Storyboard Ads StudioStoryboard Ads Studio

AI-generated video ads: what you need to label before December 2, 2026

Team at Storyboard Ads StudioWritten with AI assistance16 min read

In summary

  • The Article 50 of the European AI Regulation (Regulation EU 2024/1689) applies from August 2, 2026, and the transitional period for technical marking for systems that were already on the market before August expires on December 2, 2026, according to the FAQ and guidelines on AESIA transparency.
  • Not everything has a visible label: the machine-readable marking is the general obligation of the system provider; the visible notice falls to you as the advertiser when the video contains a deepfake or someone who appears to be a real person.
  • A faceless advertisement —product, environment, B-roll, animation without people— is the least exposed case: it is almost always sufficient for the content to be detectable as synthetic.
  • The maximum penalty under the Regulation is 15 million euros or 3% of global revenue, whichever is greater.
  • If you sell outside the EU: Pennsylvania requires notice in advertising with misleading synthetic content from December 21, 2026, and New York mandates warning about "synthetic interpreters" from June 9, 2026.

If you produce video ads with AI —as a creator, as the marketing manager of an ecommerce, or from a small agency— the deadline has crept up on you without anyone sending a warning. This guide translates into production language what the regulation states: what needs to be labeled, what does not, where to place the label, and what to keep from the process. In the end, you will see how a tool like Storyboard Ads Studio, the studio for creating video ads with AI scene by scene by AIsToHire, fits in, because compliance is much better resolved within the editing than as a later patch.

Vertical video ad generated with AI on a mobile phone with a discreet label at the bottom of the screen
Vertical video ad generated with AI on a mobile phone with a discreet label at the bottom of the screen

What exactly does Article 50 of the European AI Regulation require of you?

Article 50 does not say "label everything you do with AI." It establishes four distinct obligations depending on your role and the type of content: to notify when a person interacts with an AI system, to mark synthetic results in machine-readable format, to inform in emotion recognition systems or biometric categorization, and to clearly label deepfakes and publicly relevant texts published without human review. For a video ad, the ones that affect you are the second and the fourth.

The key difference, and the one that generates the most confusion, is who bears each obligation:

  • The provider of the AI system (the tool that generates the video) is the one who must ensure that the output is identified in a machine-readable format and is detectable as artificially generated or manipulated, with markings that AESIA describes as "effective, reliable, robust, and interoperable."
  • The deployer (you, who publishes the ad) is the one who must visibly reveal that the content is artificial when it comes to a deepfake.

As the analysis by Economist & Jurist on what the Regulation requires and what penalties it foresees (August 11, 2026) reminds us, machine-readable marking and the visible label are not the same and neither automatically satisfies the other: the applicable obligation depends on the role and the type of content.

To ground the "how," the European Commission published in June 2026 the Code of Good Practices on Transparency of AI-Generated Content, a voluntary instrument that covers sections 50(2), 50(4), and 50(5). Signing it provides predictability; not signing it requires demonstrating compliance through other means, and supervisory authorities may request additional information from non-signatories.


Why is December 2, 2026 the date that matters?

Because on August 2, 2026, the general application of Article 50 began, but a limited transitional period was left open for the marking of content generated by systems that were already marketed before that date: this margin ends on December 2, 2026, according to AESIA guidelines. From that point on, the detectability of synthetic content is required without a buffer.

In practice, this means two very concrete things for those producing advertisements:

  1. The tools you use should already be marking their output. If your video generator does not incorporate content credentials or technical watermarking, after December 2, you will be left holding the problem.
  2. Your visible notices in cases where they are required did not have a grace period: they have been required since August.
DateWhat happensSource
June 10, 2026The Code of Good Practices on transparency of AI-generated content (voluntary) is publishedEuropean Commission
August 2, 2026General application of Article 50: machine-readable marking and labeling of deepfakesAESIA
December 2, 2026The transitional period for technical marking for systems already marketed before August expiresAESIA
December 21, 2026The law against misleading advertising with AI from Pennsylvania comes into forceReed Smith

And the penalties?

The ceiling for administrative fines is 15 million euros or 3% of the total worldwide turnover of the previous financial year, applying the higher figure. These are ceilings, not automatic fines: the authority assesses the severity, intent, and mitigation measures. For a small agency or an e-commerce, the realistic expectation is not the maximum fine, but the requirement, the withdrawal of the advertisement, and the cost of redoing the campaign.

In Spain, the Organic Law Project for the proper use and governance of AI —approved by the Council of Ministers on May 26, 2026, and published in the Official Bulletin of the Cortes on June 12, 2026— designated AESIA as the authority and specified the national sanctioning regime in three tiers (up to 35 M€ or 7% for prohibited practices; up to 15 M€ or 3% for serious infringements; up to 7.5 M€ or 1.5% for minor ones), but it lapsed on October 6, 2026 when the Cortes were dissolved, without being approved. According to the analysis of the sanctioning limbo from August, until there is a national law, AESIA's ability to impose financial penalties is limited and its actions mainly consist of requirements and warnings. Translation for you: the obligation has existed since August; the Spanish sanctioning muscle comes later. It is not an excuse for non-compliance; it is an indication that right now what you will receive first is a requirement or the withdrawal of the creative by the platform.


What needs to be visibly warned and what is sufficient to mark technically?

The practical rule: if the video can make the viewer believe that what they see is real when it is not, the warning must be visible. If the video is evidently a product piece, an environmental shot, or an animation without people, the relevant obligation is technical detectability, which mainly falls on the tool that generated it.

Comparison between an ad with a synthetic person that requires a visible notice and a product-only ad that only needs technical marking
Comparison between an ad with a synthetic person that requires a visible notice and a product-only ad that only needs technical marking
What appears in your adRiskWhat you need
Synthetic person speaking to the camera as if realHighVisible notice + technical marking
Testimonial or review from a generated "customer"HighVisible notice + technical marking
Face or voice of a recreated real personVery highVisible notice, technical marking, and image rights
Demonstration that the video falsifies (the product does something it does not do)Very highRedo the scene; no label fixes a false claim
Product shot, still life, macro of textureLowTechnical marking
B-roll of environment, backgrounds, skies, interiorsLowTechnical marking
Animation of graphics or packaging, without peopleLowTechnical marking

Be careful with the fourth row, which is the one that takes down the most campaigns: the label "content generated with AI" does not legalize a false claim about the product. If the scene shows a capability that the product does not have, the problem is no longer about transparency regarding AI; it is about misleading advertising, and there the consumer protection regulations come into play.

Why the faceless ad is the least exposed case

An ad built with product, environment, and typography —the faceless format that dominates today in Reels, Shorts, and TikTok— is the lowest risk profile of those you can produce with AI: without synthetic people, there is no deepfake, without testimonials there is no deception about the authenticity of a review, and what remains is a marking obligation that is resolved with the tool and a discreet label.

This is not a forced interpretation. The New York law itself, according to the summary from the Honigman law firm on the Synthetic Performer Disclosure Law, targets synthetic human interpretations, not AI content in general: a "synthetic performer" must create the impression that a person appears in the ad, so common uses such as image editing, content enhancement, background generation, or non-human graphics may fall outside its scope.


What if you sell in the United States? Pennsylvania and New York already have rules

If your ads are served in the United States, the EU is not your only front: New York requires as of June 9, 2026 to visibly notify when an ad uses a synthetic performer, and Pennsylvania requires as of December 21, 2026 a clear notice when synthetic content creates a false impression about what is being sold. These are state laws, but they apply where the ad is seen, not where your company is located.

Three regulatory frameworks for AI advertising labeling connected to a world map: European Union, Pennsylvania, and New York
Three regulatory frameworks for AI advertising labeling connected to a world map: European Union, Pennsylvania, and New York
European Union (art. 50)Pennsylvania (AI Deceptive Advertising Act)New York (Synthetic Performer Disclosure Law)
In effectAugust 2, 2026; transitional labeling until December 2, 2026December 21, 2026 (signed on October 22, 2026)June 9, 2026
What triggers the obligationSynthetic content (labeling) and deepfakes (visible notice)Synthetic content that creates a false impression about the product or serviceAd with a synthetic performer that appears human
Type of noticeMachine-readable labeling + clear label on deepfakes"Clear and visible" notice that the ad contains AI-generated or modified content"Conspicuous" notice within the ad itself
Relevant exceptionsVisible notice is not required if it is evidentObviously fictitious content, parody, satire, commentary, criticism, and text-only contentPlatforms and publishers that only host the ad
ConsequencesUp to €15 million or 3% of global revenueAction by the attorney general or district attorney; civil fines of up to $3,000 plus costs and fees$1,000 for the first offense and $5,000 for subsequent offenses

Sources: AESIA for the EU; Reed Smith on the enforcement of Pennsylvania's law on December 21 and Morgan Lewis on Pennsylvania's new advertising disclosure law; Honigman and Kelley Drye on the enforcement of New York's law for New York.

Two details from Pennsylvania that are worth noting: the law does not apply to text-only content or to what is obviously fictitious, parodic, or satirical, and the trigger is not "having used AI," but that the synthetic content generates a false perception about the nature, origin, or authenticity of what is being sold.


Why is this no longer a theoretical debate right now?

Because AI video generation has gone from being a specialist task to being a button within the ad manager. In October 2026, during Advertising Week in New York, Meta announced that in Advantage+ creatives "advertisers and agencies can now generate video from existing static assets, a feature that is already generally available," according to their announcement of new AI capabilities for campaigns.

That is to say: any advertiser can turn a product photo into a video without leaving Ads Manager. Meta also added, in beta phase, the bulk conversion of catalog images into product videos from the Advantage+ video generator and the product video creator in Commerce Manager.

And platforms are not waiting for you to declare anything. TikTok explains in its information on how to identify AI-generated content that it has labeled more than 3 billion videos as AI-generated content by combining Content Credentials (the C2PA standard), creators' self-labeling tools, and invisible watermarks that only TikTok can read, which are harder to remove. Translation: if you do not label, it is very likely that the platform will label it on its own, and then you lose control over how that notice appears in your creative.


How to place the notice without ruining the conversion?

What kills the conversion is not the notice, it is putting it wrong: a giant watermark over the product or a label that occupies the hook for the first two seconds. A notice works and does not hinder if it is readable, discreet, stable, and appears when the viewer has already entered the video.

Storyboard of five scenes of an ad without people with the notice label placed in a scene after the initial hook
Storyboard of five scenes of an ad without people with the notice label placed in a scene after the initial hook

Production checklist, in the order you will need it:

  1. Decide scene by scene if people appear. It is the decision that determines everything else: with realistic synthetic people you enter the territory of visible notice; without them, you do not.
  2. Discreet label, not watermark. Small text in a lower band or in a corner, with enough contrast to be read on mobile. "Content generated with AI" or "Images generated with AI" comply and fit.
  3. Never in the first two seconds. The hook is what decides whether the ad is watched. Place the label starting from the second scene and keep it visible long enough to be read.
  4. Do not place it only at the end. A notice that appears when 80% of the audience has already scrolled does not inform anyone.
  5. Repeat the notice in the post text. A line in the description or in the ad copy, in addition to the on-screen label. It’s free, and it’s the first thing anyone questioning it will look at.
  6. Also use the platform’s auto-labeling. If you mark the content as AI-generated, you control the message; if the platform detects it, it applies the label itself.
  7. Review product claims scene by scene. Any shot that shows a capability that the product does not have is redone, not labeled.

What to keep from the process

Keep what allows you to reconstruct how the ad was made if a platform reviews it or a consumer questions it: the brief or starting idea, the breakdown of scenes with their description, the intermediate versions of each scene, and the final piece with the notice incorporated. There is no mandatory file format for a small business, but being able to show that a scene is a product render and not the recording of a real person resolves in an email what otherwise would be a discussion.

This is, at its core, a matter of tools: if the storyboard, the versions of each scene, and the labels of the edit live in different places, the file does not exist. If they live in the same flow, it exists without you doing anything extra.


Frequently asked questions about labeling video ads made with AI

Do I have to label an ad if I only used AI for the background or to retouch the image?

The general obligation that affects that case is technical labeling, and it falls on the provider of the system that generated the content. The visible notice of article 50 is directed at deepfakes and content that can make it seem like something real happened. A generated background or a retouch does not turn your ad into a deepfake, and in the case of New York, the generation of backgrounds and non-human graphics may fall outside the reach of the synthetic performers law.

What happens if I do not label my AI-generated video ad?

In the EU, the maximum penalty is 15 million euros or 3% of global revenue, but those are maximums and not automatic fines: the authority weighs severity, intent, and mitigation. What you will see first, especially being a small business, is a request, the withdrawal of the creative, or the label that the platform applies on its own when it detects synthetic content.

Is it enough to put "generated with AI" in the post description?

As the only notice, it is not sufficient when the video contains a deepfake or a person who seems real: in that case, the rule requires a clear disclosure within the ad. The description is a highly recommended reinforcement, never a substitute for the on-screen label.

Does an ad without people, just a product, need to be labeled?

It is the case with the least exposure. Without synthetic people, there is no deepfake to disclose, so the relevant obligation is that the content is detectable as artificially generated, which is covered by the tool's labeling. Adding a discreet label is still good practice, and platforms reward it more than they penalize it.

Does Pennsylvania law affect me if my company is in Spain?

If your ads are served to consumers in Pennsylvania, yes: from December 21, 2026, the notice is required when synthetic content creates a false impression about what you are selling. Text-only content, clearly fictional content, and parody, satire, or criticism are excluded.

How long should the label be on screen?

The rule does not set seconds. The criterion is that it is clear and perceivable: large enough and contrasted to be read on a mobile and long enough for an average viewer to read it. In a vertical video of 15 to 30 seconds, a visible lower band for several seconds starting from the second scene meets that criterion without covering the product.

What documentation should I keep for each ad?

The starting brief, the breakdown of scenes, the intermediate versions of each shot, and the final piece with the notice. With that, you can demonstrate what part of the video is synthetic, what decisions were made, and that the notice was there from the edit.

Fulfilling this is a step of the editing process, not a later patch

Everything above boils down to three production decisions: what appears in each scene, where the label goes, and what is recorded from the process. If all three are made in the same tool where you edit the ad, compliance stops being a separate task.

This is where Storyboard Ads Studio fits in, the ad studio of AIsToHire: the ad is defined scene by scene, with its description and movement, so you decide from the start whether people appear or if you stay in the low-risk territory of the product and the environment; the labels and texts are added in the final edit, along with subtitles and the logo, so the AI notice comes in as one more element of the edit and not as a last-minute touch-up; and the storyboard and the version history of each scene are recorded, which is exactly the material you will be asked for if a platform or a consumer questions the ad.

If you are going to continue producing video with AI after December 2 —and you will continue—, it is worth doing it in a flow where editing a video ad scene by scene with the included AI notice is the normal path and not the exception.

Build your ad scene by scene, with the AI notice already included in the editing — create your free account

Build your ad scene by scene, with the AI notice already included in the editing — create your free account

↑ Back to top