Do you have to label AI-generated images and video?

Marketing teams now generate images and video in minutes, and three different rule sets have started to ask what viewers are told. One is EU law. One is each platform's own policy. The third is US consumer-protection law, which asks a different question entirely, about invented customers rather than invented pixels.

These rules overlap but do not match, and the loose summary that all AI content must be labeled is wrong in both directions. This article separates them. It describes what each source says, and it stops short of telling you how a regulator or a court would apply it to your case.

What EU law requires, and from when

Article 50 of the EU AI Act sets transparency duties for AI-generated content, and its obligations became generally applicable and enforceable by national authorities on 2 August 2026. It splits the duty in two. Providers of the AI systems must make outputs machine-readable as artificially generated. Deployers, the people and companies using those systems, must disclose certain content to the people who see it. (Goodwin, 2026)

The provider duty says outputs must be marked in a machine-readable format and be detectable as artificially generated or manipulated. For most marketing teams that duty sits with the company that built the image or video tool, not with them. The AI Omnibus gave providers of generative systems already on the market before 2 August 2026 until 2 December 2026 to comply with that marking obligation. That grace period covers providers only. (AI Act Art. 50 (accessed 2026); Goodwin, 2026)

The deployer duty is the one a marketer is likely to hold. A deployer of an AI system that generates or manipulates image, audio or video deepfake content must disclose that the content is artificially generated or manipulated. Disclosure has to be clear and distinguishable, made at the latest at the first interaction or exposure, and must meet applicable accessibility requirements. Goodwin reports that non-compliance may bring fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. (AI Act Art. 50 (accessed 2026); Goodwin, 2026)

Which images and video count as a deepfake

Only some AI-generated media is a deepfake under the Act. Article 3 defines the term as AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. A realistic synthetic photo of a real-looking street scene fits that wording. A clearly fantastical illustration does not obviously fit it. (AI Act Art. 3 (accessed 2026))

The Act also reduces the duty for works that are evidently artistic, creative, satirical or fictional. For those, the disclosure only has to acknowledge that the content exists, in a way that does not hamper the display or enjoyment of the work. Where a particular image sits between those categories is a judgment, and the careful default for anything realistic is to disclose. (AI Act Art. 50 (accessed 2026))

What YouTube, TikTok and Meta ask of you

Each platform's rule turns on realism, and the three are worded differently. YouTube says it requires creators to disclose AI-generated or altered content that seems realistic, set through the AI use question in YouTube Studio during upload. It does not require disclosure for clearly unrealistic content or for minor aesthetic edits, and it lists using AI for scripts, titles or thumbnails as production help that needs none. (YouTube Help (accessed 2026))

TikTok's community guidelines say it requires creators to label AI-generated or significantly edited content that shows realistic-looking scenes or people. In 2024 TikTok also began automatically labeling AI-generated content uploaded from certain other platforms, by reading Content Credentials, the C2PA metadata attached to a file. Meta applies its AI info label when it detects industry-standard AI image indicators or when people disclose that they are uploading AI-generated content. (TikTok Community Guidelines (accessed 2026); TikTok, 2024; Meta Transparency Center (accessed 2026))

These pages change, and each platform can enforce its own policy as it chooses. Check the current wording before you rely on any summary, including this one. Metadata-based labeling also means a file that has passed through a tool that strips its metadata can arrive unlabeled by the platform, which leaves the creator's own disclosure as the only reliable route.

Text is treated differently from images and video

The Act's text duty is much narrower than its media duty. Deployers must disclose AI-generated or manipulated text only when it is published with the purpose of informing the public on matters of public interest, and even then the duty does not apply if the text went through human review or editorial control and a person or company holds editorial responsibility for its publication. The exception sits on the text duty only, not on the deepfake duty. (AI Act Art. 50 (accessed 2026))

Product marketing copy is a commercial message, not text published to inform the public on matters of public interest, so on the statutory wording the text duty is unlikely to reach it. That is a reading of the text and not a ruling, and commentary on the Commission's guidance is not uniform. The platforms point the same way: YouTube explicitly lists AI help with scripts and titles as needing no disclosure. (AI Act Art. 50 (accessed 2026); YouTube Help (accessed 2026))

The sensible split is to treat realistic synthetic media as the thing that gets labeled, and treat the words as yours to answer for. Sheevook follows that split: it keeps an AI flag on generated images and video through to publishing and sends it where a platform's interface has a field for it, and it does not add an AI label to the text of a post. Not every platform has such a field, so check what yours shows.

A label does not fix an invented customer

US federal rules treat a fabricated customer as its own offense, whether or not anything is labeled as AI. The FTC's rule on consumer reviews and testimonials, 16 CFR Part 465, makes it a violation to write, create or sell a review or testimonial that materially misrepresents that the reviewer exists, or that they used or had experience with the product. Buying such reviews, or spreading them when you knew or should have known, is also covered. (16 CFR 465.2 (accessed 2026))

The FTC said when it announced the rule that it covers reviews by someone who does not exist, such as AI-generated fake reviews. Alston & Bird says AI is not mentioned in the rule's text, but the FTC's statement of basis says AI-generated reviews are covered. The rule took effect on 21 October 2024, and the FTC can seek civil penalties, then $51,744 per violation, against knowing violators. (FTC, 2024; Alston & Bird, 2024)

Do not use a synthetic person as a stand-in customer, a testimonial or a reviewer, and do not rely on a caption to cure it. Whether a given disclaimer changes the analysis for a particular ad is a question for a lawyer. Sheevook does not generate a synthetic customer for this reason.

A short routine before you publish

Decide by looking at the file, not at the tool that made it. Ask what a viewer would reasonably think they are seeing, and label when the answer is a real person, place or event. The steps below are a working routine and not a legal test.

1. 1

   Sort the asset

   Is it realistic, or obviously stylized or fictional? Realistic images and video are the ones platforms and the EU deepfake duty are written for.

2. 2

   Disclose realistic synthetic media

   Use each platform's own control where one exists, such as the AI use question on YouTube. Do not leave the label to chance on platforms that rely on file metadata.

3. 3

   Keep the record

   Mark generated assets as AI-generated in your own files, so the decision does not depend on someone remembering how an image was made.

4. 4

   Never invent a customer

   No synthetic reviewer, testimonial or case-study subject. Use a real customer, with real permission, or no customer at all.

5. 5

   Check the pages before relying on this

   Platform rules and the EU's guidance move. Reread the platform's current help page, and ask a lawyer where the stakes are high.