Free Resource
AI Ad Disclosure Requirements on Meta, TikTok and YouTube
What each platform's own documentation asks an advertiser to disclose about AI-made creative, who applies the label, which edits are exempt, and what a production scope should record.
By Unreal Media GroupEditorial standardsReviewed
Planning guide Written by the studio from its own practice; where it states a view, the view is the studio's. Figures quoted from a platform are attributed to that platform's documentation under Sources and should be verified at upload.
Fine Dining · 11s
Whether an ad made with generative AI has to say so is a question the platforms now answer in their own documentation, and the three answers differ. Meta labels ad images it can identify as AI-generated or significantly edited. Google gives advertisers a setting to designate assets as AI-generated or edited and shows the disclosure in its ad transparency panel. TikTok's advertising policy asks for its AIGC label or the advertiser's own disclaimer on AI-generated or significantly modified content, and says undisclosed AI content will be rejected or restricted. This guide reads those three documents side by side for the marketer briefing the work: what triggers a label, who applies it, what is exempt, and how Unreal records disclosure in a scope for AI UGC ads, disclosed as AI. It is a reading of platform policy as published on the review date, not legal advice. Platform rules change; the rule that applies to a campaign is the one in the platform's documentation at upload.
What's included
- Meta, Google and TikTok rules restated from their own documents
- The exempt edits each platform names
- Who applies the label on each platform
- Definitions a legal reviewer and a producer can share
- The studio's own disclosure and likeness policy
- A scope checklist for disclosure handover
Three platforms, three mechanisms
The three documents agree on the principle (a viewer should be able to find out that an ad was made or altered with AI) and differ on everything operational: what counts, who applies the label, and where it shows. The table restates each platform's own document; the Sources list links them.
| Platform | What triggers a label | Who applies it | Where it shows |
|---|---|---|---|
| Meta (Facebook and Instagram ads) | Ad images created or significantly edited with Meta's generative AI creative features, or created or edited with third-party AI tools. Minor changes such as resizing or colour correction do not qualify. | Meta, from what it can identify. Advertisers running ads about social issues, elections or politics must disclose AI-created or AI-edited image, video or audio themselves. | The AI info label on the About this ad screen (the three-dot menu), and sometimes beside the Sponsored label at the top of the ad. |
| Google (Search, YouTube and Discover ads) | Assets the advertiser designates as AI-generated or edited in the Ads AI label setting. Google may label assets itself where it is legally required to. | The advertiser, during campaign creation or when editing assets. | The How this ad was made section of My Ad Center. For campaigns targeting the European Union, India and New York, a visible overlay on the ad as well. |
| TikTok (TikTok ads) | Completely AI-generated images, video or audio; AI alterations that show a subject doing something they did not do; AI voice work that makes a subject say something they did not say. Lighting, brightness and colour adjustments, background removal or modification, and denoising need no disclosure. | The advertiser, by applying TikTok's AIGC label or adding a clear disclaimer, caption, watermark or sticker of its own. | On the ad itself. TikTok states that AI-generated content it identifies as undisclosed will have the ad rejected or restricted. |
What counts as significant, and what is exempt
Two of the three documents draw the line explicitly. Meta excludes minor changes and enhancements, naming resizing and colour correction. TikTok lists lighting, brightness and colour adjustments, background removal or modification, and image denoising as edits that need no disclosure, and completely generated media, fabricated actions and fabricated speech as edits that do. Google leaves the designation to the advertiser and does not publish an exemption list on the setting's help page. The practical line, which a scope can be written to, is whether a viewer would take a generated or altered element for a photograph or a recording of something real: a synthetic performer, a generated environment presented as a place, a product render presented as a product photograph, or a voice the listener would assume belongs to the person on screen. Where the answer is yes, plan to disclose on every platform, whatever the platform's own detection does.
- Synthetic performer
- A person in an ad who is generated rather than filmed, or a generated likeness. Unreal discloses every synthetic performer as synthetic and creates no likeness of a real person without a written rights agreement.
- Significant edit
- An AI alteration that changes what the viewer would believe happened: generated media, a subject shown doing or saying something they did not. Distinguished by Meta and TikTok from minor adjustments such as colour correction, resizing or background removal.
- Self-disclosure
- The advertiser declaring the AI element itself: Google's AI label setting, TikTok's AIGC label or the advertiser's own caption, watermark or sticker, and Meta's required declaration on political and social-issue ads.
- Automatic labelling
- A platform applying the label from what it detects in the file or from its own tools having been used, without the advertiser's declaration. Meta describes this for ad images; Google reserves it for regions where it is legally required.
- Content Credentials (C2PA)
- An industry standard for attaching provenance metadata to a media file, including whether it was generated with AI. Meta and TikTok have both said they read such signals to identify AI content; whether a file still carries them after editing depends on the tools in the chain.
Who applies the label matters more than the label
The operational difference between the platforms is who ticks the box. On Meta the label is applied by the platform from what it can identify, which means an advertiser cannot rely on it appearing and cannot rely on it not appearing; Meta has said publicly that it reads AI-generation signals in industry metadata standards attached by other companies' tools, and that such markers can be stripped or absent. On Google the designation is the advertiser's, made in the campaign or on the asset, so a media team that does not know an asset was generated will not designate it. On TikTok the disclosure is the advertiser's, and the policy attaches a consequence to getting it wrong. The consequence for a brand working with a production partner is that disclosure is a handover question: the studio knows which elements are generated, the media team owns the account settings, and the scope has to say which of them does what, on which account, before the first upload.
Where the rules bite hardest: people and voices
Every one of the three documents is strictest about people. TikTok's examples are a subject shown doing something they did not do and a subject made to say something they did not say. Meta's political and social-issue rule names photorealistic people and realistic-sounding audio. Google's setting covers audio assets as well as visual ones. This is the ground on which Unreal's own policy is built, independently of any platform: synthetic performers are disclosed as synthetic, and no real person's likeness or voice is created without a written rights agreement covering that use. The performers on the AI actors page and in the rights guides are documented on that basis. A brand that wants a real founder, customer or creator on screen books that person; a brand that wants a performer who does not exist gets one that says so.
How Unreal writes disclosure into a scope
Disclosure is a line in the written scope, not a note in a handover email. The scope records, for each deliverable: which elements are generated (performer, voice, environment, product render, music) and which are filmed or supplied; the label each platform account will carry and who applies it, named by role; whether on-creative disclosure text is also wanted, either because the platform's mechanism is the advertiser's own caption or because the brand prefers it visible on the piece; whether generator provenance metadata travels with the delivered file or is removed in the edit, and why; and which reviewer on the brand side signs the disclosure line before upload. The brand's legal or compliance reviewer sees that line at the same time as the creative. Where a platform's documentation changes between scope and upload, the scope names who checks it. This is the same approach the studio takes to every claim it does not control: state what was done, name who is responsible, and link the document the rule comes from.
What this guide does not tell you
It does not say what the law requires of a brand. Consumer-protection and advertising law, including endorsement and testimonial rules in the United States and the transparency obligations arriving under the European Union's AI regulation, is a separate question from platform policy and one for counsel, and it depends on the claim, the audience and the jurisdiction. It does not cover broadcaster clearance for television, which has its own process. And it does not promise that a label will or will not appear on any particular ad: two of the three platforms apply labels from their own detection, and their documentation is the authority on the day the ad runs.
Sources
- How AI-generated images in ads are identified and labeled on Meta, Meta Help Center.
- Labeling AI-generated images on Facebook, Instagram and Threads, Meta Newsroom.
- Use AI content label settings and disclosures, Google Ads Help.
- TikTok Advertising Policies: Misleading and false content, TikTok for Business.
- Partnering with our industry to advance AI transparency and literacy, TikTok Newsroom.
In practice
The service this guide supports: AI UGC ads, disclosed as AI.
More free resources
Frequently asked questions
Do AI-generated ads have to be labelled?
On the three platforms this guide covers, each platform's documentation describes a disclosure mechanism for AI-generated or significantly AI-edited ad content: Meta applies an AI info label to ad images it identifies, Google provides an advertiser setting that adds a disclosure to the ad's transparency panel, and TikTok's ad policy asks the advertiser to apply its AIGC label or a clear disclosure of its own. Minor edits such as colour correction are excluded by Meta and TikTok. Verify the current rule in the platform's documentation at upload.
Does Meta label AI ads automatically?
Meta's help documentation says it labels ad images created or significantly edited with its own generative AI features or with third-party AI tools, and that minor changes such as resizing or colour correction do not get the label. Advertisers running ads about social issues, elections or politics must disclose AI-created or AI-edited media themselves.
How do I disclose AI content in a TikTok ad?
TikTok's advertising policy says to apply its AIGC label or to add a clear disclaimer, caption, watermark or sticker of your own to completely AI-generated media or to AI alterations that show a subject doing or saying something they did not. It also states that undisclosed AI content it identifies will have the ad rejected or restricted.
Does YouTube label ads made with AI?
Google's Ads AI label setting lets an advertiser designate assets as AI-generated or edited; the ad then carries a disclosure in the How this ad was made section of My Ad Center, and for campaigns targeting the European Union, India and New York a visible overlay as well. Google says it may label assets itself where it is legally required to.
Does a colour-corrected or background-removed image need a label?
Not under the two documents that list exemptions: Meta names resizing and colour correction as minor changes that do not get the label, and TikTok names lighting, brightness and colour adjustments, background removal or modification, and denoising as edits that need no disclosure. A generated background, as opposed to a removed one, is a different case.
Is it allowed to use an AI actor in an ad?
The platform documents this guide covers describe how AI-made content in ads is to be disclosed rather than prohibiting it; the strictest rules concern real people shown doing or saying things they did not. Unreal discloses every synthetic performer as synthetic and creates no likeness of a real person without a written rights agreement. Whether a particular claim or use is lawful is a question for counsel.
Start with one scoped batch.
Book a free 15-minute discovery call. We'll map the exact plan for your brand — no pitch deck, no fluff.