Do You Have to Disclose AI Ads? US, EU and Network Rules (October 2026)
There's no single "AI ad" law. Disclosure risk mostly comes from the FTC's existing deception rules and each network's content policy, plus extra rules for synthetic depictions of real people and, in the EU, AI-generated content.

None of the FTC sources reviewed for this article requires a "made with AI" label on every ad. But several rules now turn on AI specifically, and which ones apply depends on where the ad is shown, who or what it depicts and which network carries it. As of 7 October 2026: New York has required a conspicuous disclosure of "synthetic performers" in advertisements since 9 June 2026; the EU AI Act's transparency rules (Article 50) have applied since 2 August 2026 and require disclosure of AI-generated "deep fakes"; some native networks, including Outbrain/Teads and Realize (Taboola), ask advertisers to label or declare AI-generated content; and the FTC's existing rules on deceptive advertising, endorsements and fake reviews apply whether or not AI produced the ad. This article summarizes those sources and names the date of each. It is not legal advice, and a qualified lawyer should review any campaign that uses realistic synthetic people, health or finance claims, or that runs in the EU or New York.
FTC rules on deception, endorsements and fake reviews apply to AI-made ads#
Section 5 of the FTC Act declares "unfair or deceptive acts or practices in or affecting commerce" unlawful (15 U.S.C. 45(a)(1)). In judging whether an ad is deceptive, the FTC looks at it from the point of view of a reasonable consumer and at its "words, phrases, and pictures" in context, and it expects an advertiser to have a "reasonable basis" for its claims before the ad runs (FTC Advertising FAQs). In a press release of 25 September 2024, the FTC's then-Chair said its enforcement actions "make clear that there is no AI exemption from the laws on the books". On 22 December 2025, when it set aside its order against the AI writing service Rytr, the FTC said it "will continue to hold accountable actors that use AI to violate the law or deceive consumers about the capabilities of their generative AI". A fabricated "clinical study" result in an ad can be a deceptive claim whether a language model or a copywriter produced it.
The FTC's most direct statement about AI in advertising content concerns fake reviews and testimonials. Its Rule on the Use of Consumer Reviews and Testimonials (16 CFR Part 465, in effect since 21 October 2024) makes it a violation for a business to write, create or sell a consumer review, consumer testimonial or celebrity testimonial that materially misrepresents that the reviewer or testimonialist exists, used the product or had the experience described. The rule's text does not use the word AI, but the FTC's announcement of 14 August 2024 gave "AI-generated fake reviews" as an example of a review by someone who does not exist.
FTC staff guidance says the rule has no blanket ban on AI-generated avatars in marketing, that a testimonial from an avatar is prohibited only if the underlying testimonial is fake or false, that it would violate the rule to use a celebrity avatar without the celebrity's permission to speak favorably about a product if reasonable consumers would think the celebrity gave the testimonial, and that avatar use can also be deceptive under the FTC Act. The FTC set aside its 2024 consent order against Rytr on 22 December 2025, so do not assume that sellers of AI writing tools are covered.
If your native creative or advertorial leans on a "real customer" quote that a model wrote from scratch, that is the rule to check, together with the FTC's Endorsement Guides (16 CFR Part 255), which say that ads presenting endorsements by what are represented to be "actual consumers" should use actual consumers or clearly and conspicuously disclose that the people shown are not actual consumers (section 255.2(c)). For how the FTC treats advertorials and native formats, see FTC disclosure rules for advertorials and native ads; the FTC's own advertising guidance is the primary source.
When "we used AI" needs its own disclosure, and when it doesn't#
Different things get lumped together under "AI ad disclosure," and they carry different risk levels.
Using a model to resize a creative, adjust color or draft routine ad copy is a production-method choice, and the EU and FTC sources reviewed here do not require a separate AI label for routine edits like these. The European Commission's guidelines on Article 50 of the EU AI Act say that AI-powered color correction, re-scaling and adjustments or replacements of backgrounds for clearly aesthetic purposes are likely to have only a minor impact on how authentic an advertisement and its product appear. They also give AI-manipulated text that is part of a company's advertisement or product descriptions as an example of text outside the AI Act's disclosure duty for public-interest text; the example expressly leaves out claims about, for example, health, consumer safety or sustainability, which may fall inside that duty. Networks differ. Realize does not ask for a declaration for minor enhancements such as upscaling, sharpening or color correction, but does for AI background replacement and generative fill that alters key visual elements, and for AI-generated text on health, news, elections, finance or other public-interest topics. Outbrain/Teads goes further: its guidelines say all advertising content generated or materially modified using AI, including text, must be labeled as AI-generated at the point of first exposure.
So whether a model-drafted headline needs a label depends on the network, the topic and how much the model changed. In every case, the claims in the copy must be true and substantiated.
Showing a person is a different category, and it covers two cases. The first is a depiction of a real, identifiable person: a celebrity endorsement that never happened, or a synthetic voice built to sound like a specific public figure. The FTC reviews rule can apply to a celebrity avatar used without permission, as described above, and right-of-publicity, false-endorsement and platform impersonation rules can also apply. Realize does not permit AI likenesses or voice clones of real people even with disclosure, and Outbrain/Teads permits them only with the person's explicit consent and a disclosure that the person shown or heard is not genuine. Do not treat a "not an actual endorsement" label as a substitute for consent or a license, and take legal advice.
The second case is a realistic person who does not exist, such as a synthetic "customer", "doctor" or presenter. New York General Business Law section 396-b has required a conspicuous disclosure when an advertisement includes a "synthetic performer" since 9 June 2026, where the person who produces or creates the advertisement has actual knowledge. Civil penalties are $1,000 for a first violation and $5,000 for any subsequent violation, and there are exemptions, including audio advertisements and advertisements for expressive works. In the EU, the Commission's Article 50 guidelines say that realistic AI-generated human avatars or personas can be deep fakes when they would falsely appear authentic, and they list a realistic synthetic influencer testing a sponsored product as an example that does not qualify for the lighter disclosure regime for creative works. Presenting a synthetic person as a real customer or doctor also raises the FTC issues described above (16 CFR 255.2(c) and 255.3(a)). This article does not review US state laws on synthetic media in political advertising or other state laws on a person's likeness; they differ by state.
What native ad networks' policies say about AI content (read 7 October 2026)#
Some native networks now have AI-specific rules, and others rely on their general creative policies. The table below summarizes the policy pages read on 7 October 2026. These are the networks' own words, summarized, so confirm the current text before you launch.
| Network | What its public policy pages say (read 7 October 2026) |
|---|---|
| Taboola (Realize) | An "AI content" checkbox on every creative (help article dated 20 July 2026): declare images, video, audio or text created from scratch by generative AI or substantially edited by AI, including AI background replacement and AI translation or dubbing, and AI-generated text on health, news, elections, finance or other public-interest topics. Minor enhancements such as upscaling or color correction do not need declaring. Realize shows an "AI" label in markets that require one, such as the EU and New York, and says disclosure is the advertiser's responsibility. AI likenesses or voice clones of real, identifiable people are not permitted, even with disclosure. Thumbnails must not show before/after photos or feature celebrities or politicians for a product without documented permission (policy page updated 31 August 2026). |
| Outbrain / Teads | Advertiser guidelines (undated, copyright 2026): content generated or materially modified using AI, including text, images, audio and video, must be clearly labeled as AI-generated at the point of first exposure. Synthetic likenesses and voice clones of real people are prohibited unless the advertiser holds explicit consent and discloses that the person is not genuine. AI must not be used to fabricate testimonials or reviews, make unsubstantiated claims or produce synthetic before-and-after imagery, and embedded provenance markers must not be removed. AI-generated political ads must be labeled. Thumbnails cannot include before/after photos. |
| MGID | AI-generated content is listed as restricted: it must not include misleading, deceptive, absurd or unrealistic representations, and deepfakes and synthetic media that imitate real people or events are not allowed. Abusive use of AI-generated content to create deceptive or fabricated scenarios is a prohibited practice. A help article titled "AI Act Labeling Requirements" is listed, but its link was a placeholder when checked. |
| Revcontent | No AI-specific rule found on the help-center policy pages read. General rules apply: images must not be "overly edited or heavily photoshopped", fake or unverifiable testimonials and celebrity endorsements are not allowed, and before-and-after images are barred in several categories. Digitally altered political content must carry a disclosure such as "This ad does not depict real events". |
| MediaGo | No public ad-content policy page was found, so its rules on AI-generated creative are not verified. |
| Microsoft Advertising | The information integrity policy (page updated 8 April 2026) prohibits untruthful or deceptive ads, including "ads, sites or ad landing pages using deepfake technology", fake reviews or testimonials, and manipulated imagery that makes ad information untrue. No separate AI-labeling rule was found on the policy pages read. MSN placements can also be bought through other networks, so check the buying network's policy. |
| Yahoo | The AI/Synthetic Media policy (undated) says ads that use AI or synthetic media must comply with applicable laws and disclosure requirements, "including but not limited to the EU Artificial Intelligence Act (Article 50)". Advertisers must ensure creatives carry legally required AI watermarks, machine-readable metadata and visible AI labels before upload, and deepfakes are prohibited unless clearly and conspicuously disclosed. |
Several of these pages show no revision date and the wording above is summarized, so treat the table as a starting point and confirm specifics with each network before you run anything that uses realistic synthetic people or imagery. The table covers only the networks this article names; it does not review the AI-content rules of Meta, TikTok or Google Ads.
The EU AI Act and why it can matter to a US-based buyer#
If any of your ads reach people in the EU, there is a second layer. Article 50 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026, and it places separate duties on two parties. Providers of generative AI systems must mark their outputs in a machine-readable format as AI-generated; providers of systems already on the market before 2 August 2026 have until 2 December 2026 to comply with that marking duty. Deployers, which includes an advertiser that uses an AI system under its own authority, must disclose that image, audio or video content constituting a "deep fake" has been artificially generated or manipulated. A deep fake is AI-generated or manipulated content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. Where the content is part of an evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing its existence in a way that does not hamper enjoyment of the work, and the Commission's guidelines say an advertisement may qualify in some situations and not in others. The provider's machine-readable marking does not replace the visible disclosure.
Deployers are in scope if they are established or located in the EU, or are located elsewhere and the output is used in the EU, which the Commission reads as covering a deployer that directs or authorizes distribution in the EU. The guidelines also say that a company which merely commissions an advertising agency to produce an ad, without deciding whether and how the agency uses AI, is not a deployer. Fines for breaching Article 50 can reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher (for small and medium-sized enterprises, whichever is lower). This sits on top of, not instead of, each network's own policy. The Commission's guidelines of 20 July 2026 are non-binding, and whether Article 50 applies to a particular campaign is a question for qualified EU counsel.
What about AI voiceovers in native video ads?#
Native video and outstream formats raise the same question in audio form. A synthetic voice reading a script is a production choice, but several rules reach it. Realize says AI-generated voiceover, narration or music is in scope of its AI-content declaration, Outbrain/Teads requires AI-generated audio to be labeled, and under the EU AI Act audio content that resembles existing persons and would falsely appear authentic can be a deep fake. A voice built to sound like a specific real person, such as a celebrity, a public figure or a recognizable brand voice, without consent is an impersonation and right-of-publicity problem however the disclosure is worded: Realize does not permit AI voice clones of real individuals even with disclosure, and Outbrain/Teads permits them only with explicit consent and a disclosure. New York's synthetic-performer law exempts audio advertisements, so it does not apply to a voiceover on its own.
A practical disclosure workflow for a media-buying team#
A short internal checklist can catch the common problems, but it is not a substitute for legal advice: obligations depend on where an ad is shown, who or what it depicts and which network carries it, and campaigns that use realistic synthetic people, health or finance claims, or EU or New York audiences warrant review by a qualified lawyer. If part of your process involves looking at how competitors handle the same question, understand the boundary between legitimate competitive research and anything murkier.
- Flag any creative that shows a person, real or synthetic. If it is not a licensed photo of an actual person who consented to appear, do not imply that it is one, and check whether a synthetic-performer or deep-fake disclosure applies where the ad runs.
- Verify every claim before it ships, not after a network flags it. A model can produce "9 out of 10 doctors recommend" text with no source behind it. Check it against something real or cut it, the same standard you would apply to a human copywriter. The FTC expects a reasonable basis for a claim before the ad runs.
- Check each network's AI rules before upload: declare or label AI content where the network asks, and do not remove provenance markers where a network says not to.
- Keep an internal log of AI-assisted creative: which tool, which prompt, who reviewed it and which label was applied. It helps your team trace where each element came from when a network asks about labeling or substantiation.
- Treat AI-written testimonials as endorsement content under 16 CFR Part 465 and the Endorsement Guides, not as generic copy, and use the same creative analysis discipline you would apply to any claim-heavy hook.
- Confirm that your image or video generator's terms allow commercial advertising use of outputs on your plan. Terms differ by tool and plan, and this article has not reviewed any generator's terms.
Before you commit to a visual style or claim pattern, it can help to see how other advertisers in your vertical have framed similar creative. OpenAdLibrary's index holds captured ads from Meta, TikTok, Google Ads and native networks. Free accounts see the 20 newest Meta, TikTok and native ads, and search and filters need the Founders plan, $29.99 per month. A capture shows that an ad was seen, with a date. It does not show that a network approved it, that it complies with the law or that it performed, so it cannot tell you what a reviewer or regulator will accept.
What can happen if you skip a required disclosure#
The consequences depend on which rule applies. A network can reject an ad or warn and suspend an account: Realize says misdeclaration "can lead to ad rejection, account-level warnings, and - in regulated markets - fines on you as the advertiser", and Outbrain/Teads describes red-flag warnings and possible suspension or a ban from its network for repeat or severe violations. In New York, the synthetic-performer law carries civil penalties of $1,000 for a first violation and $5,000 for any subsequent violation. In the EU, breaches of Article 50 can be fined up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher (for small and medium-sized enterprises, whichever is lower). In the US, the FTC's reviews rule allows courts to impose civil penalties for knowing violations, and deceptive claims or endorsements can lead to FTC enforcement. This article does not estimate how often any of these happens. If you spot an obviously fabricated claim or a deepfaked endorsement in an ad, documenting and reporting it is worth doing.
Rules on AI disclosure are changing: New York's synthetic-performer duty took effect on 9 June 2026, Realize published its AI-content declaration guidance on 20 July 2026, and the EU's Article 50 duties began on 2 August 2026. Re-check the sources linked above before each new campaign, and see the native ad creative checklist for the creative-review side.
For routine AI-assisted copy and minor edits, the FTC sources reviewed here do not require a separate "made with AI" label, provided the claims are true and substantiated, but some networks ask for a declaration or label anyway. For realistic synthetic people, deep fakes, AI-generated testimonials and ads shown in New York or the EU, specific disclosure duties can apply, and network rules differ. The sources above were read on 7 October 2026. None of this is legal advice, so have a qualified lawyer review the campaigns that carry the most risk.







