ducks under the voice, and the video is leveled to about -14 LUFS so it isn’t quieter than the Shorts around it.
Voices spend credits: 1 credit per 1,000 characters with the Fast voice, 2 with Advanced. A thousand characters is roughly a minute of speech, so a 30-second Short costs about half a credit. Pro includes 200 credits a month and Ultra 500; see pricing.
FAQ
Can I use my own voice instead? Yes. Upload a recording and Claude can transcribe it for captions and cut to it the same way.
Which languages work? The voices read many languages. Set the language when you ask, or let it be detected from the script.
Does the voice sound robotic? Short spoken sentences and a slightly expressive setting go a long way. Listen to the hook before you render the whole video.
What if I change one line? Ask Claude to remake that line and render the next version. The rest of the edit stays.
More on the features and in the FAQ.
Make your next voice over Short with Cut Express
Cut Express is the video editor you drive from Claude: a script in, a captioned Short out. See the plans or read how editing with Claude works.