TikTok, YouTube, Meta AI labels
One table: TikTok, YouTube, Meta, and Article 50. The label is not a ranking penalty. Unoriginal work is. Platform toggles are not the Act.
Read the table. TikTok, YouTube, and Meta each have a disclosure rule. Article 50 is the fourth column: a legal duty on the deployer, not a platform toggle. The label is not a ranking penalty in any of the first three columns. Unoriginal, templated content is. Clearing a platform checkbox does not clear the Act. This page is the comparison. It does not restack the TikTok checker, the Creativity Program rule, or the Meta bonuses post.
Facts in the grid come from dated sources: the Commission's Article 50 FAQ (last updated 24 July 2026); SocialScale Hub's 2026 academy recap of TikTok, Instagram, and YouTube disclosure (including Instagram's 31 August 2026 profile-label post); Vexub's 2026 monetization-policy recap; Fliki's 25 August 2026 note; Paul, Weiss (4 August 2026); Morgan Lewis (12 August 2026). Where a cell would need a number those sources do not publish, the cell says so.
How to read the table
Each row is a decision you actually make at export. Each column is a different addressee.
- TikTok binds the login. The Creativity Program is a money path on that login. Community Guidelines are a removal path on the same login.
- YouTube binds the channel. Studio's altered-content control is the composer path. The inauthentic-content policy is the earnings path.
- Meta binds the Facebook / Instagram account. C2PA/IPTC plus self-disclosure is the ingest path. Reels bonuses depend on disclosure. Instagram's AI-generated profile label (31 August 2026) is an account-level extra for AI-generated persons.
- Article 50 binds the deployer of the AI system when the output is used in the EU. It does not bind "the TikTok." It does not care which app carried the file.
A "yes" in three platform columns and a "no" in Article 50 is a possible grid: obviously fantastical work can still want a platform toggle on one app and not on another, and it can sit outside the deepfake definition. A "no" in three platform columns and a "yes" in Article 50 is also possible: a perceivable on-file line for an EU viewer when the platform's own realism test did not fire, or when the platform badge is easy to miss on mobile. Do not average the columns.
The AI content label is the platform tag. Article 50 is the legal duty. Content credentials are the machine-readable signal some columns read at ingest. Three names. Three rows in your head before you look down.
The comparison
| TikTok | YouTube | Meta (Facebook / Instagram / Reels) | Article 50 (EU AI Act) | |
|---|---|---|---|---|
| Who it binds | The account. Creativity Program is extra eligibility on that account. | The channel. YPP is extra eligibility on that channel. | The account. Reels bonuses are extra eligibility on that account. | The deployer of the AI system (professional use). Not each social login as its own legal person. |
| What triggers | AI-generated visuals and audio that depict realistic people or scenes. Captions, scripts, colour, crop generally do not. | Realistic synthetic media of real people, or a realistic scene that did not happen. Beauty filters, captions, cloning one's own voice for a voiceover sit on YouTube Help's exemption list. | Labels required for generated or meaningfully edited media. Production-assist without featuring an AI-generated person does not need the profile badge. | Deepfakes: generated or manipulated image, audio, or video that resembles existing (or plausibly existing) persons, places, entities, or events and would appear authentic. Public-interest text unless substantively reviewed. |
| How you disclose | Built-in AI label in the posting flow. TikTok also auto-reads C2PA and can apply the badge; you cannot take an auto-label off. | Studio "altered or synthetic content" checkbox. YouTube may apply a label from C2PA or detection if you skip it. | C2PA / IPTC at ingest plus self-disclosure on the post. Instagram profile-level "AI-generated profile" (renamed 31 Aug 2026) for accounts whose featured person is AI-generated. | Visible or audible label at first exposure, perceivable without a special tool or a dedicated extra tap. Machine-readable marking is the provider duty, not a substitute. Official EU icons optional. |
| Where the viewer sees it | On-video badge. | Typically in the expanded description; a more prominent player label on sensitive topics. Easy to miss on mobile. | In-post "AI info" / "Made with AI" chip; profile badge on the account when used. | On or in the content at first hit — burned-in line, held icon, or audible open. Not only metadata. |
| Money path | Creativity Program: labelled AI eligible, same rate. Undisclosed AI can mean removal or program suspension. | AI stays monetizable when it carries the creator's own vision. Inauthentic-content policy is the earnings knife, not the disclosure badge. | Reels bonuses depend on disclosure. Creation method is not a published bonus haircut. | Not a platform pool. Fines cap at €15M or 3% worldwide turnover for the party that failed the duty. No published per-post tariff. |
| Ranking / reach | Label is not a published ranking penalty. Viewers can ask to see less AIGC in Manage Topics (audience filter). Unlabelled realistic AIGC can be removed or restricted. | Disclosure is not framed as a ranking or YPP switch. Inauthentic / templated work is. | Label is not a ranking penalty for people who disclose. Instagram (31 Aug 2026): adding the AI-generated profile label does not change reach; leaving it off an AI-person profile can limit recommendations. | Not a ranking system. First-exposure perceivability is the test. |
| Exempt / quieter | AI-assisted text (scripts, captions, hashtags) in the 2026 recaps. Obviously non-real scenes. | Clearly unrealistic content, animation, special effects, beauty filters, captions, cloning one's own voice. | Using AI in the creative process without featuring an AI-generated person (profile badge). Standard filters vs generative media. | Standard assistive editing. Evidently artistic / satirical / fictional work still discloses, quieter so it does not spoil the work. Personal, non-professional use outside the Act. |
| Auto-detection | C2PA, creator tools, invisible watermarking. TikTok newsroom (10 July 2026, via SocialScale Hub): over 3 billion videos labelled AIGC on that stack. | C2PA and detection; YouTube may apply the label if you consistently skip. | C2PA / IPTC and classifiers. Instagram will notify unlabeled AI-person profiles it detects. | Provider marking (watermarks, C2PA, other) aimed at detectors. Grace to 2 Dec 2026 for systems already on the market before 2 Aug 2026. Interoperability to 2 Feb 2027. Deployer cannot rely on that mark alone. |
| Start date | Platform policy, already in force (TikTok required realistic AIGC labels well before 2026). | Platform policy, already in force (YouTube disclosure blog 18 March 2024 is still the cited page). | Platform policy, already in force; Instagram profile rename 31 August 2026. | 2 August 2026. Guidelines adopted 20 July 2026. No deployer grace. |
| Retroactivity | Not an archive rebuild in these sources. New posts follow current policy. | Same. | Same. | Generated and published before 2 Aug 2026 not retroactive. Generated before, published after: treat as covered. |
The 3 billion figure is TikTok's own 10 July 2026 newsroom number as carried by SocialScale Hub (2026). Storrito still published 1.3 billion on the same three-method stack; SocialScale Hub prints both and prefers TikTok's figure. Either way, automated detection is the default. Betting on being missed is not a cell in this table.
What the fourth column adds
Three platform columns answer "will this login keep the post up and keep the pool open." The fourth column answers "does a person in the EU, hitting this file, perceive that it was generated or manipulated."
That is why a Studio checkbox that lives behind "show more" can pass YouTube and fail Article 50 on a phone. It is why TikTok's built-in badge can pass the Creativity Program and still leave you short of first-exposure if the badge is the only mark and a download strips it. It is why Meta's C2PA ingest can apply "AI info" and still leave a deepfake unlabelled for a viewer who never reads chips.
The fourth column also names a different person. Employees of a company are not separate deployers; the legal person is. An agency that controlled whether and how AI was used can hold the duty even if the client's logo is on the end card. A US creator with a globally accessible post is in territorial scope on Bird & Bird's reading of the 20 July 2026 guidelines. Article 50 if you are not in the EU is that map. None of the first three columns ask for a passport.
Fines attach to the fourth column's addressee, not to "TikTok." Article 50 fines is the ceiling. Platform suspension attaches to the first three. Do not file a Creativity Program removal as an AI Act case.
The disclosure template is the copy for column four. Does my TikTok need an AI label? is column one as a checker. Creativity Program AI-label rule is column one's money sentence. Meta AI labels and Reels bonuses is column three's money sentence. This grid is the only place they sit side by side.
What the table is not
It is not a ranking of which app is "strictest." Strictness is a different row depending on whether you care about removal, RPM, bonuses, or a legal duty.
It is not a promise that labels will never become a ranking weight. Ranking is opaque. The published record in those recaps is: not a ranking penalty; originality is. Instagram's 31 August 2026 profile row is a penalty for skipping a badge, which is the same shape as unlabelled-AIGC removal, not a penalty for applying one.
It is not a substitute for YouTube Help 14328491, TikTok Creator Academy, Instagram's creators blog, or the Commission's FAQ. Those pages move. This table is dated 9 September 2026 against sources gathered 4 September 2026.
It is not a licence to skip advertising disclosure or likeness releases. A labelled fake testimonial is still a fake testimonial. FTC endorsement rules are a fifth column this grid does not grow.
If you generate the file in Versely's AI video generator, the export does not fill four columns for you. Flip each platform control at upload. Burn a first-exposure line when column four fires. If you need a studio seat for that overlay, sign up. One file, four questions. The table is the page.
FAQ
If I satisfy TikTok, YouTube, and Meta, have I satisfied Article 50?
Not automatically. Platform badges can fail first-exposure — a description checkbox, a chip a mute-autoplay never reads, a manifest a re-encode killed. Article 50 wants a perceivable mark at first hit for a deepfake reaching an EU audience. Stack the overlay and the toggles.
Which column decides monetization?
The money rows: Creativity Program (same rate if labelled; suspension if not), YouTube YPP via inauthentic-content rather than the badge, Meta Reels bonuses that depend on disclosure. Article 50 is not a CPM. Originality is the shared knife across the first three.
Do I need a different file per platform?
Usually no. One master with a burned-in first-exposure line travels. Then set each app's own control at upload, and the Instagram profile badge if the featured person is generated. A clean unlabelled master is the file that fails column four the first time someone downloads it.