The AI advertising rules brands spent the summer preparing for are now in force. New York’s synthetic performer disclosure law has applied since June 9, 2026. The EU AI Act’s Article 50 transparency rules and California’s AI Transparency Act followed on August 2. Two more land within weeks: Pennsylvania’s AI Deceptive Advertising Act on December 21, 2026, and California’s synthetic performer law for ads (SB 1050) on January 1, 2027.

For the first time, brands running user-generated content (UGC) and paid social campaigns face enforceable UGC AI compliance requirements across multiple jurisdictions at once. The core rule across all of them: if your ad creative uses AI to generate or substantially modify a human likeness (face, voice, body), disclosure is required. Pennsylvania and the EU also reach AI that makes your product look different from what people will actually get. If AI is used for routine production tasks like color grading, noise removal, or copy assistance, it is not.

We have already covered the regulatory landscape in detail: what each US jurisdiction requires, how the EU AI Act works, and what the rules look like across eight additional global markets. This guide focuses on what you should actually do about them.

Which AI ad rules apply now, and what’s next

Pennsylvania is the one to watch. It doesn’t mention performers at all. Instead, it asks whether AI content creates a false impression about “the nature, origin or authenticity” of what you sell (Kelley Drye).

There’s no knowledge requirement, and law firms say it may apply to ads that reach Pennsylvania consumers wherever the advertiser is based. The label must say the ad contains content modified or generated using AI, shown when that content first appears (Holland & Knight).

We cover New York and California in detail in our US AI regulations guide, and Article 50 in our EU AI Act guide.

The transparency trigger: when AI use requires disclosure

The biggest source of confusion in AI advertising compliance is figuring out which AI uses actually require a label and which ones do not. Disclosure is triggered by materiality, specifically whether AI use could mislead consumers about the content’s authenticity or a person’s identity.

That principle gives brands a practical decision tool. We call it the Transparency Trigger.

What triggers disclosure: synthetic humans, AI voice cloning, digital twins, AI-generated testimonials

If your ad creative contains an AI-generated or substantially modified human likeness, disclosure is required. This is consistent across New York’s Synthetic Performer Disclosure Law, the EU AI Act’s Article 50, and all three major ad platforms. The definitions vary slightly, but they all converge on the same trigger: AI-generated avatars, digital spokespeople, synthetic voices, and simulated extras in video ads all require disclosure.

Pennsylvania’s law, from December 21, goes wider. It also covers AI images, video or audio that could mislead people about the product itself, such as an AI-enhanced before-and-after.

Liability falls on whoever produces or creates the advertisement, meaning the brand, the agency, or the production company. Platforms are exempt. Pennsylvania puts the duty on the advertiser that creates or disseminates the ad, or causes it to be, and has no knowledge requirement.

Our posts on US, EU, and global regulations cover each jurisdiction’s specific definitions, penalties, and timelines in full.

What doesn’t trigger disclosure: color grading, noise removal, background blur, AI copywriting assist, optimization algorithms

Not every use of AI in your creative workflow triggers a compliance obligation. The IAB framework explicitly exempts routine production tasks from its disclosure requirements. AI-assisted editing such as color grading, noise removal, and background blur falls squarely outside the scope. So do optimization algorithms for bid management and audience targeting, AI-powered copywriting assistance, and background production workflows like automated transcription or format conversion.

New York’s law carves out similar exemptions for audio-only ads and AI used solely for translation, and does not apply to real human creators filmed on camera. The practical takeaway is straightforward: if your AI tool changes what the viewer sees in a way that could create a false impression about a real person or event, you need to disclose. If it makes the production process more efficient without altering what the viewer perceives as real, you do not.

The IAB materiality test

The IAB framework provides the clearest decision-making tool available. It asks one question: “Does this AI use materially affect authenticity, identity, or representation in ways that could mislead consumers?” If the answer is yes, disclose. If no, move on.

The framework introduces a two-layer compliance model. The first layer is consumer-facing: standardized text labels, badges, icons, or watermarks placed near the ad creative so viewers can see that AI was involved. The second layer is machine-readable: C2PA metadata embedded in the file itself, which supports technical compliance and allows platforms to detect and label AI content automatically.

For your creative team, the workflow looks like this: before any asset goes into production, ask whether AI will generate or substantially modify a human likeness. If yes, plan for both the consumer-facing label and the metadata layer from the start. If no, proceed normally. This two-step check takes less than a minute per asset and eliminates most compliance guesswork before it becomes a problem.

The gray areas

The obvious ends of the spectrum are clear. The following scenarios cover the middle ground where most production teams actually get stuck.

1. You use AI to remove and replace the background behind your creator in a product demo video.

The creator’s face, voice, and body are unchanged. The background is now a clean studio setting instead of a messy kitchen. This does not trigger disclosure. The AI modification affects the environment, not the human likeness. The creator is still authentically themselves.

2. You take a creator’s 60-second testimonial and use an AI tool to generate five hook variations that splice, rearrange, or synthesize the creator’s voice to say opening lines they never actually recorded.

This triggers disclosure. Even though the original voice is real, AI is generating new speech the creator never said. Under the EU AI Act, that’s content that could “falsely appear to a person to be authentic.” From December 21, Pennsylvania’s law covers it too whenever the new line could mislead people about the product, for example a claim the creator never made.

New York’s current law is narrower. It covers AI people who aren’t recognizable as anyone real, so a cloned voice of a real creator may fall outside it, though a pending bill (A11683) would extend it to digital replicas. Either way, every hook variation needs a label, and the creator’s written consent.

3. You use AI-generated 3D product renders in a b-roll section while a real creator appears on camera in other parts of the same ad.

The product renders do not trigger disclosure because they contain no human likeness. The creator segments are real footage. As long as AI is not generating or modifying the human elements of the ad, the combined piece does not require synthetic content disclosure.

4. You run your creator’s video through CapCut’s AI auto-captions feature, which adds stylized animated text overlays.

This does not trigger disclosure. AI-generated captions, text overlays, and script assistance are explicitly exempt on every major platform. YouTube’s policy confirms that AI used for productivity tasks like automatic captions does not require the synthetic content label. The same applies to AI-written ad copy, hashtag suggestions, and script drafts.

5. You use an AI beauty filter that smooths skin, adjusts lighting on the creator’s face, and subtly reshapes facial features.

This is the hardest call. Light touch-up (smoothing, lighting) is comparable to standard post-production and likely falls below the materiality threshold. But if the filter substantially changes the creator’s appearance to the point where it could mislead viewers about what the person actually looks like, it starts to cross the line. The safest approach: if you would not feel comfortable telling the creator “we didn’t change how you look,” the modification is probably material enough to disclose.

Be stricter if you sell skincare, beauty or anything where how it looks is the result. A filter that makes skin look better than your product can deliver is exactly the false impression Pennsylvania’s law targets. The Commission’s Article 50 guidelines also treat an AI image that makes a product look better than it really is as a deepfake.

6. You launch a creator’s video in an Advantage+ campaign with Meta’s creative enhancements switched on.

Some Advantage+ creative enhancements use generative AI to change the asset itself, for example by turning product photos into video, expanding images or generating new backgrounds. Some can be switched on by default. In October, Bookshop.org found that Meta had added an AI illustration to one of its human-made ads after an account update opted it in (Adweek).

Apply the same test as above. If an enhancement changes how the creator looks or sounds, or makes the product look different, it either needs a label or needs to be off. Check which enhancements are on for every creator ad before launch.

Platform terms: who carries the risk when Meta’s AI edits your ad

Laws aren’t the only rules that decide who answers for AI in an ad. Platform terms do too, and Meta has just shown where they’re heading.

In October, 2026, Meta previewed new Commercial Terms with a section on its AI features. Under that wording, advertisers would be solely responsible for whatever Meta’s AI creates or changes in their ads, including features Meta switches on itself (Adweek). Today’s terms tie that responsibility to AI tools you choose to use.

On November 5, Meta paused the update. It won’t take effect on November 20 as planned, no new date has been set, and the existing terms stay in force while Meta consults (Nieman Lab). The pause followed publisher objections to a separate clause about crawling advertisers’ websites, a court injunction in Norway and a UK Competition and Markets Authority investigation.

Imagine you hire a creator to film a 30-second demo for your serum, then launch it in an Advantage+ campaign alongside three product photos. An enhancement you never switched on turns one photo into a video with a glowier “after” than the serum delivers. Under Pennsylvania’s law and the EU AI Act, the disclosure duty for that video would likely sit with you as the advertiser, whatever Meta’s terms say.

So the pause changes the contract, not the advice. Know which AI features touch your creator ads, and switch off the ones you wouldn’t want to disclose. Meta says advertisers can turn off individual Advantage+ features in Ads Manager at any time.

Your compliance audit checklist for Q4 and Q1

If you’re running any AI-generated or AI-edited creative, the New York and EU rules already apply to it, so audit now rather than at the next deadline. Pennsylvania follows on December 21 and California’s synthetic performer law on January 1, both in the middle of holiday campaigns.

If you have AI-generated ads live without disclosure, pause them and add labels before restarting. No AI Act fines or New York enforcement actions have been reported so far, but how long a non-compliant ad runs and how many people see it both count if one comes.

1. Creative audit: flag every AI-generated asset across campaigns

Apply the Transparency Trigger from Section 1 to every asset in your library. The question for each one is the same: “Does this creative contain an AI-generated or substantially modified human likeness?” If the answer is yes, that asset needs disclosure. If no, it can proceed without additional labeling. Include versions Meta creates with Advantage+ creative enhancements, not only the files your team uploads.

2. Disclosure audit: ensure on-ad labels + platform metadata for every flagged asset

For every asset you flagged in the creative audit, verify two layers of compliance. The consumer-facing layer means an on-ad label, a text disclosure, badge, icon, or watermark, that is “conspicuous” in the legal sense: prominent, unavoidable, and noticeable, not buried in fine print. For Pennsylvania, the label needs to say the ad contains content modified or generated using AI. It must appear when that content first appears, in the same medium: on screen for video, spoken for audio.

The machine-readable layer means C2PA metadata embedded in the file itself. Check that your AI tool provider has not stripped this metadata during export or format conversion. Screenshots, social media uploads, and file conversions can all remove embedded metadata, so verify the final published version, not just the original export.

3. Operations check: contracts, platform settings, and documentation

Contracts. For every active creator relationship, confirm the agreement covers your current usage (organic, paid, whitelisting), verify geographic scope matches where your ads actually run, and check whether exclusivity terms are still active.

Platform settings. Each platform has its own disclosure mechanism and they need to be checked independently. On Meta, open Advantage+ creative for each ad and switch off enhancements that alter the creator or the product. Meta labels ads made with its own generative AI tools and, since June 2026, says it also detects ads made with other AI tools through provenance signals such as C2PA metadata. Don’t rely on detection alone: a file that has lost its metadata may not be caught, so keep your own record and disclose wherever the law requires it. On TikTok, confirm the self-disclosure toggle is on. On YouTube, select the “altered or synthetic content” disclosure in the upload flow.

Audit trail. For each AI-generated asset, log the tool used, the prompt or input, the output, who approved it, and when. Define your pass/fail criteria upfront so your team has clear thresholds rather than case-by-case judgment calls.

Why human UGC sidesteps most of this

Every regulation, framework, and platform policy covered in this guide targets AI-generated content specifically. Real people filmed on camera fall outside every AI disclosure definition, whether it is NY’s “synthetic performer,” the EU’s “deep fake,” or any platform’s labeling policy.

That holds as long as the footage stays as filmed. Once AI changes how a creator looks or sounds, or what they say, the gray areas above apply.

The performance case: attention metrics, authenticity premium, trust data

The compliance advantage of human UGC is clear, but the performance case is worth stating as well. In controlled testing, human UGC usually achieves stronger attention metrics, with AI UGC videos typically reaching 85% to 110% of the click-through rate of well-performing traditional UGC. The gap widens in categories where trust and authenticity drive the purchase decision, which includes most DTC verticals.

A 2026 Ipsos study of 20 real brand ads points the same way: human-made ads scored 14% higher on short-term creative effectiveness and 17% higher on long-term brand equity than AI-generated ones (full breakdown).

The consumer trust data reinforces this. The Yahoo and Publicis Media survey found that while proactive AI disclosure helps compared to hidden use, human-created content avoids the trust question entirely. There is no “authenticity penalty” to manage, no perception gap to bridge, and no disclosure label that might lower perceived naturalness or purchase willingness. The content is what it appears to be.

Human UGC vs AI UGC vs hybrid

Choosing between human creators, AI-generated content, and hybrid approaches used to be a performance and budget question. In 2026, it is also a compliance question. Each format carries a different regulatory footprint, and the right choice depends on where your ads run, what category you operate in, and how much disclosure overhead your team can absorb.

Which Approach Fits Which Scenario

The table above gives you the data. Here is how to apply it to common campaign scenarios.

For high-volume creative testing where you need 20 or more variants per concept, AI UGC makes economic sense. The cost advantage is enormous and the iteration speed lets you test hooks, CTAs, and angles that would be prohibitively expensive with human creators. But every single variant needs to pass through a disclosure compliance check before it goes live, so factor that operational cost into your planning. Using human content with ad variations can be another and a safer way to go here.

For hero creatives and brand campaigns where authenticity drives the message, human UGC is the stronger choice. It carries no disclosure burden beyond standard sponsored content tags, and the attention metrics consistently outperform AI alternatives. When consumers see a real person talking about a product, there is no trust penalty to manage.

Some teams are experimenting with a hybrid split: 20% of the creative budget on human UGC hero content and 80% on AI-powered iteration. On paper, that maximizes testing volume. In practice, once you factor in the disclosure compliance overhead on every AI asset, plus the trust penalty data above, the effective cost gap between human and AI creative narrows considerably.

Creator rights and licensing: what the AI wave changes

Most brands running UGC at scale already understand the basics of creator licensing: scope, duration, exclusivity, geographic rights. What the AI compliance wave changes is one specific area that many existing contracts do not cover: modification rights, specifically whether your brand can use AI tools to alter creator footage after delivery, and what happens to your compliance obligations when it does.

Why your contracts need an AI modification clause

The most common source of UGC disputes is a mismatch between what the contract permits and how the brand actually uses the content. In 2026, the highest-risk version of that mismatch is a brand taking human creator footage, running it through an AI tool that substantially alters the creator’s likeness or voice, and publishing it without disclosure. At that point, the content may cross the Transparency Trigger threshold from Section 1, turning a compliance-safe human UGC asset into one that requires synthetic content disclosure.

Your creator contracts should now explicitly address: whether the brand can use AI tools to modify the footage (trimming and color grading are different from face-swapping or voice-cloning), who is responsible for disclosure if AI modification crosses the materiality threshold, and whether the creator consents to their likeness being used as a training input or reference for AI-generated variations and whether you may run the footage through platform AI features such as Meta’s Advantage+ creative enhancements.

How Billo fits into this

Every AI-generated asset needs a disclosure check, a metadata verification, a platform toggle, and an audit trail entry. For teams running dozens of creatives per month, that compliance overhead adds up fast.

Billo sidesteps most of it by design. Every piece of content on the platform is filmed by a real human creator on camera, which means it passes the Transparency Trigger automatically. No synthetic performer disclosure, no metadata requirements, no platform labeling beyond standard #ad tags. Commercial usage rights for paid ads and organic use are included in every order, and you can add Meta partnership ads or TikTok Spark Ads permissions in the same order. The briefing and review workflow is built into the platform, so scope, deliverables, and revision terms are defined upfront.

One caveat: if you run those videos through Meta’s Advantage+ creative enhancements, gray area 6 applies, so check which ones are switched on.

One note on how Billo uses AI: the platform’s creator matching and brief optimization are powered by AI, but that AI operates on workflow and data analysis, not on the content itself. The videos your creators film are entirely human-made, which is the distinction that matters for every disclosure framework covered in this guide.

Summary & next steps

If you take one thing from this AI ad compliance guide, it is one principle: if AI creates or substantially modifies a human likeness in your advertising, or makes your product look different from what people will get, disclose it. If it does not, proceed normally.

Three steps to take now:

1. Run the audit checklist before December 21, when Pennsylvania’s law takes effect. Flag every AI-generated or AI-edited asset, including Advantage+ versions, check labels and metadata, and confirm your creator contracts cover AI edits.
2. Prepare for California on January 1, 2027. SB 1050 covers audio-only ads too, so check AI voiceovers in podcast and streaming audio spots. If you run ads in several states, one label in California’s wording covers New York as well.
3. Build the Transparency Trigger into your standard creative workflow. Before any asset goes into production, ask whether AI will generate or substantially modify a human likeness. If yes, plan for disclosure from the start. If no, move forward without it. This single question, asked consistently, eliminates most compliance guesswork.

The simplest compliance path remains the most direct one: use human creators and reserve AI for analysis and minor editing.

FAQs