.svg)
On 22 July 2026, Amazon notified all third-party sellers of a new requirement: every product image and every piece of A+ content containing a photorealistic AI-generated person must carry metadata tags before it is uploaded, as reported by the Consumer News and Business Channel (CNBC). This is not a shift in how Amazon views AI content. It is the direct consequence of a state law in the United States.
For Amazon sellers operating storefronts in the US market, the change moves a legal obligation from the platform onto the merchant. The use of AI images, long treated as a way to cut creative production costs, now carries a disclosure duty, a penalty exposure and a direct effect on conversion.
1. What Amazon requires for the use of AI images in listings
Amazon's mechanism is primarily a technical requirement rather than a wholesale change of policy on AI-generated content. Specifically, before uploading an image to Amazon, sellers must use image editing software that supports the IPTC standard (International Press Telecommunications Council - the standard governing how information is stored inside an image file) to add the label contains-synthetic-performer, indicating that the image contains an AI-generated figure, into the file's metadata. Metadata is the hidden layer of information travelling with each image file: it is not visible to the viewer, but systems can read and process it.
One point deserves attention: the identification duty sits entirely with the seller. Amazon does not detect these assets on the merchant's behalf, which means any misclassification is the storefront's own exposure.
Image source: CNBC
1.1. Cases where AI tagging is mandatory
Under Amazon's requirement, sellers must apply the metadata tag where an image or piece of detail-page content contains a photorealistic AI-generated human figure. This applies to the following content types:
-
Product images, including both the main image and secondary images in a listing.
-
A+ Content, covering images, video and the enhanced graphics used to present the product on the detail page.
-
Fully AI-generated people who look like real humans and are not based on any actual individual.
-
Any AI-generated figure appearing in the image, whether a main subject in the foreground or a person in the background of a lifestyle scene (a pedestrian on the street, a customer in a café, a figure seated in the distance).
1.2. Cases where no tagging is required
Amazon does not require the metadata tag in the following cases:
-
Images contain only real people, even where the image has been edited with AI tools such as object removal, upscaling, lighting adjustment or background replacement.
-
Images with no people in them, such as product-only or object-only shots.
-
Illustrations, animation, comic art or stylised graphics that are not photorealistic.
-
Characters from film, television programmes or video games, which are not AI figures created to simulate a real person.
The line between the two groups is narrow. A photograph of a real model with an AI-replaced background and an added pedestrian in the distance falls into the tagging group, even though the primary subject is a real person.
2. Why Amazon requires AI image tagging: the rule originates in New York's "Synthetic Performer" law
Amazon's policy traces to Section 396-b of New York's General Business Law, effective 9 June 2026. The statute requires the party that creates or produces a commercial advertisement to make a conspicuous disclosure where it has actual knowledge that the advertisement contains a synthetic performer.
The statutory definition is the part sellers should read closely: a synthetic performer is a human-like digital asset, created by generative AI or a software algorithm, that engages in a visual or audiovisual performance. The law targets synthetic human performances rather than AI-generated content in general. Common practices such as image retouching, quality enhancement, background generation and non-human graphics may fall outside its scope.
The statute also sets out its exemptions:
-
Advertising for expressive works: motion pictures, television programmes, streaming content, documentaries and video games - The synthetic performer's use is consistent with the underlying work.
-
Audio-only advertisements.
-
Cases where AI is used solely to translate the dialogue of a real human performer.
-
Parties that merely publish or disseminate advertisements: newspapers, magazines, television networks, streaming services, billboards and transit advertising.
Civil penalties run to USD 1,000 for a first violation and USD 5,000 for each subsequent one. The statute does not create a private right of action, so enforcement is likely to rest with the New York Attorney General. The liability structure matters more than the figures: the final exemption above likely covers Amazon in its capacity as a distributor, and Amazon has pushed the identification step onto sellers (the party that actually creates the advertising content).
This means there is currently no settled legal standard for determining whether a business has fully complied. Until the New York Attorney General's office or the courts provide detailed guidance, no method of disclosure can be treated as legally watertight. That is why sellers should follow Amazon's prescribed technical mechanism precisely rather than designing their own approach to disclosure.
3. Why sellers outside the United States are still affected
Amazon has not scoped this requirement to shoppers in New York. Tagging applies at catalogue level, which means a storefront operated from Vietnam, China or Singapore must still process its entire image library against the same standard. The use of AI images in listings is therefore a general operational matter rather than an issue confined to one state's market. Note too that Amazon's requirement is broader than the statute in places: a synthetic figure appearing in a background lifestyle scene must still be tagged, even though it is not a performance in the narrow statutory sense.
The legislative direction is widening beyond New York. The California AI Transparency Act (SB 942), as amended by AB 853, becomes operative on 2 August 2026. The scope is worth understanding correctly: it places no direct obligation on sellers. It regulates providers of generative AI systems with more than one million monthly users — that is, the very image tools sellers are using. Those providers must embed provenance data into the content their systems generate and offer a tool that lets anyone check whether a file was AI-created.
From 1 January 2027, the Act adds duties for "large online platforms": social media platforms, file-sharing platforms, mass messaging platforms and stand-alone search engines with more than two million unique monthly users. They must detect provenance data attached to content and surface that information to users. A marketplace such as Amazon does not obviously fall inside that definition, but the regulatory direction is clear.
The practical implication for sellers lies elsewhere: AI provenance is increasingly embedded into image files at the point of creation, regardless of whether the seller declares it. An untagged image can still carry provenance data inside the file itself and be identified further downstream. Accurate declaration is therefore the lower-risk option compared with staying quiet and hoping the platform fails to notice.
Content platforms have also moved ahead of legislators. Meta, TikTok, Pinterest, Google and YouTube all apply AI labels to uploaded video and images. Even so, TikTok and Meta have recently been criticised for inadequately labelling ads that use AI-generated influencers to promote dubious products — in some cases without the named brand's knowledge. TikTok has said it banned accounts making misleading health claims, and Meta has said it labels AI video. The lesson for sellers: platform labelling still has gaps, so the pressure to tighten it will grow rather than ease.
No US federal law currently mandates disclosure of AI-generated advertising content, but the working compliance standard is being set by platforms faster than by legislators. Sellers should treat this as the new baseline rather than a temporary rule.
4. The commercial impact of labelled AI imagery
4.1 Compliance and listing risk
The immediate risk is not the New York penalty, since that targets advertisers with actual knowledge of the content. The practical risk sits with Amazon: a storefront that misdeclares or overlooks assets can face content removal, policy violation warnings or listing disruption in the middle of peak season. For accounts running thousands of SKUs, a manual audit of the entire image library is a substantial workload.
4.2 Buyer trust risk
Consumer research indicates that advertising carrying an AI label tends to draw lower trust and lower purchase intent, particularly for higher-value or high-consideration categories. That creates a concrete economic question: the saving from the use of AI images has to be weighed against the revenue that may be lost once the label appears on the detail page.
The answer differs by category. For low-ticket consumables, the effect may be negligible. For fashion, cosmetics or higher-priced home appliances, photographing real models is likely to remain the financially stronger choice.
5. Five steps to standardise the use of AI images on an Amazon storefront
-
Audit the image library: list every live image and A+ asset, recording the origin of each file and the tools used to create or edit it.
-
Classify against the definition: test each file against the scope in Section 1 and separate the assets that require tagging from those that are exempt.
-
Move tagging upstream: require the design team or outsourced studio to embed the metadata at export, rather than reworking files after they are already live.
-
Keep provenance records: retain prompts, source files and edit logs to evidence the basis of classification if the platform reviews the account.
-
Measure the impact: track conversion and revenue by SKU before and after the label appears, to decide which categories should revert to real photography.
The fifth step is the one most often skipped and the one with the clearest financial value. Without sales data broken out by SKU and by creative asset type, a business has no basis on which to judge whether the use of AI images is saving money or eroding margin. This is where standardised sales data and SKU-level revenue reconciliation come in — the problem that Genbook and Sliner's accounting automation service address.
6. Content transparency becomes a compliance line item
The new rule does not prohibit the use of AI images. It converts them from a creative choice into a compliance item with documentation, process and financial consequences. Amazon sellers that build the process early avoid the risk of listing disruption and retain control over their creative options.
Sliner works with cross-border e-commerce businesses to review compliance obligations market by market and to design a legal structure that fits a multi-platform selling model. As more US states legislate on content transparency, establishing which entity carries responsibility for advertising activity in each market becomes part of the corporate structuring question.
Explore Sliner's Corporate Structuring & Tax Planning service, or contact the advisory team to review the compliance position of your Amazon storefront.
Data source: CNBC
