AI disclosure rules by platform: Google, YouTube, TikTok, Meta, Amazon and more

Platform rules change faster than laws, and several platforms refuse automated reads. We read each help page we cite on October 21, 2026 and say where we could not. This is general information, not legal advice, and a platform's own text is the authority on what it asks. For the laws, start at the AI disclosure laws overview.

The platforms at a glance

The tag Google asks for, and the values around it

The IPTC Photo Metadata Standard has a field, Iptc4xmpExt:DigitalSourceType, that holds a URI from IPTC's controlled vocabulary. Each value is the base address http://cv.iptc.org/newscodes/digitalsourcetype/ followed by a term. The four terms that matter for AI imagery, with IPTC's labels and definitions:

Two other signals show up. C2PA Content Credentials are a signed manifest inside the file, and an action in the manifest records a digitalSourceType, using the same IPTC terms. Amazon's reported keyword is a plain text entry in the XMP dc:subject list. A tool can read what a file declares. It cannot tell you whether the declaration is true, and a file with no marker proves nothing about how the image was made. Our AI image metadata guide explains the fields, the checker reads them from a file, and the writer adds them without re-encoding the image and warns before it writes into a file that carries a C2PA manifest, because adding XMP to such a file breaks its signature.

Sources: IPTC Digital Source Type vocabulary; IPTC Photo Metadata Standard 2025.1; C2PA specification 2.4. Verified October 21, 2026.

Google Merchant Center

Google's help page on preserving metadata tags says all AI-generated images must carry the IPTC DigitalSourceType value trainedAlgorithmicMedia, and that metadata such as DigitalSourceType must not be removed from AI-generated images, including those made with Product Studio. It applies to the image_link, additional_image_link and lifestyle_image_link attributes. A second help page covers AI-generated text: titles and descriptions generated with AI go in structured_title and structured_description, with the sub-attribute digital_source_type set to trained_algorithmic_media. The second page lists three IPTC values to preserve: TrainedAlgorithmicMedia, CompositeSynthetic and AlgorithmicMedia.

What the pages leave open: they name no value for an AI-edited real photo, such as a real product with an AI-generated background, so which tag Merchant Center expects there is unstated; and they do not say whether Google reads the tag in the uploaded file or in the file served at the image link, so a CDN step that strips XMP between your master file and the served image is a risk to test. Google's pages describe a requirement and state no penalty, and we found no enforcement action.

Sources: Merchant Center: preserving metadata tags for AI-generated images (dated February 2024); Merchant Center: AI-generated content. Verified October 21, 2026.

Google Search

Google's image metadata documentation, last updated December 10, 2025, lists four DigitalSourceType values: trainedAlgorithmicMedia, compositeSynthetic, algorithmicMedia and compositeWithTrainedAlgorithmicMedia. It also says "If an image contains C2PA metadata, Google can extract those details" for the "About this image" feature, which shows signers on the C2PA Trust List. Its guidance on generative AI content adds, for ecommerce, that AI-generated images must contain the IPTC TrainedAlgorithmicMedia metadata, and that AI-generated product data such as titles and descriptions must be specified separately and labeled as AI-generated. It also advises human review of AI text that can show in results, including alt text.

Sources: Google Search Central: image metadata; Google Search Central: generative AI content. Verified October 21, 2026.

Google Ads and YouTube ads

In July 2026 Google rolled out an AI label setting across Google Ads, Display & Video 360, Campaign Manager 360, Merchant Center and Ads Editor. An advertiser can mark an asset "Label this asset as created or edited with AI". A designated asset gets a disclosure in "How this ad was made" worldwide, and visible overlays on ads in campaigns that target the European Union, India and New York. Google also permits text or visual AI labels inside image and video creatives without breaking its text overlay and watermark policies. Google may label assets itself, including when it gets signals from other platforms, and labels cannot be overwritten. Google states that the setting does not guarantee compliance and tells advertisers to seek legal guidance.

The one plainly mandatory Google disclosure on the pages we read is for election ads: advertisers must tick "Altered or synthetic content" when an ad inauthentically depicts real or realistic-looking people or events. We do not say that every Google ad must carry an AI label, because the pages do not.

Sources: Google Ads policy change log, July 2026; Google Ads Help: AI content label settings. Verified October 21, 2026.

YouTube

YouTube asks creators, including businesses that upload marketing video, to answer "AI use" in YouTube Studio when realistic content was generated or meaningfully altered with AI. Its triggers include a real person appearing to say or do something they did not, altered footage of a real event or place, and a generated realistic scene that did not occur. Non-realistic content such as animation, and minor aesthetic edits such as color or lighting adjustments, are outside it. YouTube may also label content made with its own AI tools, content carrying C2PA metadata and content its systems detect, and creators generally cannot change those labels. It carries forward AI disclosures from tools that use C2PA version 2.1 or higher. Creators who consistently do not disclose may face penalties, including suspension from the YouTube Partner Program. We found no named enforcement action.

Sources: YouTube Help: disclosing use of altered or synthetic content; YouTube Help: how this content was made. Verified October 21, 2026.

TikTok

TikTok's Community Guidelines require creators to label AI-generated or significantly edited content that shows realistic-looking scenes or people, and say unlabeled content may be removed, restricted or labeled depending on harm. For ads that are not Spark ads, an advertiser whose image, video or audio is completely AI-generated or significantly modified by AI ticks "This ad contains AI-generated content" in Ads Manager when creating the ad; the setting cannot be unset after submission unless the creative changes. TikTok has said it reads C2PA Content Credentials to label AI content uploaded from other platforms. The ads help page shows "Last updated: September 2025".

Sources: TikTok Community Guidelines: edited media and AI-generated content; TikTok Ads Manager: add disclaimers to ads. Verified October 21, 2026.

Meta

Organic content. Meta's systems read industry-standard signals in uploaded files, including C2PA and IPTC "AI generated" information, and label content "AI info". Meta may require a label on photorealistic video and realistic-sounding audio that was digitally created or modified. Its page states "This requirement does not apply to images", although images may still be labeled if its systems detect AI generation or modification. The page warns of penalties for content shared without a required label.

Ads. Ad images created or significantly edited with Meta's generative AI features, or with third-party AI tools, can be labeled "AI info" on the "About this ad" screen, though not all ads get the label. Advertisers on social issue, election or political ads must disclose photorealistic or realistic-sounding media made or edited with third-party generative AI tools; non-disclosure leads to rejection and repeated failure may bring penalties. Meta's advertising standards page also says that from June 1, 2026 Meta uses automated detection for ad media made with third-party AI tools. A fresh fetch of Meta's pages returned nothing, so we did not re-read that sentence on October 21, and the page does not name the signals the detector uses.

Sources: Meta Help Center: how to identify AI content; Meta Help Center: AI-generated images in ads; Meta Advertising Standards: social issues, elections or politics. Verified October 21, 2026.

Pinterest

Pinterest says that when its systems detect, through a Pin's metadata (following the IPTC standard) or through classifiers, that a non-promoted Pin was generated or modified with AI, an "AI modified" label shows. The label also appears if the content owner indicated that AI was used. For ads, the label reflects what the owner indicated. The advertising guidelines page we read had no AI disclosure rule.

Source: Pinterest Help: Gen AI labels. Verified October 21, 2026.

Amazon listings and A+ Content

On July 22, 2026 Amazon reportedly told sellers to add the keyword contains-synthetic-performer to the XMP dc:subject field of any image or video, including A+ Content, that shows a photorealistic person created entirely by AI. Reported exclusions are real people edited with AI, television, film and game characters, images with no people and non-photorealistic people. Amazon reportedly says it will show a disclosure on listings "where applicable". The rule rests on CNBC's report of July 23, 2026 and on seller guidance that quotes Amazon's help page. We could not read Amazon's own text, because the two Seller Central help pages cited for it redirect to a seller sign-in.

Three cautions follow. First, we do not tell sellers that tagging older images is required: reports about older images disagree, and Amazon's own page is the place to check. Second, writing a keyword into a file that carries a C2PA manifest breaks the manifest's signature, so check the file first. Third, Amazon's resized listing images may not keep embedded metadata; see what happens to a tag in transit. A tag in a file is also not the New York or California label for an ad, which is about what viewers see; see US rules on AI-generated people in ads.

Source: Amazon Seller Central help page (needs a seller login), not read; reported by CNBC on July 23, 2026. Verified October 21, 2026.

Amazon KDP

Kindle Direct Publishing requires publishers to tell KDP about AI-generated content, meaning text, images or translations created by an AI-based tool, "even if you applied substantial edits afterwards", when they publish a new book or edit and republish one. AI-assisted content, where you created the work and used AI to edit, refine, error-check or brainstorm, does not need to be disclosed. The page does not say that customers see the disclosure.

Source: KDP content guidelines: artificial intelligence content. Verified October 21, 2026.

Etsy

Etsy's Seller Policy says that if an item is created through the use of artificial intelligence, you must disclose this in your relevant listings, and its Creativity Standards place seller-prompted AI creations in the "Designed by a seller" category with the same disclosure in the listing description. Etsy's pages mention no image metadata, no dedicated AI field and no checkbox. Etsy refuses automated reads, so we read dated copies of its own pages; the AI sentence in the Seller Policy reads the same in every copy we compared from December 2025 to October 2026. We did not check the listing editor, so we cannot say whether it shows an AI attribute. Blog posts that describe a January 2026 AI field, or counts of listings removed, have no source in Etsy's own policy text.

Sources: Etsy Seller Policy; Etsy Creativity Standards. Verified October 21, 2026.

Steam

Steam's content survey has a section on generative AI, concerned with AI used to create content that ships with a game. Developers describe "Pre-Generated" content, made with AI tools during development, and "Live-Generated" content, created while the game runs, for which they also describe the guardrails against illegal content. Efficiency gains from AI tools used in development are not the focus of the section. We did not confirm how, or whether, the store page shows the disclosure to players.

Source: Steamworks content survey. Verified October 21, 2026.

Spotify

Spotify supports an industry standard for AI disclosures in music credits, developed through DDEX and submitted by labels, distributors and partners. It also says vocal impersonation is allowed only when the impersonated artist has authorized it. A later update on the page says Spotify launched a beta on April 16 (the page does not print the year) in which credits for vocals, lyrics or production appear in Song Credits on mobile. The page notes that the absence of a credit does not mean AI was not used. No file metadata is read.

Source: Spotify newsroom: Spotify strengthens AI protections. Verified October 21, 2026.

What happens to a tag in transit

A tag only helps if it survives to the file the platform or a crawler reads. What we found, from the platforms' and vendors' documentation and from our own October 2026 checks, which used small samples:

The practical consequence is to test the file that is actually served, not only the master. The AI image metadata checker has a mode that fetches an image the way a store serves it. Never use a tool or setting that strips markings: the EU Code of Practice asks signatories not to remove them (Section 1, Measure 1.2), and China's Labeling Measures bar deleting or tampering with labels (China).

Sources: Cloudflare Polish; Cloudflare image transformations; Cloudflare: preserve Content Credentials; WordPress core, Imagick editor; sharp: output options. Verified October 21, 2026.

Platforms we could not read

These gave no readable text to an automated read on October 21, 2026, so we say nothing about their AI rules: X, Medium, Adobe Stock, Shutterstock's contributor terms, Microsoft Advertising and Amazon's advertising creative policy. eBay's US policy pages sit behind a bot check; its UK picture policy says nothing about AI and asks that photos accurately represent the item. LinkedIn's community policies ask for disclosure of synthetic media that depicts a person saying or doing something they did not, and we did not verify whether LinkedIn reads Content Credentials. Shopify's help page says its own AI image generation applies an invisible watermark, and we found no Shopify rule that makes merchants disclose AI product images. Checked October 21, 2026. Sources: Shopify Help Center: media generation; LinkedIn Professional Community Policies; eBay UK picture policy.

Claims we do not repeat

• "Etsy added an AI disclosure field in January 2026." Not in Etsy's own policy text.
• "Amazon is the only major channel that reads a tag from the file." Google Merchant Center, YouTube, Meta, TikTok and Pinterest all say they read signals in files.
• "Every Google ad must carry an AI label." The label setting is optional; the only plainly mandatory Google disclosure we read is for election ads.
• "Meta labels every ad that used a third-party AI tool." Meta's page says not all ads get the label.
• "The EU Code of Practice requires C2PA." It names no file format.
• "Adding the XMP tag satisfies Article 50." The Commission says deployers cannot rely on the marking embedded in the content to meet their visible disclosure duty.