---
title: "Disclosing synthetic media and following platform policies"
description: "Why disclosure matters Synthetic voices and avatars are now realistic enough to fool people. That creates real harm: scams using cloned voices, fake…"
url: https://optimizeall.com/learn/ai-video-and-voice-production/disclosure-and-platform-policies
updated: 2026-10-05
---

AI Video & Voice Production: ElevenLabs, Veo, Runway and More · Quality, accessibility and disclosure · lesson 14 of 16 · 10 min

# Disclosing synthetic media and following platform policies

## Why disclosure matters

Synthetic voices and avatars are now realistic enough to fool people. That creates real harm: scams using cloned voices, fake endorsements, political deepfakes and loss of trust in genuine content. Platforms, regulators and audiences increasingly expect clear labeling of AI-generated or altered media, especially when it is realistic.

For creators and brands, good disclosure is also a trust asset. Being upfront ("made with my AI voice") rarely hurts performance as much as being caught hiding it.

## What platforms generally require

Policies change often, so always check the current help center of each platform. As of 2026, the broad direction across major platforms is consistent:

- **YouTube** asks creators to disclose content that is realistic and meaningfully altered or synthetic, such as making a real person appear to say or do something they did not, or realistic depictions of events that did not happen, using its disclosure setting at upload. Labels may appear on the video. Obviously unrealistic, animated or minor edits, such as color correction or beauty filters, generally do not need disclosure.
- **Meta (Facebook, Instagram, Threads)** applies "AI info" style labels, using industry signals and creator self-disclosure, and expects disclosure of realistic AI-generated or altered video and audio.
- **TikTok** requires creators to label AI-generated content that shows realistic scenes, and can auto-label content carrying content-credential metadata. It prohibits certain harmful synthetic content, such as fake endorsements or misleading depictions of real people in some contexts.
- **Ad platforms** (Google, Meta, TikTok) have additional rules for ads, especially political and social-issue ads, which in many regions must disclose synthetic content, and they prohibit deceptive impersonation.

## Content credentials

Many tools and platforms support **content credentials** based on the C2PA standard: tamper-evident metadata recording how content was created and edited, including AI involvement. Some AI tools add credentials or invisible watermarks automatically, and platforms may read them to apply labels. Do not strip this metadata to avoid labels; that undermines trust and may breach platform rules.

## Legal direction

Laws are evolving quickly. At principle level, as of September 2026:

- The **EU AI Act's transparency obligations (Article 50)** apply from **2 August 2026**. Providers of generative AI systems must mark synthetic audio, image, video and text outputs in a machine-readable, detectable way, and deployers who publish **deepfakes** (realistic content depicting real people, places or events) must disclose that the content is artificially generated or manipulated, with lighter rules for evidently artistic, satirical or fictional work. A Commission **Code of Practice on transparency of AI-generated content** and guidelines support implementation. A provisional political agreement on an "AI omnibus" in May 2026 proposed extra time, to December 2026, for some machine-readable marking duties for systems already on the market; check the final adopted text and guidance for your situation.
- **Advertising regulators** such as the UK ASA/CAP, the US FTC and Gulf media regulators apply existing rules: ads must not mislead. A synthetic presenter implying a real customer, expert or celebrity endorsement is misleading regardless of AI-specific laws.
- **Influencer and sponsorship disclosure** rules still apply independently: #ad, paid partnership labels and local regulator requirements (for example the UAE Media Council's advertiser permit requirements for influencers) are separate from AI labels. You may need both.
- **Election and political content** faces the strictest rules in many countries and on every major platform.

## A practical disclosure policy

Label, or use the platform's disclosure tool, when:

1. A real, identifiable person appears to say or do something they did not actually say or do, even if it is yourself via your avatar or clone, and especially in a new language.
2. Realistic footage depicts events, places or products in a way that did not happen.
3. The platform's policy requires it.
4. A reasonable viewer would feel misled if they later learned it was AI.

Labeling is usually unnecessary (though harmless) for obviously animated or stylized content, minor AI edits (noise removal, color grading) and AI-assisted scripting where a real human presents.

**Wording examples:**
- "This video uses an AI version of my voice. Script written and checked by me."
- "Presented by our AI avatar. Information reviewed by our team."
- "AI-dubbed from English with the presenter's permission."
- "Some scenes in this video are AI-generated."

## Worked example

A Pakistani creator with a large YouTube following uses her HeyGen digital twin and ElevenLabs voice clone for a weekly news-roundup Short in Urdu and English. Her routine: she toggles the platform's altered or synthetic content disclosure, adds "Made with my AI avatar" to the description, keeps a pinned FAQ explaining she never asks for money via DMs or voice notes, and records genuine videos for sponsored segments, which also carry #ad and the paid partnership label.

## Content credentials and watermarks in practice

Two technologies travel with AI media:

- **C2PA content credentials**: signed metadata describing how a file was created and edited, including AI involvement. Some generators and editors attach them; some platforms read them to apply labels automatically.
- **Invisible watermarks**: for example Google's **SynthID**, which Google applies to content generated by its models such as Veo. These survive some editing and can be detected by compatible tools.

Workflow rules: keep credentials and watermarks intact, do not re-encode files through tools that strip metadata unless unavoidable, and when a platform labels your video automatically, do not try to remove the label; add your own clear description line as well.

## Hands-on: disclosure decision tree

```text
1. Does a real, identifiable person appear to say or do something they did not? -> LABEL (platform tool + description)
2. Does realistic footage depict an event, place or product result that did not happen? -> LABEL, or cut it
3. Is it a paid promotion? -> Sponsorship disclosure (#ad / paid partnership) IN ADDITION to any AI label
4. Is it political, electoral or social-issue content? -> Follow platform political-ad rules; stricter disclosure
5. Obviously animated/stylized, or AI only used for cleanup/scripting? -> Label optional; keep production record
Always: keep consent and production records; never strip C2PA/SynthID.
```

## Second worked example: an EU-facing campaign

A Dubai agency runs a campaign for a German client featuring a realistic AI-generated "customer" in a café. Under the client's policy and the EU transparency rules for deepfake-style content, the agency replaces the fake customer with a clearly stylized animated character, adds "Created with AI" to the description, applies platform AI labels, and keeps the real customer testimonial in a separate, genuinely filmed clip. The campaign still performs, and legal signs off in one round.

## Pitfalls

- Relying on a hashtag only when the platform provides a dedicated disclosure tool.
- Assuming AI labels replace sponsorship disclosures.
- Using an avatar to deliver a "personal" testimonial that you never gave.

## Video lecture: Disclosing synthetic media and following platform policies

Lecture coming soon · 15 chapters · about 9 minutes. Read the full transcript below.

1. Disclosure and platform policies
2. Why it matters
3. The nutrition label idea
4. Platform direction
5. EU AI Act, Article 50
6. Beyond the EU
7. Credentials and watermarks
8. Disclosure decision tree
9. Example 1: a creator's weekly Short
10. Example 2: an EU-facing campaign
11. Watch me do it
12. Making it operational
13. Common mistakes
14. Recap
15. Try this now

## Lecture transcript

### Disclosure and platform policies

Imagine you find out that a video you trusted, a founder's heartfelt message, was actually an AI avatar reading a script someone else wrote. How would you feel about that brand? Now imagine the same video opened with a small line: presented by our AI avatar, script written and checked by our team. Different feeling, right? In this lecture you'll learn what platforms require, where the law is heading as of September twenty twenty-six, how content credentials and watermarks work, and a simple decision tree for when and how to disclose.

### Why it matters

Why does this matter so much now? Because synthetic voices and avatars are realistic enough to fool people. That creates real harm: scams with cloned voices, fake endorsements, political deepfakes, and a general loss of trust in genuine content. Platforms, regulators and audiences increasingly expect clear labeling of AI-generated or altered media, especially when it's realistic. And for creators and brands, disclosure is also a trust asset. Being upfront rarely hurts performance as much as being caught hiding it.

### The nutrition label idea

Here's an analogy. Disclosure works like the nutrition label on food. Nobody thinks a label ruins the snack. It lets people make an informed choice, and brands that hide ingredients lose trust when someone finds out. There are really two kinds of labels in video. The AI label says how the content was made. The sponsorship label says who paid for it. They're separate obligations, and you may need both on the same post. Mixing them up is one of the most common mistakes I see.

### Platform direction

What do platforms require? Always check each help center, because policies change. But the direction is consistent. YouTube asks creators to disclose realistic content that's meaningfully altered or synthetic, like making a real person appear to say something they didn't, using a setting at upload. Obviously unrealistic, animated, or minor edits like color correction generally don't need it. Meta applies AI info style labels using industry signals and self-disclosure. TikTok requires labels on realistic AI-generated content and can auto-label files carrying content credentials. And ad platforms add stricter rules, especially for political ads.

### EU AI Act, Article 50

Now the law, as of September twenty twenty-six. The EU AI Act's transparency obligations, in Article fifty, apply from the second of August twenty twenty-six. Providers of generative AI must mark synthetic outputs in a machine-readable way. And people who publish deepfakes, meaning realistic content depicting real people, places or events, must disclose that it's artificial, with lighter rules for clearly artistic or satirical work. A provisional agreement in May proposed some extra time for certain marking duties for systems already on the market. Check the final guidance for your case.

### Beyond the EU

And beyond the EU, existing advertising law already does a lot of work. The UK's advertising regulator, the US Federal Trade Commission, and Gulf media regulators all apply the same principle: ads must not mislead. A synthetic presenter implying a real customer, expert or celebrity endorsement is misleading, AI law or not. Influencer disclosure rules, like hashtag ad, paid partnership labels, and permit requirements such as the UAE's advertiser permit for influencers, apply independently. And election content faces the strictest rules almost everywhere.

### Credentials and watermarks

Two technologies travel with AI media, and you should know both. Content credentials, based on the C2PA standard, are signed metadata describing how a file was made and edited, including AI involvement. Some generators and editors attach them, and some platforms read them to apply labels. Invisible watermarks, such as Google's SynthID, are embedded in content generated by models like Veo, and can survive some editing. The workflow rule is simple: keep them intact. Don't strip metadata to dodge a label. That undermines trust and can break platform rules.

### Disclosure decision tree

Here's the decision tree from the lesson. One: does a real, identifiable person appear to say or do something they didn't? Label it, with the platform tool and a description line. Two: does realistic footage show an event, place or product result that didn't happen? Label it, or cut it. Three: is it paid? Add sponsorship disclosure too. Four: is it political or social-issue content? Follow the stricter rules. Five: is it obviously stylized, or was AI only used for cleanup or scripting? Labeling is optional, but keep your production record.

### Example 1: a creator's weekly Short

First example, a simple one. A Pakistani creator uses her own digital twin and her own voice clone for a weekly news roundup Short in Urdu and English. Her routine: toggle the platform's altered or synthetic content setting, add made with my AI avatar to the description, and keep a pinned FAQ saying she never asks for money by DM or voice note. For sponsored segments, she films herself for real, and those carry hashtag ad and the paid partnership label. AI label for how it's made, sponsorship label for who paid.

### Example 2: an EU-facing campaign

Second example, a business case. A Dubai agency runs a campaign for a German client featuring a realistic AI-generated customer in a café. Because the content targets EU audiences and could read as a real person, the agency replaces the fake customer with a clearly stylized animated character, adds created with AI to the description, applies the platform labels, and uses a genuinely filmed customer testimonial in a separate clip. The campaign still performs, and legal signs off in one round instead of three.

### Watch me do it

Watch me write a disclosure policy for a small brand. I open a blank page with three sections. When we label: I copy the five questions from the decision tree. How we label: platform tool first, then a standard description line. And our wording library. Four lines. This video uses an AI version of my voice, script written and checked by me. Presented by our AI avatar, information reviewed by our team. AI-dubbed from English with the presenter's permission. Some scenes in this video are AI-generated. Short, specific, honest.

### Making it operational

Then I add two operational rules. First, the publish gate in our tracker stays red until the labels field is filled. Second, we never re-export through tools that strip content credentials unless we have to, and if a platform labels us automatically, we leave it and add our own line. Finally, I set a reminder to review the policy every quarter, because platform rules and regulations keep moving. A policy nobody reviews is just a document.

### Common mistakes

Common mistakes. Relying on a hashtag when the platform provides a dedicated disclosure tool. Assuming AI labels replace sponsorship disclosures. Using an avatar to deliver a personal testimonial you never gave. Stripping metadata to avoid an automatic label. And treating obviously stylized content the same as realistic deepfakes, which leads to either over-labeling everything or labeling nothing.

### Recap

Recap. Disclose realistic synthetic or altered media, especially when a real person appears to say or do something new. Use platform tools plus a clear description line. Remember AI labels and sponsorship labels are separate. The EU's transparency rules apply from August twenty twenty-six, and advertising law already bans misleading endorsements. Keep C2PA credentials and SynthID watermarks intact. And run a simple decision tree before every publish.

### Try this now

Try this now. Write your own synthetic-media disclosure policy on one page: the five decision questions, how you label on each platform you use, and three standard wording lines. Then apply it to your next AI-assisted video, and add a labels field to your production tracker so nothing publishes without it.

## Key takeaways

- Disclose realistic synthetic or altered media, especially when a real person appears to say or do something new.
- Major platforms provide disclosure tools and labels; use them and check policies regularly.
- Content credentials (C2PA) record AI involvement; do not strip them.
- AI labels and sponsorship disclosures are separate obligations, and you may need both.

## Try it

Write your personal synthetic-media disclosure policy with three standard label wordings, and apply it to your next AI-assisted video.

- [Previous: Captions, accessibility and quality control](https://optimizeall.com/learn/ai-video-and-voice-production/captions-accessibility-and-qc)
- [Next: Playbook: producing course and training videos with AI narration](https://optimizeall.com/learn/ai-video-and-voice-production/course-video-production-playbook)
- [All lessons of AI Video & Voice Production: ElevenLabs, Veo, Runway and More](https://optimizeall.com/learn/ai-video-and-voice-production)