AI Content Disclosure Rules: What Each Platform Requires (2026)

If you publish AI-generated content, two different rule sets apply to you at the same time, and they are easy to confuse.

The first layer is platform policy. Each platform decides what it wants labeled, applies its own labels when it detects AI, and enforces with the tools it has: removal, reduced distribution, or loss of monetization. Breaking these rules costs you an account, not a fine.

The second layer is law. In the European Union, Article 50 of the AI Act creates a legal transparency duty for anyone who publishes AI-generated images, audio or video that could pass for real. Breaking that one costs money, and it applies whether or not the platform you posted on cares.

Here is where the four largest platforms stand today. Every cell below is taken from the platform's own published policy, quoted in the sections that follow.

Two patterns are worth noticing before the detail. First, every platform that requires disclosure scopes the requirement to REALISTIC content. Clearly stylized or animated output is treated differently, and on YouTube it is explicitly exempt. Second, none of these platforms penalises the label itself. The penalty is always for hiding the AI, or for posting content with nothing original in it. If you are building a repeatable posting workflow, disclosure belongs in the template, not in the exception list.

What is required. TikTok's Community Guidelines put it plainly: "we require creators to label AI-generated or significantly edited content that shows realistic-looking scenes or people." The requirement is scoped to realistic output. TikTok's help documentation additionally "encourages" labeling for anything completely generated or significantly edited by AI, which is a softer ask than the realistic-content requirement.

How you disclose. The post settings carry an AI-generated content switch. Turned on, the post shows a label reading "creator labeled as AI-generated." TikTok also accepts disclosure written into the post itself: "Creators can disclose content as AI-generated directly on the post by adding text, a hashtag sticker, or context in the post's description."

The automatic label. TikTok applies the label itself in two situations: when you use TikTok's own AI effects, and when you upload a file carrying Content Credentials, the provenance metadata standard from the Coalition for Content Provenance and Authenticity (C2PA). In TikTok's words, this happens "when a creator uses TikTok AI effects or uploads AI-generated content that has Content Credentials attached." Once that automatic label lands, you cannot take it off: "Once your content is labeled as AI-generated with an auto label, you won't be able to remove the label from your post." Practically, this means the decision is often already made for you by the tool that produced the file.

What happens if you do not label. "Unlabeled content may be removed, restricted, or labeled by our team, depending on the harm it could cause." That is a graduated response tied to harm, not a fixed ladder of strikes. Labeling content that is NOT AI-generated is also a violation: misleadingly applying the label to unaltered content "is a violation of our Terms of Service and may result in the removal of content."

Does the label cost you reach? This is the question every operator actually asks, and TikTok answers it directly: "Turning on the AI-generated content setting won't affect the distribution of your video as long as it doesn't violate our Community Guidelines." Treat that as the published position, not as a measured guarantee about your specific account.

The rule that quietly matters more. Separate from AI labeling, TikTok states that "Content is also ineligible for the FYF if it includes unoriginal or reused material without anything new." The For You Feed is where reach comes from. For a high-volume AI account, thin repetition is a bigger structural risk than the disclosure toggle ever was.

Dates. The current Integrity and Authenticity guidelines were released on 14 August 2025 and took effect on 13 September 2025. The same page announces a further update effective 24 September 2026, whose text we have not reviewed.

What is required. Meta's Community Standards are unusually direct about the obligation and the consequence in the same sentence: Meta says it will "require people to disclose, using our AI-disclosure tool, whenever they post organic content with photorealistic video or realistic-sounding audio that was digitally created or altered, and we may apply penalties if they fail to do so."

Read that scope carefully, because it is narrower than most summaries claim. As written, the mandatory disclosure covers photorealistic VIDEO and realistic-sounding AUDIO. Still images are not named in that sentence. That does not make AI images invisible on Meta: they are covered by the labeling system described below, and by the separate rule on content that could mislead on a matter of public importance. But the explicit "you must disclose or face penalties" wording is tied to video and audio.

The label. Meta applies an AI Info label across Facebook, Instagram and Threads. Its Transparency Center describes the trigger: labels were added "when we detected industry standard AI image indicators or when people disclosed that they were uploading AI-generated content." So the label can arrive from your disclosure or from the file's own metadata, without you choosing it.

The high-risk carve-out. For content that does not otherwise break the rules, Meta says it "may place an informative label on the face of content, or reject content submitted as an advertisement," when the content is photorealistic or realistic-sounding, digitally created or altered, and "creates a particularly high risk of materially deceiving the public on a matter of public importance." A synthetic persona posting lifestyle content is not the target here. A synthetic persona commenting on an election is.

Direction of travel. In July 2026 Meta announced it is signing the EU AI Act Code of Practice on Transparency of AI-Generated Content, and described its provenance work through the Partnership on AI and C2PA. The practical read: metadata-driven automatic labeling is going to get better, not worse, which shifts the value of self-disclosure from "optional honesty" toward "getting ahead of a label you will receive anyway."

What is required. "To help keep viewers informed about the content they're viewing, we require creators to disclose when they use AI to meaningfully alter or generate photorealistic content." YouTube then names three specific triggers. You must disclose content that:

• Makes a real person appear to say or do something they did not do.
• Alters footage of a real event or place.
• Generates a realistic scene that did not actually occur.

How you disclose. During upload, under Attributes in YouTube Studio, the AI use setting takes a Yes or No.

What you do NOT have to disclose. YouTube is the only platform of the four that publishes a usable exemption list: "Creators don't need to disclose non-realistic content that's made with AI, or edits to realistic content that are minor." The examples it gives as exempt include beauty filters, colour and lighting adjustment, special-effects filters, sharpening, upscaling and video repair, audio repair, cloning your own voice for voiceovers or dubs, and production assistance such as scripts, outlines and thumbnails. Fully animated and clearly unrealistic content is also exempt. For anyone running a stylized rather than photorealistic character, that list is the single most useful paragraph on this page.

Where the label shows up. For photorealistic AI content, "a label in the video player may also appear." For non-photorealistic or animated content, the label sits in the expanded description instead.

When you cannot change it back. If YouTube's systems label your video and you disagree, you can usually correct the disclosure survey. Three cases are locked: "content made with YouTube's AI tools, content containing C2PA metadata, or content labeled after manual review can not be adjusted."

What happens if you do not disclose. This is the strictest published consequence of the four platforms: "Creators who consistently choose not to disclose this information may be subject to manual application of a label, or penalties from YouTube, including removal of content or suspension from the YouTube Partner Program." Losing YPP is losing the revenue, not just the video.

The monetization rule that catches AI channels. On 15 July 2025 YouTube renamed its "repetitious content" policy to "inauthentic content," clarifying that it "includes content that is repetitive or mass-produced." YouTube framed this as a clarification rather than a new restriction: "This type of content has always been ineligible for monetization under our existing policies, where creators are rewarded for original and authentic content." If your channel's videos feel interchangeable from one to the next, that is the policy you will meet, and it has nothing to do with whether you ticked the AI box. The same logic drives how you should think about monetization in the first place.

Likeness. YouTube also documents a likeness detection tool that "helps creators find content on YouTube where their face appears to be altered or generated by AI." It currently matches faces only, and YouTube states it aims to extend detection to audio.

X is the outlier, and the difference is structural rather than a matter of degree.

Across the X Rules, the Authenticity policy and the enforcement documentation we reviewed, we found no general requirement to label AI-generated content and no creator-facing AI disclosure toggle equivalent to TikTok's switch, Meta's disclosure tool or YouTube's AI use attribute. What exists instead is a harm test.

The rule. "Synthetic and Manipulated Media: You may not deceptively share synthetic or manipulated media that are likely to cause harm. In addition, we may label posts containing synthetic and manipulated media to help people understand their authenticity and to provide additional context." The policy detail names, among the categories not allowed, "media depicting a real person that has been fabricated or simulated, especially through use of algorithms or broader artificial intelligence."

Note the two hinges: DECEPTIVELY, and LIKELY TO CAUSE HARM. An openly fictional AI character is a different object under this policy than a fabricated clip of a real person. X also concedes the limits of its own detection: "In situations where we are unable to reliably determine if it is misleading media, we may not take action."

What enforcement looks like. X publishes its enforcement options, and the reach penalty is the one that bites a growing account first. Limiting post visibility can mean "excluding the post from search results, trends, and recommended notifications," removing the post from the For you and Following timelines, restricting its discoverability to the author's profile, downranking it in replies, and restricting likes, replies, reposts, quote posts, bookmarks and sharing. Above that sit required post removal and account-level action.

How to read this practically. Less mandatory labeling is not less risk; it is less predictable risk. On TikTok or YouTube you can comply by ticking a box. On X, compliance is a judgment about whether your content deceives someone about a real person in a way likely to cause harm, and the first sign that a judgment went against you may be a reach drop rather than a notification. Separate policies also require content warnings on certain sensitive media, so a post can be perfectly fine on the synthetic-media rule and still require marking under another one.

This section is general information, not legal advice. If real money depends on the answer, pay a lawyer in the relevant jurisdiction.

The obligation. Article 50(4) of Regulation (EU) 2024/1689 reads: "Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated." A deep fake is defined broadly, as 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."

The creative carve-out, and its limit. The same paragraph continues: "Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the transparency obligations set out in this paragraph are limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work." This is a lighter obligation, not an exemption. You still disclose. You are simply allowed to do it in a way that does not wreck the work.

How the disclosure has to reach the viewer. This is the point most summaries miss, and it changes what compliance actually looks like. The European Commission's own guidance states that "Deployers must disclose deepfake content to a natural person upon first exposure at the latest," in a manner that is "understandable and perceivable by natural persons (e.g. with visible or audible labels), without need for any specific technical tools." It then closes the obvious loophole: "deployers cannot simply rely on the machine-readable marking embedded in the content by the provider under Article 50(2) of the AI Act to fulfil their disclosure obligation." An invisible watermark baked in by your generator does not discharge your duty. A visible label does.

Dates. "Article 50 of the AI Act applies as from 2 August 2026." A limited grace period exists only for the separate machine-marking duty on providers, and only for AI systems placed on the market before that date; those providers must comply from 2 December 2026. Content generated before 2 August 2026 "does not need to be labelled retroactively," although the Commission encourages it.

Penalties. Article 99(4) sets administrative fines "of up to EUR 15 000 000 or, if the offender is an undertaking, up to 3 % of its total worldwide annual turnover for the preceding financial year, whichever is higher," and item (g) of that list is "transparency obligations for providers and deployers pursuant to Article 50." Article 99(6) softens this for smaller operators: "In the case of SMEs, including start-ups, each fine referred to in this Article shall be up to the percentages or amount referred to in paragraphs 3, 4 and 5, whichever thereof is lower."

Whether it applies to you. Article 2 extends the Regulation to "providers and deployers of AI systems that have their place of establishment or are located in a third country, where the output produced by the AI system is used in the Union." The trigger is where the output lands, not where you sit. A creator in the United States whose synthetic persona has European followers is inside the scope. This is the part that surprises people who assumed a non-EU address was a shield, and it is why disclosure is worth building into your marketing process rather than bolting on later.

1. Turn disclosure on by default, per platform. TikTok's switch, Meta's disclosure tool, YouTube's AI use attribute. Make it part of the upload checklist rather than a case-by-case judgment, because the judgment is where mistakes happen.
2. Put a visible line in the bio and in the caption. A platform toggle satisfies the platform. A human-readable statement is what Article 50 asks for, and it travels with a screenshot when the post is reshared somewhere the toggle does not exist.
3. Know whether your output is photorealistic. It is the hinge in every policy on this page. Stylized and animated work carries a materially lighter obligation, and on YouTube it is explicitly exempt.
4. Assume the file will disclose for you. C2PA Content Credentials already trigger automatic labels on TikTok and lock the disclosure on YouTube. Check what your generator writes into the file before you argue with a label.
5. Never label real footage as AI. TikTok treats false labeling as a Terms of Service violation with removal on the table. The label is a factual claim in both directions.
6. Treat the originality rules as the bigger threat. TikTok's For You Feed ineligibility for unoriginal material and YouTube's inauthentic content policy are the two rules most likely to quietly cap a high-volume AI account, and neither one is fixed by disclosing.
7. Do not confuse a watermark with a disclosure. Under Article 50 they are different obligations sitting on different parties. Yours is the visible one.

If the vocabulary in these policies is unfamiliar, the AI influencer glossary defines the terms, and what is an AI influencer covers the category itself.

A page like this is only worth citing if it is honest about its edges. These are the things we went looking for and did not confirm against a platform's own documentation on 5 September 2026.

TikTok's "four-tier penalty system." Several compliance-tool websites describe a fixed four-stage escalation for AI disclosure violations on TikTok. We could not find that structure, or any numbered tier system, anywhere in TikTok's own Community Guidelines or help documentation. TikTok's published language is discretionary and harm-scaled: content "may be removed, restricted, or labeled by our team, depending on the harm it could cause." We have left the four-tier claim out rather than repeat it.

TikTok Creator Rewards eligibility for AI content. The widely repeated claim is that fully AI-generated video is excluded from Creator Rewards while AI-assisted editing stays eligible. We could not load TikTok's Creator Rewards Program terms to check the wording. The claim is consistent with TikTok's verified statement that unoriginal or reused material is ineligible for the For You Feed, but we are not presenting it as confirmed.

Whether TikTok requires labels on clearly stylized AI content. TikTok's requirement is scoped to "realistic-looking scenes or people," yet its own definition of AI-generated content explicitly includes work in artistic styles such as painting, cartoon and anime. The two statements do not resolve each other, and TikTok does not spell out which way it falls. Assume the safer reading if your account sits near the line.

Meta's ad-rejection statistics. Specific figures circulate for how often undisclosed AI triggers an ad rejection, along with a three-strike account cadence. We found no Meta-published source for either, and have not reproduced them.

Meta's rules for AI-generated still images in organic posts. Meta's mandatory-disclosure sentence names photorealistic video and realistic-sounding audio. We could not find equivalent mandatory-disclosure wording for still images in the Community Standards, though images clearly fall inside the AI Info labeling system. We have described the gap rather than filling it with an assumption.

X's position on AI labeling generally. Our conclusion that X operates no general AI-disclosure requirement is based on the X Rules, the Authenticity policy and the enforcement-options documentation. It is an absence of a found rule, which is weaker evidence than a quoted rule, and X changes policy quickly.

• Integrity and Authenticity, TikTok Community Guidelines (released 14 August 2025, effective 13 September 2025; labeling requirement, unlabeled-content enforcement, For You Feed originality rule).
• About AI-generated content, TikTok Help Center (the AI-generated content switch, C2PA automatic labeling, distribution statement, false-labeling rule).
• Misinformation, Meta Community Standards (the AI-disclosure tool requirement and the high-risk labeling carve-out).
• Labeling AI Content, Meta Transparency Center, updated 19 February 2025 (the AI Info label and its triggers).
• Meta is Signing the EU AI Act Code of Practice on Transparency of AI-Generated Content, Meta Newsroom, July 2026.
• Disclosing use of GenAI content, YouTube Help (disclosure requirement, exemption list, label placement, risks of not disclosing).
• YouTube channel monetization policies, YouTube Help (the 15 July 2025 rename of repetitious content to inauthentic content).
• Likeness detection, YouTube Help (face matching today, audio stated as planned).
• The X Rules, X Help Center (the synthetic and manipulated media rule as stated to users).
• Authenticity, X Help Center (misleading-media definition and the categories not allowed).
• Our range of enforcement options for violations, X Help Center (visibility limits, required post removal).
• Article 50: Transparency Obligations, reproducing Regulation (EU) 2024/1689 (deployer deep-fake disclosure and the creative carve-out).
• Transparency obligations under Article 50 of the AI Act, European Commission FAQ (application date, first-exposure disclosure, the machine-readable marking point, no retroactive labeling).
• Article 99: Penalties, European Commission AI Act Service Desk (the EUR 15 000 000 or 3 percent ceiling and the SME rule).
• Article 2: Scope, European Commission AI Act Service Desk (third-country deployers where the output is used in the Union).

Platform policies change without notice. Where a claim on this page matters to a decision you are making, open the source and confirm the wording yourself.