How to Make YouTube Shorts Ads With AI (Vertical Creatives That Survive the Skip Button)

YouTube Shorts now serves more than 70 billion daily views, and Google has built its Demand Gen campaign type around that inventory, placing short-form video directly inside a performance buying workflow with purchase-intent signals from search. For DTC brands already running TikTok and Reels, adding Shorts is less a creative challenge and more a production one: you need vertical 9:16 video, you need enough hook variants to actually learn, and you need to ship them fast enough that your test data is still relevant by the time the campaign exits the learning phase.

That is exactly where making youtube shorts ads with ai changes the economics. Instead of booking a creator for every new angle, you feed one product photo and a short brief into an ai shorts ad generator and get back a batch of ready-to-post videos, each opening on a different hook, formatted for Shorts from the first frame. This guide walks the full process, from asset prep to Demand Gen setup, with practical hook timing notes and a budget split you can use as a starting point.

Step 1: Prepare Your Product Asset and Brief

What you need before you generate anything

The single most important input for any AI-generated video is a clean product photo. Aim for:

• A hero shot on a white or near-white background, easier for generative models to composite into scene-native footage
• 1080x1080 resolution or higher
• No heavy drop shadows or complex glass reflections that will confuse depth estimation

Beyond the photo, write a brief of roughly 50 to 100 words covering: who the product is for, the core benefit you want communicated, the tone (energetic, calm, educational, playful), and the call to action you want voiced at the end. The brief does not need to be a full script. The AI generates hook and narration logic from the brief combined with the product visual context.

If you are unsure which hook structures convert on short-form video, reviewing a solid resource on ugc ad script structure before writing your brief will save you iteration cycles later.

Step 2: Generate a Batch of Hook Variants

Why batching beats polishing one video

Hook performance on Shorts is highly variable. Google's own creative guidance makes clear that the first three seconds must earn the next three. Producing a single polished video is a low-information strategy because you have no signal until you spend real budget finding out you had the wrong hook.

A proper ai shorts ad generator like Sepia (sepia-lab.com) produces multiple video variants in one job, each opening on a different hook type. Common hook categories to request in your brief include:

• Problem statement: "If you still do X, you are losing money every month"
• Curiosity gap: "We tested 30 of these so you do not have to"
• Social proof reference: "This is why 40,000 people made the switch"
• Demonstration open: show the product doing something surprising in the first two seconds before any narration

Sepia uses models like Seedance, Veo, and Kling for the footage layer, ElevenLabs for voice, and auto-generates captions and background music. Because it works from a single product photo, there is no actor booking, no location scout, and no shoot day. You pay per batch on a credit system with no subscription required.

Data on hook rate benchmarks consistently shows that the gap between a strong and an average hook can be two to three times the three-second view rate. Generating five to eight variants rather than one is the fastest path to a winner without inflating production cost.

Step 3: Format for YouTube Shorts Compliance

Specs that will get your ad rejected if you miss them

Before uploading to Google Ads, run through this checklist:

Requirement	Spec
Aspect ratio	9:16 vertical only
Resolution	1080x1920 minimum
Duration	15 to 60 seconds for most Demand Gen placements
Captions	Burned-in or SRT file; auto-captions do not satisfy ad compliance
Audio	Must have an audio track; silent videos are suppressed algorithmically
Safe zone	Keep all text and UI elements at least 20% from the top and bottom edges

Sepia outputs at 9:16 natively with burned-in captions and a music layer, so most of the format work is handled automatically. The one thing to double-check manually is the safe zone: Shorts overlays the channel name and engagement buttons over the bottom 15 to 20 percent of the frame, and a subscribe prompt over the top right. Any price callout or text hook placed in those zones will be hidden for real viewers even though it appears fine in preview.

Step 4: Configure Your YouTube Shorts Demand Gen Campaign

Settings that specifically affect Shorts delivery

Demand Gen is the campaign type that gives you the clearest path to Shorts inventory. Inside Google Ads:

1. Create a new campaign and select Demand Gen as the objective.
2. At the ad group level, select Shorts as the only placement if you want isolated Shorts performance data. Google defaults to mixing Shorts, in-stream, and in-feed; isolating the placement gives you cleaner creative signal for each format.
3. Upload each hook variant as a separate ad, not as multiple assets inside one ad unit. Separate ads produce separate performance rows in reporting, which is how you identify the winning hook.
4. Set bidding to Target CPA or Maximize Conversions for learning-phase efficiency. Manual CPV is harder to optimize against a purchase goal.
5. Enable Optimized Targeting only after the campaign has accumulated at least 50 conversions. Before that threshold, it can drift audience outside your intended segment.

Budget for the learning phase at roughly ten times your target CPA per day, or a minimum of 50 USD per day if your CPA target is below that. Shorts CPMs run lower than in-stream inventory in most verticals, which is a meaningful part of the format's appeal for DTC brands.

Step 5: Time Your Hook for the Skip Button

What the first three seconds actually need to do

YouTube Shorts ads are skippable in most placements within the first few seconds. The opening frame must create enough pattern interrupt that the viewer pauses mid-swipe. Specific techniques that work in AI-generated footage:

• Open on motion, not a static product reveal. A hand picking up the product, a pour shot, or a fast zoom-in all outperform flat product displays because motion registers in peripheral vision before conscious attention kicks in.
• Voice the hook within the first 1.5 seconds, not after an establishing shot. Narration carries the viewer past the skip impulse before they make the decision.
• Layer a text overlay as a second hook signal. A bolded callout in the first two seconds reinforces the audio hook visually and catches muted viewers simultaneously.
• Skip long logo animations or intro cards entirely. These are reliable skip triggers in the Shorts feed.

When generating with Sepia, you can specify in the brief that the opening should be action-first. The underlying video models (Kling, Veo, Seedance) produce motion-native clips rather than slideshow-style transitions, which helps significantly with that first-frame pattern interrupt.

Step 6: Split Your Budget Across TikTok, Reels, and Shorts

A starting allocation for multi-platform testing

If you are already spending on TikTok and Reels and want to add Shorts without disrupting your existing performance baseline, a practical starting split for most DTC brands is:

Platform	Starting allocation	Rationale
TikTok	50%	Largest short-form audience in most DTC verticals and the most mature creative optimization loop
Meta Reels	30%	Strong retargeting integration with the Meta pixel ecosystem and lookalike infrastructure
YouTube Shorts	20%	Lower CPMs, a growing logged-in audience, and Demand Gen's purchase-intent signal from adjacent search behavior

As data accumulates, shift budget toward the platform generating the lowest cost per purchase, not the lowest CPM. Shorts often wins on CPM but must be evaluated on downstream conversion quality relative to your specific product category and customer acquisition cost targets.

For context on how much creative volume you need to sustain a multi-platform test without hitting fatigue before reaching statistical significance, the ad creative volume benchmarks guide provides platform-specific thresholds that are useful as planning inputs.

FAQ

Can AI-generated video actually run as a YouTube Shorts ad?

Yes. Google Ads does not require human-filmed content. The policy requirement is that the video must be original content, not reposted from another creator's channel, and compliant with Google's advertising policies for the relevant category. AI-generated footage built around a real product meets both conditions. Brands have been running AI-generated UGC-style ads on Shorts, TikTok, and Reels without policy issues as long as the content does not violate category restrictions (supplements, financial products, etc.) or make prohibited claims.

How many hook variants should I test per product?

A minimum of three to five variants per product. Below three, you cannot distinguish hook performance from natural CPM variance in the data. Above eight, you need a significantly larger daily budget to give each variant enough impressions to reach any kind of significance. For most DTC brands running Shorts at 50 to 200 USD per day, four to six variants is the practical sweet spot that balances learning speed against budget dilution.

How is Sepia different from an avatar platform?

Sepia is not an avatar library. It does not offer a roster of digital human characters you place in front of a branded background. Instead, it generates product-forward footage using video generation models (Seedance, Veo, Kling) from your actual product image, adds AI voice narration via ElevenLabs, and layers in burned-in captions and background music automatically. The output is UGC-style creative built around the product itself, without requiring a creator, a character selection, or a studio setup.

Do Shorts ads need to be different from the TikTok or Reels versions of the same creative?

The format spec is identical across all three platforms (9:16, vertical, captions, audio), so the same render can technically run on all three without re-encoding. In practice, the hook style that performs best can vary by platform: Shorts viewers often arrive with adjacent search intent and respond to slightly more informational or benefit-forward hooks, while TikTok skews toward trend-native and pattern-interrupt formats. The efficient workflow is to generate a batch with varied hook styles using a tool like Sepia, distribute the same batch across platforms, and let each platform's algorithm surface whichever hook style resonates most with its specific audience composition.