Do you have to label AI content? What Meta, TikTok and YouTube say, and a one-page team policy

An AI label is a notice that tells viewers a piece of content was made or changed with AI. Some labels are added by you, with a switch or checkbox when you upload. Others are added by the platform on its own. Some come from a law, not a platform. People often mix these up, which is how a team ends up either over-labelling everything or missing the one case that mattered.

This page is about platform and public labels: what appears on the post or ad that the public sees. It is not about telling a client how you made the work. That is a different question, covered in should agencies disclose AI use. It is also not about who owns the output, which the commercial use checklist covers. It is information, not legal advice. Platform rules change often, so use the links in the sources and re-read them.

The short rule of thumb

Ask three questions about each asset.

1. Does it look real? A photo-style person, a believable street, a news-style scene.
2. Did the real thing happen? A real person saying something they never said, a place that does not exist shown as real, an event staged by AI.
3. Could a viewer be misled? If a sensible viewer would feel tricked on learning how it was made, label it.

Two yeses on the first two, label it. That is the strictest common ground across the platforms below, and it is the safest default for a team. Stylised, animated or clearly unreal content, and small edits such as colour, cropping or noise reduction, are usually outside the rules, but the platforms word the exceptions differently.

Rules matrix, by platform

This table gives the position at a glance. The sections after it carry the detail and the source for each row. Always check the linked page before you rely on a row.

Platform	Who applies the label	Trigger	Notes
YouTube	You, at upload; YouTube can add one too	Realistic altered or synthetic content	Exemptions for clearly unrealistic content, minor edits, help with scripts and captions, and cloning your own voice. Repeated non-disclosure can lead to removal or Partner Program suspension.
TikTok	You; TikTok adds one to its own AI effects and to uploads carrying content credentials	Realistic AI content	From TikTok's support page. Check the page for ad-specific wording.
Meta (Facebook, Instagram), ads	Meta applies "AI info" itself	Ads made or significantly edited with Meta's own tools; third-party AI tools detected by industry signals	The post also mentions "Paid for by" disclosures for social issue, election and political ads. Varies by region. The post shows no advertiser AI switch, so check Ads Manager.
Meta, organic posts	Check Meta's help centre	Rules for organic posts differ from the ad rules	Read Instagram and Facebook help for organic posts.
EU (law)	The person publishing a deepfake	Realistic AI image, audio or video of people, places or events	Disclose clearly, at first exposure at the latest. Applies from 2 August 2026. Lighter duty for evidently artistic, satirical or fictional work.
New York (law)	The advertiser	A synthetic performer in an ad, when the advertiser knows	In force from 9 June 2026. Penalties are set in the statute.

YouTube

YouTube, Google's video platform, asks creators to tell it when realistic content has been altered or made synthetically. Its help page lists what needs a disclosure and what does not. Content is exempt when it is clearly unrealistic, when the edit is a minor aesthetic one, when AI only helped with scripts or captions, and when you clone your own voice for a voiceover. The disclosure is a setting in the upload flow, and YouTube shows a label to viewers. It can also add a label itself, and it says that labels are applied automatically to content carrying C2PA (an industry standard for content credentials that record how a file was made).

The consequence worth knowing: YouTube says that repeated failure to disclose can lead to content removal or suspension from the YouTube Partner Program (the programme that pays creators). For a business channel, that is a real cost. If you publish on YouTube, start with AI video for YouTubers and course creators. Also read the line about a real person: realistic content that shows a real person saying or doing something they did not is a clear case for disclosure. If you are making that kind of content, the consent steps come first, and the consent checklist covers them.

Meta: Facebook and Instagram

Meta's own announcement about gen AI transparency in ads (updated 1 June 2026) is the page I rely on here. In short, Meta shows an "AI info" label in "About this ad" for ads that were made or significantly edited with its own AI tools. It says it also detects AI from other tools through industry signals, which are markers some tools embed in the file. And when Meta's own tools put a photorealistic AI human in an ad, the label sits next to the "Sponsored" tag instead of behind the three-dot menu. An ad with no significant edit and no photorealistic human gets no label. This varies by region.

What that means for a business: Meta, not you, applies the label to most ads, so you cannot "choose" to skip it if your file carries the right signals. The post also lists "Paid for by" disclosures for social issue, election and political campaigns inside the same "About this ad" panel. Marketing blogs describe manual advertiser AI controls too. The post does not mention one, so treat that as unconfirmed and read Meta's own Ads Manager help before acting on it.

For organic Instagram and Facebook posts, rules differ from the ad rules. Read the current help centre pages for creators, and apply your own policy below as the floor. The ad-specific traps, such as claims and before-and-after images, are in a separate page, AI-generated ads: platform policy traps.

TikTok

TikTok's support page on AI-generated content says that realistic AI content needs a label. TikTok also adds its own label to videos made with its AI effects and to uploads that carry C2PA content credentials, and it does so automatically. A label is not a penalty. TikTok's page says the AI-generated content setting "won't affect the distribution" of your video as long as it follows the Community Guidelines. It also says an auto label cannot be removed once applied, and that falsely labelling unaltered content as AI is itself a Terms of Service breach. Separately, TikTok does not allow AI content showing adult private figures without their permission, or anyone under 18. Marketing blogs say undisclosed AI in TikTok ads is rejected or restricted. TikTok's ads policy changes, so for paid content read it directly.

For sellers there is a second set of rules. TikTok Shop Academy's page on requirements for high-quality videos and LIVEs says an AI voice or pre-recorded audio is banned in LIVE streams, that in a LIVE stills, slideshows and animated figures can fill at most half the screen, and that posted videos need real-world footage with at least three seconds of dynamic content. The page showed a date of 26 October 2026, so treat the details as current but re-check them. A Mexico-version page also carried an AI disclosure rule for Shop content, which I mention only as a hint to look at your own region's page. For the video side of this, see TikTok Shop product videos: what must be real footage.

Laws that go further than platforms

Platform rules are only part of it. In some places the law itself requires a label, and the platform's toggle does not satisfy the law.

European Union

Article 50 of the EU AI Act covers transparency. People who publish deepfakes, meaning realistic AI image, audio or video of people, places or events, must disclose that the content is artificially generated or manipulated. The disclosure must be clear and appear at first exposure at the latest. The duties for these publishers apply from 2 August 2026. A later EU law, the Digital Omnibus (Regulation (EU) 2026/1744), did not delay those duties. It did give providers whose systems were already on the market until 2 December 2026 to add machine-readable marking to what their tools produce. Artistic and satirical work gets a lighter duty. If you sell into the EU, read the Article 50 text and ask a lawyer, because the details of "deployer", exceptions and enforcement matter.

New York

New York's synthetic performer ad law was signed on 11 December 2025 and took effect on 9 June 2026. It requires a disclosure in an ad when the advertiser knows a synthetic performer, an AI-generated person, appears in it. The bill text sets penalties of $1,000 for a first violation and $5,000 for each later one. If you advertise to New York audiences with an AI person on screen, add the disclosure and keep proof of when you did.

Marketplaces

Marketplaces have their own rules. Etsy's listing image policy (updated 22 October 2026) says most listings need one unedited real photo, while other listing images may use AI if they are accurate. Amazon, in a notice on its Seller Central forum in July 2026, asked for the keyword contains-synthetic-performer in the XMP metadata field dc:subject on images or videos showing photorealistic AI-generated people. Whether that scope widens is unconfirmed. Details are in AI product images for Amazon and Etsy.

What gets labelled automatically, and why that is not a plan

C2PA content credentials are a record of where a file came from, which can travel with the file. Some AI tools add them, and some platforms read them to add labels, as YouTube and TikTok say they do. For example, Higgsfield's terms (sections 5.5 and 6.4) say it may embed provenance marks, without a promise that they will last, and that users must not remove them. So the same file may be auto-labelled on one platform and not on another.

Do not build your policy around those automatic labels. Secondary sources report that many platforms strip file data on upload and that screenshots lose it, so the absence of a credential proves nothing. Tell your client before posting if you know your tool embeds credentials, because a label may appear without anyone clicking a button. I cover the details in the page on provenance marks and why not to strip them.

A one-page team policy you can copy

Paste this into your team's handbook and edit it. Keep it short, so people read it.

1. Scope. This applies to every image, video and audio file we publish for ourselves or a client, organic or paid.
2. The test. If the asset looks real and shows a person, place or event that did not happen, it is labelled. If it is clearly stylised, animated or only lightly edited, a label is optional, and the owner of the account decides.
3. Strictest rule wins. When we post the same asset on several platforms, we use the label rule of the strictest platform for all of them.
4. Real people. No recognisable real person is shown in AI content without a signed consent form on file.
5. Where the label goes. Use the platform's own label setting first. Add words in the caption or on screen when the platform has no setting, or when a law requires it. The label must be visible at first exposure, not hidden in a long description.
6. Log. For each asset we keep: tool, date, who made it, whether a real person is shown, and where a label was applied.
7. Clients. We tell clients which platform labels may be added, including automatic ones, before an asset goes live.
8. Review. We re-read the platform and legal pages every three months and update this policy. Next review date: [date].

Pair this with an approval step. The client sign-off page shows a simple way to add a "label check" before approval.

Where to re-read each rule

Pages change. A page I list as updated in June 2026 may read differently by the time you check. Put these in a calendar reminder.

• YouTube: the help page "Disclosing use of altered or synthetic content". It also says disclosure alone does not limit a video's audience or affect monetization.
• TikTok: the support page on AI-generated content, and TikTok Shop Academy if you sell there.
• Meta: the gen AI transparency post for ads, plus the ad standards pages.
• EU: the text of Article 50 and your national regulator's guidance.
• New York: the bill text for the synthetic performer law.
• Marketplaces: Etsy's listing image policy and Amazon Seller Central notices.
• Google Ads: the misrepresentation policy, which bars manipulating media to deceive, defraud or mislead. It has no AI-specific wording, so claims fall under it.

Platform account managers can answer case-specific questions, and for EU or New York paid ads, a lawyer is worth the cost. A short summary page like this one is a starting map, not the final word.

How I'd approach this

I would not argue with a platform about what counts as "realistic." I would pick the strict rule, label the content, and move on. TikTok's own page says its label setting does not affect distribution if the content follows its guidelines, and a label costs less than a removed video or a suspended channel.

Then I would write the policy above, put the log in a shared sheet, and add a quarterly reminder. Add a "label check" line to your approval step, so it cannot be skipped. If you want that process built into a shared asset library or approval portal, Blur Studio can build it around the tools you already use, and you can read more here. A simple log and a calendar reminder go a long way on their own. For the ad side, read the policy traps page next.