AI content disclosure rules by platform
Most platforms do not ban AI-made content, but many ask you to label realistic synthetic media, and some read Content Credentials to label it for you. Here is what each one asks for, where the setting is, and what happens if you skip it.
Not legal advice. Summaries of each platform's own policy pages, checked 2026-09-25; policies change, so follow the official link before you publish.
YouTubeTikTokInstagram, Facebook & Threads (Meta)LinkedInPinterestSnapchatSpotifyApple PodcastsAmazon KDP (books)Audible / ACX (audiobooks)Google SearchGoogle Ads (election ads)Meta Ads (social issue, election & political ads)EU AI Act, Article 50China: AI-generated content labelling measuresUS FTC rule on fake reviews & testimonialsUS state election deepfake laws
Video platforms
YouTube
YouTube requires creators to disclose realistic content that is meaningfully altered or synthetically generated, and shows an 'altered or synthetic content' label to viewers.
- What must be labelled
- Realistic AI-generated or altered content a viewer could mistake for real: a real person appearing to say or do something they didn't, altered footage of real events or places, or realistic scenes that never happened. Not needed for clearly unrealistic or fantastical content, beauty/colour/lighting filters, script or thumbnail help, cloning your own voice, or audio/video repair and upscaling.
- How to disclose
- In YouTube Studio when uploading, go to Details > Attributes > 'AI use' (altered or synthetic content) and select 'Yes'.
- Automatic labels
- Yes. YouTube labels content made with its own generative AI tools, content that carries C2PA metadata, and content its systems detect as AI-generated or altered.
- If you don't
- Creators who consistently fail to disclose can have a label added for them, content removed, or be suspended from the YouTube Partner Program.
- Monetisation
- YouTube says disclosure does not limit reach or monetisation eligibility. Separately, the YPP 'inauthentic content' policy (renamed from 'repetitious content' in July 2025) excludes mass-produced or templated AI content that adds no original value.
TikTok
TikTok's Community Guidelines (2026 H2) require a clear label when AI or significant editing is used to show realistic-looking people or scenes, and ban AIGC that misleads on matters of public importance or harms people.
- What must be labelled
- AI-generated or significantly edited content showing realistic people or scenes, e.g. face swaps, making someone appear to say something they didn't, misleadingly adding or removing people or objects, or AI audio mimicking a real person's voice. Not needed for small edits (colour, crop, reframe), artistic styles such as anime, or generic TTS narration that isn't a recognisable person's voice.
- How to disclose
- Use TikTok's AIGC label (the 'AI-generated content' setting when posting), or add your own clear caption, sticker or watermark.
- Automatic labels
- Yes. TikTok labels content made with its own AI effects and reads C2PA Content Credentials to label AI content made elsewhere.
- If you don't
- Unlabelled content may be removed, restricted (e.g. made ineligible for the For You feed) or labelled by TikTok, depending on the harm it could cause.
Social networks
Instagram, Facebook & Threads (Meta)
Meta shows an 'AI info' label on Facebook, Instagram and Threads, based on creator self-disclosure or on industry signals it detects in the file.
- What must be labelled
- Organic posts with photorealistic video or realistic-sounding audio that was digitally created or altered. Meta may add a more prominent label to high-risk content that could materially deceive the public on an important matter.
- How to disclose
- Use Meta's AI disclosure and label tool when posting (in Instagram it is an 'AI label' option on the share screen; wording differs between apps).
- Automatic labels
- Yes. Meta detects C2PA and IPTC signals from major AI tools and adds 'AI info'. Content only edited with AI shows the label in the post menu rather than on the post itself.
- If you don't
- Meta says it 'may apply penalties' if people fail to disclose; misleading AI content can also be fact-checked.
LinkedIn shows Content Credentials (C2PA) on images and videos that carry them, and its Professional Community Policies ban undisclosed deepfakes of people.
- What must be labelled
- Synthetic or manipulated media showing a person saying or doing something they didn't, unless you clearly disclose that it is fake or altered.
- How to disclose
- No dedicated toggle. Disclose clearly in the post text or on the media itself. Uploading files that keep their C2PA credentials also shows provenance.
- Automatic labels
- Partly. A C2PA icon appears on images and videos with Content Credentials, including single-image and video ads. Clicking it shows whether AI was used, plus the app and issuer. LinkedIn says it cannot yet identify all AI content.
- If you don't
- Posts that break the policy can be reported and removed under the Professional Community Policies.
Pinterest adds an 'AI modified' label to Pins that it detects as generated or modified with AI, or that the creator discloses.
- What must be labelled
- No strict duty is stated. Creators can voluntarily mark Pins as created or modified with AI.
- How to disclose
- Self-disclose when creating the Pin, or keep the AI metadata in the file.
- Automatic labels
- Yes. Pinterest reads IPTC metadata and uses its own classifiers, even when there are no visible markers. The label appears at the bottom left of the Pin close-up. For ads it sits behind the '…' menu, e.g. 'contains an AI-generated person'.
- If you don't
- The label is added automatically. Creators can appeal a wrong label through support.
Snapchat
Snapchat's Community Guidelines expect responsible AI use and make creators responsible for meeting local AI-labelling laws. Spotlight and Discover recommendations favour authentic content.
- What must be labelled
- Snapchat has no general labelling rule of its own. You must follow AI disclosure or labelling laws where they apply in your region. Manipulated or AI-generated imagery used to harass people is banned.
- How to disclose
- No dedicated AI toggle was documented on the page we read. Use a caption or on-screen text where disclosure is needed.
- Automatic labels
- Snap says it adds transparency to content made with Snapchat's own AI tools.
- If you don't
- Enforcement against accounts that misuse AI, up to account termination. Recommendation systems prioritise authentic content.
Music, podcasts & audiobooks
Spotify
Spotify bans unauthorised AI voice clones and impersonation, filters spam uploads, and shows AI credits (DDEX standard) that artists submit through distributors.
- What must be labelled
- There is no general duty to label AI music. Vocal impersonation of a real artist is allowed only with that artist's permission. AI-use credits (vocals, instrumentation, post-production) are voluntary.
- How to disclose
- Submit AI disclosures through your label or distributor using the DDEX AI-credits standard. Spotify shows them in the song credits.
- Automatic labels
- No automatic AI label was described. Spotify's music spam filter flags mass uploads, duplicates, SEO tricks and artificially short tracks.
- If you don't
- Unauthorised voice clones and deepfakes are removed. Spam tracks are no longer recommended or are removed (Spotify reported 75M+ removed in a year).
- Monetisation
- Spotify says disclosing AI use does not down-rank a track. Spam tactics aimed at the royalty pool are targeted.
Apple Podcasts
Apple Podcasts content guidelines (sections 1.11 and 1.12) require prominent disclosure of AI-generated audio or video and ban misleading uses of AI.
- What must be labelled
- Audio or video generated with AI, including synthetic voices, AI hosts or on-screen personas, and AI replicas of real people. Using AI to fabricate news or manipulate clips into false narratives is banned.
- How to disclose
- Disclose prominently in the content itself and in the metadata (description/notes) of every episode and the show.
- Automatic labels
- No automatic AI label was described.
- If you don't
- Apple may label or remove content, suspend subscription sales, or suspend or terminate the account.
Books & publishing
Amazon KDP (books)
KDP requires authors to tell Amazon when a book contains AI-generated text, images or translations. AI-assisted work does not need to be disclosed.
- What must be labelled
- AI-generated content: text, images or translations created by an AI tool, even if you edited them heavily afterwards. Not required for AI-assisted work, such as content you created yourself and refined, edited or brainstormed with AI.
- How to disclose
- Answer the AI-generated content questions in the KDP publishing flow when you publish a new book or edit and republish an existing one.
- Automatic labels
- No automatic detection or public label is described. KDP does not say whether the disclosure is shown to customers.
- If you don't
- Books that break the content guidelines can be rejected or removed from sale. You are responsible for all AI-generated and AI-assisted content meeting the guidelines.
Audible / ACX (audiobooks)
ACX titles must be human-narrated unless ACX authorises otherwise. Amazon's own AI route is KDP 'virtual voice', whose audiobooks are clearly labelled.
- What must be labelled
- ACX bans unauthorised text-to-speech, AI or automated recordings (submission requirements, April 2026). Virtual-voice audiobooks made through KDP are labelled as computer-narrated.
- How to disclose
- Do not submit AI narration to ACX. For AI narration, use KDP's virtual voice beta (US, eligible eBooks), which applies the label for you.
- Automatic labels
- Virtual-voice titles are labelled automatically for listeners.
- If you don't
- Unauthorised AI-narrated ACX submissions are prohibited and can be rejected.
- Monetisation
- Virtual voice pays a 40% royalty on à la carte sales, with a list price of $3.99-$14.99. KDP Select titles earn from a separate Audible Plus catalogue fund.
Search & ads
Google Search
Google Search does not penalise content just for being AI-generated, but using generative AI to mass-produce pages without adding value is scaled content abuse.
- What must be labelled
- A visible AI disclosure is recommended, not required. Two metadata rules apply: AI-generated images must carry IPTC DigitalSourceType 'TrainedAlgorithmicMedia', and AI-generated product titles and descriptions (e-commerce) must be specified separately and labelled as AI-generated.
- How to disclose
- Optionally explain on the page how the content was made. Keep IPTC metadata in AI images. Label AI product data in Merchant Center feeds.
- Automatic labels
- Google does not add a label in Search results for AI text.
- If you don't
- Scaled content abuse (many low-value AI, scraped or stitched pages made to manipulate rankings) can lead to ranking demotion or removal under the spam policies.
Laws & regulations
Google Ads (election ads)
Election advertisers on Google must disclose ads with synthetic or digitally altered content that inauthentically depicts real or realistic-looking people or events.
- What must be labelled
- Images, video or audio that make a person appear to say or do something they didn't, or that alter footage of real events or show realistic events that never happened. Inconsequential edits (resizing, cropping, colour or brightness, red-eye fixes, background edits that don't create realistic events) are exempt.
- How to disclose
- Tick the synthetic-content checkbox in campaign settings. Google auto-generates the disclosure for feeds, Shorts and in-stream video. In other formats you must add a clear, conspicuous disclosure such as 'This video content was synthetically generated'.
- Automatic labels
- Google auto-generates the disclosure text in some formats once the checkbox is ticked.
- If you don't
- Ads that don't comply can be disapproved, and the account can face political-content policy enforcement.
Meta Ads (social issue, election & political ads)
Meta requires advertisers to disclose when a social-issue, electoral or political ad has been digitally created or altered, including with AI.
- What must be labelled
- Realistic images, video or audio showing a real person saying or doing something they didn't, realistic fake people or events, altered footage of real events, or realistic depictions of events that didn't happen. Minor edits (crop, resize, colour, sharpening) are exempt unless they change the ad's claim.
- How to disclose
- Tick the digitally created or altered disclosure when creating the ad in Ads Manager.
- Automatic labels
- Meta adds a label to the ad and records the disclosure in the Ad Library. For AI images in ordinary ads, 'AI info' can appear under 'About this ad'.
- If you don't
- Ads that don't comply are rejected. Repeated failure to disclose can lead to penalties on the account.
EU AI Act, Article 50
Article 50 transparency duties apply from 2 August 2026. The 2025-26 Digital Omnibus delayed the high-risk rules but not Article 50, apart from a grace period to 2 December 2026 for machine-readable marking by generative systems already on the market.
- What must be labelled
- Deployers (anyone using AI professionally, not purely personal use) must disclose deepfake image, audio or video. For artistic, satirical or fictional works, a lighter disclosure applies. AI text published to inform the public on matters of public interest must be disclosed unless it was human-reviewed and someone holds editorial responsibility.
- How to disclose
- Disclose clearly and distinguishably at the latest at first exposure, e.g. an on-screen or caption label. Providers of generative tools must also mark outputs in a machine-readable way (Art. 50(2)).
- Automatic labels
- Not a platform. Generative-AI providers must embed machine-readable marks (watermarks or metadata) that platforms can read. An EU Code of Practice on marking and labelling is being drafted.
- If you don't
- Fines for breaching transparency duties can reach EUR 15M or 3% of worldwide annual turnover (Art. 99), enforced by national authorities.
China: AI-generated content labelling measures
China's Measures for Labeling AI-Generated Synthetic Content (CAC, MIIT, MPS, NRTA) took effect on 1 September 2025, with the mandatory standard GB 45438-2025.
- What must be labelled
- AI-generated or synthetic text, images, audio, video and virtual scenes published on services in China. Providers add explicit (visible) and implicit (metadata) labels, and users must declare AI content when they publish it.
- How to disclose
- When posting, declare the content as AI-generated and use the platform's labelling function (Art. 10). Do not delete, alter, forge or hide labels.
- Automatic labels
- Yes. Platforms must check files for implicit labels, show prominent notices, and flag content as 'suspected' AI when they detect signs of it.
- If you don't
- Platforms can add labels or act under their own rules, and regulators enforce under China's cybersecurity and AI service rules. Tampering with labels is prohibited.
US FTC rule on fake reviews & testimonials
The FTC's Trade Regulation Rule on Consumer Reviews and Testimonials (final August 2024, in force October 2024) bans fake reviews and testimonials, including AI-generated ones.
- What must be labelled
- It covers reviews or testimonials from people who don't exist (e.g. AI-generated) or who have no real experience with the product, undisclosed insider reviews, reviews bought for a set sentiment, fake 'independent' review sites, and buying or selling fake followers or views.
- How to disclose
- Don't create, buy or post AI-written fake reviews or testimonials. Clearly disclose material connections, following the FTC's endorsement guidance.
- Automatic labels
- Not applicable.
- If you don't
- The FTC can seek civil penalties from people who knowingly break the rule.
US state election deepfake laws
Public Citizen's tracker (updated 22 September 2026) lists enacted election-deepfake laws in about 33 states, and at least one (Hawaii) has been struck down. Most require a disclaimer on deceptive AI media about candidates in a pre-election window.
- What must be labelled
- It varies by state. Typically: materially deceptive AI-generated or altered media of a candidate distributed within a set period before an election, unless it carries a clear 'manipulated' disclaimer. Satire and news reporting are often exempt.
- How to disclose
- Check the state's statute. Usually a clear on-screen or audio disclaimer that the media was AI-generated or manipulated.
- Automatic labels
- Not applicable.
- If you don't
- Depending on the state: injunctions, civil liability or criminal penalties. Several laws face court challenges.
Rules of thumb
- Label anything realistic that a viewer could mistake for a real person, place or event, even if only part of it was generated.
- Script help, grammar fixes, colour correction, captions and clearly fantastical animation usually do not need a label.
- Keep Content Credentials (C2PA) on your exports when the tool adds them: several platforms use them to label automatically.
- Never use a real person's face or voice without consent, AI or not.
Checking a tool's own terms? See commercial use and watermarks by app.