Do AI generated video ads work on Facebook and Instagram?

AI generated video ads can work on Facebook and Instagram when they follow the same rules as any good mobile ad: a hook in the first second, the product shown accurately and large, captions for sound off viewing and one clear claim. The model is a production tool. Whether the ad sells depends on the brief, the proof it shows and the testing behind it.

What should a brief for an AI video ad include?

A brief for an AI video ad should include one named buyer, their problem in their own words, a single claim the brand can substantiate, the proof assets available, a locked list of product details, the hook written as its own shot, a shot list of four to six short clips with camera and lighting, the copy layer, the planned variants and the quality checks each clip must pass.

How long should an AI video ad for Meta be?

There is no single right length, but Meta's creative guidance favours short video for mobile placements. A brief of four to six clips of a few seconds each usually cuts to around 15 seconds. Length matters less than structure. The hook has to land in the first second or two and the claim has to be clear before the viewer scrolls away.

Does Meta label AI generated ads?

Meta can add an "AI info" label when it detects industry standard signals that content was made with AI, or when the advertiser discloses it. For ads about social issues, elections or politics, Meta requires disclosure of realistic digitally created or altered images, video and audio. Commercial ads must also follow Meta's Advertising Standards on misleading claims.

Which aspect ratio should I generate AI video ads in for Meta?

Generate AI video ads in 9:16 for Reels and Stories and in 4:5 for Feed, because these are the vertical formats Meta recommends for mobile placements. Generating in the final ratio from the start avoids cropping the product out of frame, which often happens when a 16:9 clip is cut down to vertical.

How many AI video ad variants should I test on Meta?

Brief a small grid where one element changes at a time, for example three hooks with the same body and end card. That gives Meta several distinct creatives to deliver and gives you a clear lesson about which hook works. Changing several elements at once produces more ads but less learning. ## Sources - [Meta Ads Guide: specifications and recommendations for each placement](https://www.facebook.com/business/ads-guide/) - [Meta Advertising Standards](https://transparency.meta.com/policies/ad-standards/) - [Meta: our approach to labelling AI generated content and manipulated media](https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/) - [EU AI Act, Article 50: transparency obligations](https://artificialintelligenceact.eu/article/50/) - [Google DeepMind: Veo video generation model](https://deepmind.google/models/veo/)

AI Video Ads for Meta and How to Brief Them So They Sell

AI video ads for Meta work when the brief fixes three things before anyone opens a model: one buyer, one claim the footage can prove, and a shot list of short clips the model can actually render. Most AI video ads that fail on Facebook and Instagram fail at the brief, because a vague prompt produces pretty footage with no hook, no product truth and no reason to buy. A good brief is a one page document that turns a selling idea into four to six shots of a few seconds each, with the first shot written as the hook.

Key Takeaways

Brief the ad, not the model. Decide the buyer, the problem and the single claim first, then translate that into shots a video model can render in one generation each.
Write the first two seconds as a separate shot with its own description, because on Reels and Feed the opening frame decides whether anyone sees the rest.
Lock the product truth (shape, colour, label, scale) in the brief and check every clip against it, since AI models drift on exactly the details that matter to a buyer.

What an AI video ad for Meta actually is

An AI video ad is a paid social video where some or all of the footage is generated by a video model such as Veo, Kling, Sora, Seedance or Runway, then edited into a finished ad with text, captions, sound and an end card. In practice most ads that perform are hybrids: a real product photo or packshot is used as the starting image, the model adds motion, setting or a person using the product, and a human editor cuts it together.

A brief is the written instruction that connects the marketing idea to the production. For AI production it does two jobs at once. It tells the creative team what the ad must say, and it gives the person prompting the model enough detail that each clip can be generated, checked and regenerated without guessing.

Meta itself sits at the end of this chain. Meta's delivery system decides who sees which ad, and on Advantage+ campaigns it leans heavily on the creative to find the audience. That makes the creative, and therefore the brief, the main lever an advertiser still controls.

What works in AI video ads on Facebook and Instagram

The ads that hold attention on Meta placements share a handful of traits, whether the footage was shot or generated. The difference with AI is that each of these has to be written into the brief, because the model will not supply it on its own.

A hook in the first frame. The product, the problem or a surprising visual appears in the opening second. A slow establishing shot is the most common way AI ads waste their budget.
The product is the hero, not the scenery. Models are good at beautiful environments and weaker at keeping a product accurate. Ads that sell keep the product large, in focus and true to life.
Built for sound off. Meta's creative guidance recommends designing video so it makes sense without sound. That means burned in captions or on screen text for every claim, not just a voiceover.
Vertical first. 9:16 for Reels and Stories and 4:5 for Feed are the formats Meta recommends for mobile placements. Generate in the target ratio from the start, because cropping a 16:9 generation usually cuts off the product.
One idea per ad. A 15 second ad can carry one problem and one claim. Ads that try to list five benefits end up proving none of them.
Proof the viewer can see. A texture close up, a before and after, a product in use at real scale. The footage has to show the claim, because a caption alone does not persuade.

If you want a deeper look at opening shots, our guide to Meta ad hooks that stop the scroll using AI video breaks down hook types shot by shot.

Which AI video ad formats suit which job

Not every format suits every product or every stage of the funnel. The table below compares the formats we see most often in AI produced Meta ads, by the criteria that decide the choice.

Format	What it is	Best for	Main risk	Brief must include
Product hero motion	Packshot animated with camera moves, light and simple physics	Products where look and finish sell (beauty, jewellery, design)	Model warps label or shape	Reference image, locked product details, allowed camera moves
Product in use	A generated person or hands using the product in a setting	Products where the result or ritual sells (skincare, kitchen, fitness)	Hands, faces and scale look wrong	Who uses it, where, the exact action, scale reference
UGC style	Talking head or selfie style clip with a script	Explaining a problem and a reason to believe	Looks synthetic, trust drops	Script, tone, setting, disclosure plan
Problem and solution	Short cut from the problem to the product fixing it	Clear functional benefits	Exaggerated claims in the visual	The real problem, what the product honestly does
Text led motion	Kinetic type over simple generated backgrounds	Offers, launches, retargeting	Looks like every other ad	The one line, the offer, brand colours and font

Pick the format from the claim, not from what the model does best. If the claim is "this serum absorbs in seconds", a product in use shot of skin is the proof, and a cinematic hero shot of the bottle is not.

How to brief an AI video ad for Meta

A brief for AI production is shorter than a traditional film brief in some places and far more precise in others. The steps below are the order we write it in.

Name one buyer. Write a single sentence about who the ad is for and what they already believe. "Women in their thirties who have tried retinol and found it too harsh" is a buyer. "Skincare lovers" is not.
State the problem in their words. One line, phrased the way the buyer would say it to a friend or type into a search bar.
Write the single claim. One sentence the ad must make the viewer believe, and only claims the brand can substantiate under Meta's Advertising Standards and local advertising rules.
List the proof you have. Real product photos, real reviews the brand can quote, ingredient or material facts, real footage. The AI generation builds on this; it never replaces it.
Lock the product truth. Shape, colour, label text, pack size, how it is held, how it opens. Attach the reference images. This list is what every clip gets checked against.
Write the hook as shot one. Describe the first one to two seconds as a separate shot: what is in frame, how big the product is, what moves, and the on screen text.
Write the shot list. Four to six shots, each a few seconds long, each with subject, action, setting, camera move, lighting and aspect ratio. Most current video models render clips of roughly five to ten seconds per generation, so one shot per generation keeps control.
Write the copy layer. On screen text per shot, captions, voiceover script if any, and the end card with the offer and call to action.
Set the variants. Decide in advance what changes between versions, for example three hooks with the same body and end card.
Define done. The checks a clip must pass before it goes into the edit: product accurate, no warped hands or text, safe zones respected, claim visible on screen.

Our performance creative brief template for AI production has these fields laid out as a document you can copy.

How to turn a shot list into model prompts

A shot from the brief becomes a prompt by keeping the same order every time: subject, action, setting, camera, light, style, format. A consistent order makes it easy to change one variable and see what it does.

Here is one shot from a brief for a fictional face serum, written as a prompt:

Close up of a woman's hand pressing a glass dropper onto the cheek, the serum spreading and absorbing into the skin, bright bathroom by a window, slow push in, soft morning daylight, natural skin texture with visible pores, realistic, 9:16 vertical.

Three habits make the difference between usable clips and wasted generations:

Start from an image where you can. Image to video, using a real packshot or an approved still as the first frame, keeps the product accurate far more often than text alone.
Describe motion simply. One camera move and one subject action per shot. Stacked instructions such as "orbit while zooming while she turns" are where models produce artefacts.
Ask for texture. Words like "visible pores", "fabric weave" or "condensation on the glass" push models away from the plastic look that tells viewers the footage is synthetic.

For model specific settings and more prompt patterns, see our prompting guide for AI video models in advertising.

How many variants to brief and why

Meta's delivery system needs several genuinely different creatives to find what works, and AI production makes variants cheap. That does not mean every variant is worth making. Brief variants as a small grid where one element changes at a time.

A worked example: three hooks, two bodies and one end card give 3 × 2 × 1 = 6 ads. If one hook wins across both bodies, you have learned something about the opening, not just that "ad four did well". If you change the hook, the body, the music and the offer all at once, six ads give you six results and no lesson.

A practical rule for the brief is to change the element that matters most first. For cold audiences that is almost always the hook. For retargeting it is usually the offer or the proof. Our AI creative testing framework for DTC brands covers how to read the results and what to brief next.

Finishing the ad for Meta placements

Generated clips are raw material. The finish is where most of the selling happens, and it belongs in the brief too.

Cut to the hook. Trim every clip to the moment it starts moving. Models often spend the first fraction of a second settling.
Check against the product truth list. Reject any clip where the label, colour or shape has drifted, however good it looks.
Add captions and on screen text. Keep text out of the areas covered by the Reels interface at the top and bottom of the frame. Meta's Ads Guide shows the safe zones for each placement.
Mix sound for both states. Music and voiceover for viewers with sound on, captions that carry the whole message for viewers without.
Export per placement. 9:16 for Reels and Stories, 4:5 or 1:1 for Feed, at the resolution Meta lists for each.
Add the end card. Product, offer, call to action. Hold it long enough to read.

Our AI video post processing workflow for ads goes through upscaling, colour and sound in detail.

AI disclosure and Meta's rules

Meta labels content with "AI info" when it detects industry standard signals that an image or video was made with AI, or when the advertiser discloses it. Meta has explained this approach in its announcement on labelling AI generated content. For ads about social issues, elections or politics, Meta requires advertisers to disclose when a realistic image, video or audio has been digitally created or altered.

Commercial ads still fall under Meta's Advertising Standards, which prohibit misleading claims. An AI video that shows a result the product cannot deliver is a misleading ad, whether or not anyone labels it. In the EU, the AI Act adds transparency duties for certain AI generated content, set out in Article 50. Write the disclosure decision into the brief, so it is made once and on purpose rather than at upload.

Adsome is a film and ad production studio in Stockholm that makes AI produced ad creative for brands that spend on paid social, and this brief structure is the one we work from.

Common Mistakes

Briefing the model instead of the ad. A prompt is not a brief. Without a buyer, a problem and a claim, the model produces nice footage that sells nothing.
A slow first shot. Establishing shots, logo reveals and fades in from black spend the only seconds most viewers give the ad.
Letting the product drift. Accepting a clip where the label is misspelled or the bottle is the wrong shape, because the lighting looks good.
Generating in 16:9 and cropping. The product ends up off centre or cut off in the vertical version.
Claims the footage cannot prove. On screen text that promises a result the visuals do not show, or visuals that show a result the product cannot deliver.
Too many variables per variant. Changing hook, body, music and offer at once, so the test teaches nothing.
No sound off version. Relying on a voiceover to carry the message when much of the audience watches muted.

Want this strategy running in your account? Adsome makes the ads and runs the campaigns for e-commerce brands.