How to Make YouTube Shorts With AI

YouTube Shorts rewards a specific set of behaviours, and most of them have nothing to do with production quality. This is a process for making Shorts with AI that respects what the format actually optimises for.

What Shorts is measuring

Before the craft, the mechanics. Shorts distribution is driven primarily by:

• Retention in the opening seconds. If viewers swipe immediately, the video stops being shown.
• Watch-through and loops. Short videos that get rewatched signal quality more strongly than long ones that get finished once.
• Consistency of posting. The system needs repeated signal to work out who your audience is.

Notice what is missing: resolution, cinematography, edit polish. Those affect whether people subscribe, not whether the video gets shown. Optimise for distribution first.

Step 1 — Write the hook before anything else

The first line does more work than the rest of the video combined. Write it first, in isolation, and write several.

Hooks that work fall into a few shapes:

• The claim. "You are using AI video generators wrong."
• The question with a stake. "Why do your Shorts stop at 200 views?"
• The result first. "This took four minutes to make."
• The contradiction. "Longer videos are easier to rank than Shorts."

Hooks that do not work: introductions, context-setting, "in this video I'm going to". There is no budget for a preamble. Start at the point.

Step 2 — Keep the script to the length

A sixty-second Short is roughly 130 to 150 spoken words. Most drafts are two to three times that, which produces narration so rushed the captions cannot keep up.

Write the script, count the words, and cut to the number. This is the least glamorous step and the one that most improves the finished video.

One point per Short. If you have three points, that is three Shorts, and three Shorts will outperform one video containing all three.

Step 3 — Pick the generation mode to match the content

This is where AI tooling gives you a real choice, and the wrong choice is the most common reason a Short underperforms.

Story, history, facts, explainers. Use AI image shorts — a generated still per scene with camera motion. This is the format behind most of the story content in your feed. It renders fast, costs little, and avoids the motion artefacts that pull viewers out.

Anything advice-shaped — finance, health, education. Use stock footage shorts. Real footage reads as more credible, and in categories where trust drives the follow, viewers spotting AI-generated imagery works against you.

You need real motion on screen. Use the AI shorts generator — generated footage handles movement that stills and stock cannot.

Direct address to camera. Use avatar shorts. Faces hold attention, and a talking head is the most-used Shorts format for a reason.

Step 4 — Get the vertical framing right

Do not crop a widescreen video to 9:16. Cropping throws away the sides of every frame, and the subject is frequently in the part you threw away. Generating natively vertical means the composition was built for a phone.

Two things to check on every Short:

• Captions inside the safe zone. YouTube overlays a title, channel name, and action buttons on the video. Text underneath any of that is text nobody reads. Keep captions in the middle band.
• Subject sized for a small screen. What reads on a monitor can be illegible on a phone held at arm's length. When in doubt, go closer.

Step 5 — Captions are not optional

A large share of Shorts viewing happens with sound off. The caption carries the content in those sessions, which makes it a primary element rather than an accessibility addition.

Burned-in captions styled for vertical, timed to the narration, positioned clear of the interface. Every generator on Vidnebu produces these automatically, but check the placement on the finished video before posting.

Step 6 — Publish enough to learn something

This is where most people fail, and it is not a craft problem.

Which hook works is genuinely unpredictable. Not hard to predict — unpredictable. The only reliable method is to publish several and read the retention data. That means your production process has to support volume, which is the actual argument for generating rather than filming: the marginal cost of a fifth variation is credits, not another afternoon.

A workable cadence:

1. Write five hooks against one idea.
2. Generate all five.
3. Post them across a week.
4. Keep the shape that retained best; discard the rest.
5. Write five variations of the winner.

What to check before posting

• Does the first line work with no context?
• Is the script under 150 words?
• Are captions clear of the top and bottom interface zones?
• Does it make exactly one point?
• Would it still make sense with the sound off?

Cost

Shorts are the cheapest thing to produce here because they are short — cost scales with length. AI image shorts and stock shorts are the most economical modes, which is what makes a daily cadence realistic. Estimate a specific video with the credit calculator before committing to a batch.

For the vertical-specific product detail, see the AI shorts generator, and for the underlying mechanics read how AI video generation actually works.