AI-generated content disclosure rules in 2026: what the EU AI Act, Meta and the FTC ask of marketers

AI-generated content disclosure rules in 2026, side by side

AI-generated content disclosure rules for marketers come from three places in 2026: EU law, the platforms, and US advertising law. Each one tests something different. The EU's AI Act asks whether synthetic content could pass as real. Meta asks which tool made an ad and whether it is political. The FTC asks whether an ad or an endorsement misleads, whatever made it. A single "made with AI" tag does not satisfy all three, and some posts need none of them.

The rest of this guide takes each rule in turn, then sets out a checklist a team can run before anything publishes. It is a reading of the primary sources linked in each section, not legal advice.

What EU AI Act Article 50 requires of marketers

EU AI Act Article 50 requires the business that uses an AI tool, which the Act calls the deployer, to disclose deepfakes and some AI-written text. The tool's maker, the provider, has separate duties: under Article 50(2) it must mark its outputs in a machine-readable way so they can be detected as AI-made. Article 50 has applied since August 2, 2026, according to the Commission's Article 50 FAQ, last updated on July 24, 2026.

The Commission's FAQ settles three questions marketers ask first:

• An advertising company that uses an AI system is the deployer. Its designers, animators and content creators acting under its instructions are not separate deployers, and it stays the deployer when freelancers run the tool on its behalf.
• A person using AI for personal, non-professional purposes is outside the Act. A creator who earns from the activity on a regular basis is acting professionally and counts as a deployer.
• The provider's machine-readable mark does not satisfy the deployer's duty. The disclosure must be perceivable by people, for example a visible or audible label, without special tools or dedicated actions.

Article 50(5) sets the form: the information must be clear and distinguishable, reach people at the latest at their first exposure, and meet accessibility requirements. The Commission published a voluntary Code of Practice on Transparency of AI-generated Content on June 10, 2026, with a section of rules for deployers labelling deepfakes and text. The Commission and the AI Board assessed it as an adequate way to show compliance, and the Commission reports that about 190 organisations had signed it by the end of July 2026. Deployers may use an EU set of icons as the label.

When an AI ad image counts as a deepfake

An AI ad image counts as a deepfake under the AI Act when it meets three conditions together: it closely resembles something, that something exists or plausibly could, and the image would falsely appear authentic to the person who sees it. Article 3(60) of the Act lists persons, objects, places, entities and events, so the definition reaches a product, a shop front or a city street, not only a face.

The third condition depends on context. The Commission's FAQ says the assessment can weigh the level of resemblance, the message, where the content is shown and what its intended audience expects. Its own example is background scenes and special effects in film production, which are not likely to make content falsely appear authentic. Read with ad formats in mind, the guidance sorts roughly like this:

• Likely in scope: a photorealistic "customer photo" of a person who does not exist, a realistic AI voice of a real person, or a product shown in a real-looking setting where viewers would take the picture as a photograph.
• Less likely: a stylised illustration, an obviously impossible scene, or an edit that only tidies a real photo. Article 50(2) also exempts AI that performs an assistive function for standard editing from the provider's marking duty.
• Lighter duty: content that is part of an evidently artistic, creative, satirical or fictional work needs disclosure only in a way that does not spoil the work. The Act does not say that ordinary advertising qualifies.

Does human review remove the EU label duty?

Human review removes the EU label duty for AI-written text only. The exemption in Article 50(4) covers text published to inform the public on matters of public interest that has undergone human review or editorial control, where a person or company holds editorial responsibility. There is no matching exemption for deepfake images, audio or video, so approving a realistic AI image does not remove the duty to label it.

The Commission's FAQ narrows both halves of the text rule:

• Matters of public interest include politics, public health, consumer safety, environmental protection, and economic, financial or scientific developments that may be the subject of public debate. Most product captions are not written to inform the public on such matters; a post about a product recall or a health claim may be.
• Human review means a person with relevant knowledge examines the substance of the content. Spell-checking, grammar fixes and other purely formal or procedural checks do not count.
• Editorial responsibility means a person holds the ultimate legal responsibility for the publication.

For a marketing team, the practical rule is simple. An approval step with a named reviewer who reads for substance covers AI-written text. It does nothing for a realistic AI picture, which still needs a visible label.

What Meta's AI info label means and when it applies

Meta's AI info label tells viewers that an ad image was created or materially edited with Meta's own generative AI creative features, such as background generation and image generation. Meta adds it, not the advertiser. According to Meta's ads help page, the label sits in the ad's three-dot menu unless the image includes an AI-generated photorealistic human, in which case it appears next to the Sponsored label. Meta says the feature is being introduced gradually and that not every qualifying ad will carry it.

Meta's other rules split by the kind of content:

• Political and social issue ads: under Meta's policy for ads about social issues, elections or politics, advertisers must disclose photorealistic images or video, or realistic audio, made or edited with third-party AI tools to show people or events that are not real. Meta rejects ads that should have disclosed. The same page says such ads are not allowed to run in the EU, apart from voting information placed by election authorities.
• Organic posts: Meta said in February 2024 that it requires people to use its disclosure tool when they post photorealistic video or realistic audio that was digitally created or altered, and that it may apply penalties when they do not. Its April 2024 update says labels on content that AI only edited moved to the post's menu from September 2024.

Meta's pages disagree on third-party tools. The ads help page lists ads created or modified with third-party AI tools among those that will not have AI info. The political ads policy says that, from June 1, 2026, Meta also uses automated detection to find ad media made or edited with third-party generative AI tools and adds AI info in "About this ad". Either way, an EU deployer cannot treat a label inside a menu as its own disclosure, because the Commission says a deepfake label must be perceivable without any dedicated action.

What the FTC expects from AI influencers and AI testimonials

The FTC expects AI-made endorsements to follow the same rules as human ones, and none of the FTC pages read for this guide requires a label saying content is AI-generated. The test is whether the ad misleads. The Endorsement Guides in 16 CFR Part 255 define the endorser as the party whose views the message appears to reflect, who could be or appear to be an individual. On that wording, a virtual influencer's post for a brand is an endorsement.

Three parts of the FTC's rules bite hardest on AI content:

• Section 255.1 says endorsements must reflect the honest opinions, findings, beliefs or experience of the endorser. When an ad says the endorser uses the product, the endorser must have been a bona fide user of it. An AI persona that claims to use a product sits badly with that rule.
• The fake reviews rule in 16 CFR Part 465 bans businesses from writing, creating or selling reviews or testimonials that misrepresent that the reviewer exists or used the product. The FTC's August 14, 2024 announcement names AI-generated fake reviews as an example and says the rule lets it seek civil penalties against knowing violators. The rule took effect 60 days after its August 22, 2024 publication in the Federal Register, on October 21, 2024.
• Section 255.0(f) defines a clear and conspicuous disclosure as one that is difficult to miss and easily understood. A claim made in pictures needs the disclosure at least in the pictures, and a claim made out loud needs it at least in the audio.

Is #ad enough for an AI-generated post?

#ad is not enough for an AI-generated post that needs an AI label, because #ad answers a different question. #ad tells viewers that the poster has a paid or other material connection to the brand, which is what the FTC's Endorsement Guides require. An AI label tells viewers that the picture, voice or person is synthetic, which is what the AI Act, Meta and YouTube ask about. A sponsored post that shows a photorealistic AI person to an EU audience can need both.

The FTC's Endorsement Guides FAQ also warns against leaning on platform tools. It says a platform's built-in paid partnership feature is no guarantee of an effective disclosure, that the responsibility rests with the influencer and the brand rather than the platform, and that it is safest to add your own disclosure as well. It recommends putting the disclosure in the video itself rather than only in the description, and says a picture may need a disclosure superimposed over it. The same placement logic suits an AI label: on the image or in the video, not only in a caption a viewer may never expand.

YouTube's disclosure page shows how a platform label works in practice. Creators select Yes under AI use when they upload realistic AI content, and YouTube adds a label. YouTube may also apply the label itself to content made with its own AI tools, content carrying C2PA metadata, or content its systems detect as AI-made. A creator who consistently fails to disclose may lose content or be suspended from the YouTube Partner Program.

A pre-publish checklist for AI-generated ads and posts

A pre-publish checklist for AI-generated content works best as one step in an approval flow, run by the person who signs the post off. The order below goes from the questions that decide whether a label is needed to the ones that decide where it goes:

1. Note which parts of the post an AI tool made: the text, the image, the video, the voice, the music.
2. Ask whether any image, audio or video would pass as a real photograph or recording of a real person, product, place or event. If yes and the post reaches the EU, it needs a visible label.
3. Check whether the post shows a person who does not exist presented as a customer, reviewer or user. If yes, rewrite it: under the FTC's rules, that testimonial misrepresents that its author exists.
4. Decide whether AI-written text informs the public on a matter of public interest, such as health, safety or finance. If yes, make sure a person reviews its substance and the team holds editorial responsibility, or label it.
5. Check for a paid or other material connection to the brand. If there is one, add #ad or a similar disclosure, separate from any AI label.
6. Put each label where it is seen at first exposure: on the image, in the video or in the first line of the caption, not only behind a menu or in a description.
7. Apply the platform's own setting as well: Meta's AI disclosure tool for realistic video or audio, the political ads disclosure where it applies, and YouTube's AI use field.
8. Record who approved the post and when, so the editorial responsibility for text is on file.

Where overads fits

overads is AI marketing automation for small teams: its assistant drafts posts and makes images and short videos in chat, and what the assistant and workflows draft waits for a person's approval by default. That gives the checklist above a fixed place to run. overads has two approval controls, described in its scheduling and approvals guide:

• A per-post Require approval switch in the composer, for the one post that needs a second pair of eyes.
• A workspace rule, turned on by an owner or admin in Settings, Rules, Require approval. While it is on, every post waits as pending until a team member approves it, including one sent with Publish now, whether a person, the assistant or a workflow made it. The rule is off until someone turns it on.

A post a workflow writes waits for approval by default, as the workflows guide explains. When a reviewer approves a post, overads sends a post.approved webhook that names who approved it, which a team can keep as its record. overads Publish posts to Instagram, Facebook, LinkedIn, X, Bluesky, Mastodon and Pinterest.

What overads does not do:

• It does not put a visible AI label on the images and videos it makes. The team adds the disclosure, on the image or in the caption.
• It does not set Meta's AI disclosure, YouTube's AI use field or any other platform's AI label when it publishes.
• It does not check a post for missing disclosures or tell a team whether a post needs one.
• It does not publish to YouTube or TikTok today; both are built and in platform review.
• It does not give legal advice.

A team that publishes mostly to YouTube, or that needs a label applied automatically on every AI image, is better served by the platform's own upload settings or an asset tool that stamps a visible mark. An agency running client approvals across several brands can compare options in our guide to social media approval workflow tools, and the overads page for agencies shows how each client gets its own workspace. Our guide to human-in-the-loop AI for marketing covers where the approval step belongs.

What this guide does not cover

This guide covers EU AI Act Article 50, Meta's and YouTube's published rules, and two FTC rules, as read on October 19, 2026. It does not cover US state laws on AI in political ads, rules in other countries, TikTok's, LinkedIn's or X's policies, or sector rules for finance and health advertising. It is not legal advice: a team with ads in regulated categories, or doubts about a specific campaign, should ask a lawyer who practises in the markets the ads run in.