---
title: "AI in advertising: disclosure, synthetic media and…"
description: "Four rulebooks at once When you use AI in marketing, four layers of rules can apply at the same time: 1. AI-specific law : EU AI Act Article 50…"
url: https://optimizeall.com/learn/ai-governance-eu-ai-act/ai-in-advertising-and-synthetic-media
updated: 2026-10-05
---

AI Governance & Regulation: EU AI Act, NIST AI RMF and ISO/IEC 42001 · Marketing-specific AI compliance · lesson 15 of 17 · 16 min

# AI in advertising: disclosure, synthetic media and platform rules

## Four rulebooks at once

When you use AI in marketing, four layers of rules can apply at the same time:

1. **AI-specific law**: EU AI Act Article 50 (deepfakes, chatbots), California's AI Transparency Act for large GenAI providers, state voice and likeness laws.
2. **Advertising and consumer law**: misleading advertising rules everywhere (EU Unfair Commercial Practices Directive, UK CAP Code and consumer law, US FTC Act, GCC consumer protection laws), plus endorsement and influencer rules.
3. **Platform policies**: Meta, Google/YouTube, TikTok, LinkedIn and others have their own AI and synthetic media rules, often stricter than law and enforced faster.
4. **Rights of real people**: image rights, voice and likeness laws, data protection, defamation.

The simplest mental model: **AI does not change what is misleading; it changes how easy it is to mislead.**

## Misleading ads: the timeless test

An ad is misleading if it is likely to deceive the average consumer about something material to their decision. AI-specific examples:

- **Exaggerated product imagery**: AI-enhanced "after" photos for a skincare product, or a food shot that looks nothing like the real portion. The UK ASA has repeatedly ruled against ads using filters and edits that exaggerate efficacy; the same logic applies to AI imagery.
- **Synthetic testimonials**: AI-generated "customers" giving reviews. In the US, the FTC's rule on fake reviews and testimonials (effective October 2024) covers AI-generated fake reviews; the EU and UK treat fake reviews as unfair practices.
- **Fake scarcity or urgency** generated dynamically by AI.
- **Virtual influencers**: an AI-generated persona endorsing a product must still disclose the commercial relationship, and should not mislead people into thinking it is a real person's genuine experience.

## Disclosure layers for synthetic media

| Layer | Who | Example |
|---|---|---|
| Machine-readable provenance | Tool provider | C2PA Content Credentials metadata, invisible watermarks |
| Legal disclosure | Deployer/advertiser | "This video contains AI-generated imagery" for deepfakes under AI Act Art. 50(4) |
| Platform label | Platform, often triggered by your self-disclosure | "AI info", "Altered or synthetic content", "AI-generated" labels |
| Advertising disclosure | Advertiser/creator | #ad, "Paid partnership", "Sponsored" for commercial relationships |

These are separate. An #ad label does not disclose AI generation; an AI label does not disclose sponsorship.

## Platform policies to know (verify current versions)

- **YouTube** requires creators to disclose realistic altered or synthetic content (for example making a real person appear to say or do something they did not, altering footage of real events, or generating realistic scenes) using its disclosure setting; labels may appear on the video.
- **Meta** (Facebook, Instagram, Threads) labels content with "AI info" based on industry signals or self-disclosure, and requires advertisers to disclose digitally created or altered content in ads about social issues, elections or politics.
- **Google Ads** requires election advertisers to prominently disclose synthetic content that realistically depicts real people or events.
- **TikTok** requires labeling of AI-generated content that contains realistic images, audio or video, and prohibits certain misleading synthetic content, especially of public figures.

Policies change often. Keep a one-page "platform AI rules" sheet with links and a last-checked date.

## Real people: consent and likeness

- **Get written consent** before creating synthetic images, video or voice of a real person, covering purpose, media, territories, duration, revocation and payment.
- **Public figures** are higher risk, not lower: personality rights, false endorsement, and platform bans on deceptive depictions of public figures.
- **Employees**: a staff member's likeness in an AI avatar needs specific consent that survives the power imbalance. Offer a genuine choice and an exit.
- Laws like the **Tennessee ELVIS Act**, image rights under UAE and KSA law, and **SDAIA deepfake guidance** reinforce this.

## Influencer and creator rules with AI

- Commercial relationships still need clear disclosure (FTC Endorsement Guides, UK CAP Code and CMA guidance, similar GCC rules; the UAE also requires a media license for paid influencer activity, so check current rules).
- If a creator uses AI to make a sponsored piece look like a real product demonstration that did not happen, the product claim can be misleading even with #ad.
- Brands remain responsible for claims made by influencers on their behalf.

## Worked example: a skincare launch across the UK, UAE and US

| Asset | Issues | Decision |
|---|---|---|
| AI-generated "before/after" images | Misleading efficacy; ASA/FTC risk | Replaced with real trial photos, unedited, with test conditions stated |
| AI avatar presenter (fictional) | Not a deepfake of a real person; transparency expected by platforms | "AI-generated presenter" on-screen line; self-disclosed on platforms |
| Founder's cloned voice for radio | Real person's voice: consent + deepfake disclosure in EU contexts | Written consent; "AI-generated voice" spoken tag |
| Virtual influencer partnership | Sponsorship + synthetic persona | "#ad" plus "virtual creator" in bio and caption |

## Hands-on: a synthetic media pre-flight checklist

```text
[ ] Does the asset depict a real, identifiable person, place or event? If yes: consent/rights + deepfake disclosure rules.
[ ] Could the asset mislead about the product's performance, size, results or availability?
[ ] Are any testimonials, reviews or "customers" synthetic? If yes: remove or clearly present as dramatization.
[ ] Commercial relationship disclosed (#ad / Paid partnership)?
[ ] Platform AI disclosure setting used where required?
[ ] Provenance metadata preserved in export?
[ ] Election, political or social-issue content? If yes: platform-specific rules and extra review.
[ ] Entry added to the synthetic asset register (tool, label, consent record)?
```

## Measuring success

Zero upheld ad complaints or platform takedowns for synthetic media; 100% of real-person likenesses backed by a consent record; checklist completion rate on campaign launches.

## Pitfalls

- Thinking a platform label covers your legal duty.
- Using AI "customer" faces in testimonials.
- Cloning a staff member's voice on a casual verbal OK.

## Video lecture: AI in advertising: disclosure, synthetic media and platform rules

Lecture coming soon · 14 chapters · about 9 minutes. Read the full transcript below.

1. AI in advertising
2. Why it matters
3. Four rulebooks
4. Analogy: burger food styling
5. Where AI ads go wrong
6. Disclosure layers
7. Platform rules (verify current)
8. Real people = consent
9. Worked example: skincare launch
10. Example 2: Karachi clothing brand
11. Common mistakes
12. Watch me do it: campaign pre-flight
13. Recap and next step
14. Try this now

## Lecture transcript

### AI in advertising

AI didn't change what counts as a misleading ad. It changed how easy it is to make one. With a few prompts you can create a flawless before and after photo, a glowing customer testimonial, or a celebrity-sounding voice-over. In this lesson you'll learn the four rulebooks that apply when you use AI in marketing, how disclosure layers work, what major platforms require, and a pre-flight checklist for every synthetic asset.

### Why it matters

Why does this matter so much in marketing? Because advertising is where AI's persuasive power meets the public at scale, and it's heavily policed by regulators, platforms and competitors. An ad complaint can get a campaign pulled within days. A platform can disable an ad account. And a deceptive synthetic testimonial can do lasting brand damage when it's exposed. The good news: the rules are well understood, and honest AI creative is both allowed and effective.

### Four rulebooks

Four rulebooks can apply at once. AI-specific law, like the EU AI Act's deepfake and chatbot duties. Advertising and consumer law, which everywhere prohibits misleading the average consumer about something material. Platform policies from Meta, YouTube, TikTok, Google and others, which are often stricter than the law and enforced faster. And the rights of real people: image and voice rights, data protection and defamation. Your campaign has to pass all four.

### Analogy: burger food styling

An analogy: using AI in ads is like using professional food styling in a burger commercial. Some styling is fine, everyone expects a nicely lit photo. But if the burger in the ad is twice the size of the one you get, that's misleading, whether it was done with glue, lighting tricks or an AI tool. Regulators don't care about the technique. They care about whether the average customer is misled about something that matters to their decision.

### Where AI ads go wrong

The timeless test is simple: is the ad likely to deceive the average consumer about something that matters to their decision? AI makes three failures easy. Exaggerated product imagery, like AI-enhanced skincare results. The UK's ASA has ruled against filters that exaggerate results, and AI edits follow the same logic. Synthetic testimonials, AI-generated customers praising a product. In the US, the FTC's fake reviews rule covers these explicitly. And virtual influencers, who must still disclose sponsorship and shouldn't pass off an invented experience as real.

### Disclosure layers

Disclosure works in layers, and each layer does a different job. Tool providers add machine-readable provenance, like Content Credentials metadata and invisible watermarks. You, the advertiser, add legal disclosures, like this video contains AI-generated imagery, where deepfake rules apply. Platforms add their own labels, often triggered by your self-disclosure. And advertising disclosure, like hashtag ad or paid partnership, covers the commercial relationship. These don't substitute for each other. A hashtag ad doesn't say it's AI, and an AI label doesn't say it's sponsored.

### Platform rules (verify current)

Platforms have their own rules, and they change often, so verify the current versions. YouTube requires creators to disclose realistic altered or synthetic content, like making a real person appear to say something they didn't. Meta labels AI content and requires advertisers to disclose digitally created or altered content in ads about social issues, elections or politics. Google Ads requires election advertisers to disclose synthetic content that realistically depicts people or events. TikTok requires labels on realistic AI-generated content and restricts misleading synthetic content of public figures. Keep a one-page sheet with links and a last-checked date.

### Real people = consent

Real people need consent. Before creating synthetic images, video or voice of anyone, get written consent covering purpose, media, territories, duration, revocation and payment. Public figures are higher risk, not lower: false endorsement claims and platform bans. Employees need special care, because a staff member may feel they can't say no. Offer a genuine choice and a way out. Laws like Tennessee's ELVIS Act, image rights in the UAE and Saudi Arabia, and SDAIA's deepfake guidance all point the same way.

### Worked example: skincare launch

Here's a skincare launch across the UK, UAE and US. The AI before and after images were scrapped and replaced with real, unedited trial photos with test conditions stated. The fictional AI presenter got an on-screen line and platform self-disclosure. The founder's cloned voice for radio got written consent and a spoken AI voice tag. And the virtual influencer partnership used hashtag ad plus virtual creator in the bio and caption. Nothing was banned. Everything was made honest.

### Example 2: Karachi clothing brand

A simpler example. A Karachi clothing brand uses AI to generate lifestyle backgrounds, like a beach at sunset, behind real product photos of its kurtas. The garments themselves are photographed as they are, colors unaltered. Is this misleading? Not about anything material: customers see the real product accurately. The brand keeps a note in its asset register, preserves the provenance metadata, and uses the platform's AI disclosure setting where the background looks realistic. Proportionate, honest and quick.

### Common mistakes

Three mistakes to avoid. Thinking a platform's AI label covers your legal duties, when it may not appear everywhere your ad runs and won't cure a misleading claim. Using AI-generated customer faces in testimonials, which is deception even with a label. And cloning a staff member's voice on a casual verbal OK, without a written, specific, revocable consent they genuinely felt free to refuse.

### Watch me do it: campaign pre-flight

Watch me do it. I open last month's skincare campaign folder and run each asset through the pre-flight checklist, logging results in the synthetic asset register. Asset one: an AI-enhanced before-and-after image. Question: could it mislead about results? Yes, the skin texture change is far beyond what the trial showed. Decision: replace with unedited trial photos and state the test conditions. Asset two: a fictional AI presenter in a product video. Real person? No. Misleading? Not about the product. Label: on-screen line AI-generated presenter, and the platform disclosure setting switched on. Asset three: three customer quotes on the landing page. I check the source: two are real reviews with permission, one was drafted by AI as a placeholder and never replaced. That one gets deleted today. Asset four: a sponsored post by a virtual creator. Sponsorship label: hashtag ad present. Synthetic persona disclosed: only in the bio, so I add virtual creator to the caption too. Four assets, two fixes, one deletion, and every row now has a label and a link to evidence.

### Recap and next step

Recap. AI doesn't change the misleading test. It makes failing it easier. Four rulebooks apply: AI law, advertising law, platform policies and people's rights. Disclosure layers work together but don't substitute for each other. And real people's likeness and voice need written consent. Your next step: copy the synthetic media pre-flight checklist from the lesson text into your campaign launch template, and run it on your next campaign.

### Try this now

Try this now, step by step. Open your last campaign and list every asset that used AI. For each one, ask four questions from the pre-flight checklist: does it depict a real person, place or event; could it mislead about the product; are any testimonials synthetic; and is the commercial relationship disclosed? Fix anything that fails. Then add each asset to your synthetic asset register with the tool, the label you used and any consent record. It should take under an hour, and it becomes your template for every future launch.

## Key takeaways

- AI does not change what is misleading; exaggerated imagery, synthetic testimonials and undisclosed virtual influencers are classic failures.
- Provenance metadata, legal AI disclosure, platform labels and sponsorship disclosure are separate layers that do not substitute for each other.
- Major platforms require disclosure of realistic synthetic content, with stricter rules for political, election and social-issue ads.
- Synthetic media of real people needs specific, written, revocable consent; employees need a genuine choice.

## Try it

Add the synthetic media pre-flight checklist to your campaign launch template and run it on one live or recent campaign. Log findings in your synthetic asset register.

- [Previous: Model and vendor due diligence](https://optimizeall.com/learn/ai-governance-eu-ai-act/model-and-vendor-due-diligence)
- [Next: Data protection and AI: GDPR and its cousins](https://optimizeall.com/learn/ai-governance-eu-ai-act/data-protection-and-ai-in-marketing)
- [All lessons of AI Governance & Regulation: EU AI Act, NIST AI RMF and ISO/IEC 42001](https://optimizeall.com/learn/ai-governance-eu-ai-act)