Amazon Requires Sellers To Label AI-Generated People In Listing Images - Forbes

The requirement puts a disclosure decision inside every listing image that features an AI-generated person.

Amazon this week began requiring third-party sellers to identify product images and videos containing photorealistic AI-generated people. The company notified sellers Wednesday that qualifying images and A+ content — the enhanced graphics and video modules on listing pages — must carry specific metadata keywords before upload, according to CNBC, which reviewed the announcement. In the notice, Amazon wrote that 'recent legislation requires disclosure when images or videos in advertisements contain photorealistic AI-generated people.'

Amazon also plans to add a shopper-facing indicator to listings where applicable, per the same report. The criteria that will determine when that indicator appears have not been detailed publicly, and the company did not immediately provide comment to CNBC. eWeek reported that sellers are directed to embed an IPTC-compatible metadata keyword flagging synthetic performers in qualifying assets.

Two exclusions matter to creative teams. The requirement does not cover fictional characters from film, television, or video games, and it does not cover real people whose appearance has been altered with AI.

The law behind the requirement

New York's synthetic performer disclosure law took effect June 9. Gov. Kathy Hochul signed S.8420-A/A.8887-B in December 2025, and her office describes it as first-in-the-nation. It amends General Business Law § 396-b and applies to advertisements in any medium. Announcing the effective date, Hochul said the state is 'setting the rules of the road' rather than letting AI run the show.

The statute defines a synthetic performer as a digital asset created, reproduced, or modified by computer using generative AI or a software algorithm, intended to give the impression of a human performance by someone who is not recognizable as an identifiable person. Anyone who produces or creates a covered advertisement and knows it contains a synthetic performer must conspicuously disclose that fact. Civil penalties run $1,000 for a first violation and $5,000 for each subsequent one.

That definition also explains the shape of the carve-outs. Digital replicas of identifiable real people sit outside the synthetic performer definition, with separate right-of-publicity and digital replica laws governing that use, and advertisements for expressive works are excepted where the synthetic performer's role matches the underlying film, show, or game.

The statute reaches distribution as well. Attorneys at Reed Smith note that a publisher or distributor put on notice of an undisclosed synthetic performer has five days to remove the ad or bring it into compliance. Background figures count too, which means a synthetic bystander in a lifestyle scene falls inside the definition alongside a synthetic hero model.

The performance case and the label sit on the same asset

Amazon markets generative AI creative tools to the same sellers now responsible for tagging. On its advertising site, the company reports that Sponsored Brands campaigns using AI-generated images delivered 10.3% higher return on ad spend than campaigns without them, that brands using its AI creative tools advertised five times as many products, and that advertisers who adopted its Sponsored Brands image generator saw 10% more sales per advertiser a month after adoption. Those figures come from Amazon's own internal U.S. measurements across 2024 and 2025.

Third-party sellers account for more than 60% of units sold on the marketplace, and AI lifestyle imagery has become routine among them because it costs a fraction of a photo shoot and turns around in hours rather than weeks. The disclosure requirement lands on exactly that inventory. For marketing leaders, the efficiency case and the disclosure decision now attach to the same asset, which makes this a creative budgeting question as much as a compliance one.

What AI labels do to shoppers

The available evidence on labels points one direction, with caveats worth reading closely. In a 2×2 experiment published in Equilibrium in December 2025, Khalil Israfilzade of Vytautas Magnus University found that advertisements labeled as AI-generated drew lower consumer trust and purchase intent than those labeled human-made, with the gap widest for high-involvement products such as laptops. Advertisements labeled human-made scored higher even when AI had in fact produced them.

Broader label research suggests the penalty may not track actual deception. In two preregistered experiments with 4,976 U.S. and U.K. participants published in PNAS Nexus, Sacha Altay and Fabrizio Gilardi found that labeling headlines as AI-generated lowered perceived accuracy and willingness to share them regardless of whether the headline was true or human-written. The effect was roughly a third the size of labeling a headline false, and the authors traced it to an assumption of full automation with no human oversight. They also caution that the finding was tested on headlines and may not generalize to images or video.

The practical read for a CMO is that a shopper-facing indicator functions as a conversion variable, not only a compliance artifact, and that it likely behaves differently by category. A $14 phone case and a $1,400 mattress do not invite the same scrutiny.

The workflow this creates

Five steps follow from the announcement:

- Audit the catalog. Identify which ASINs use synthetic human figures, starting with highest-traffic listings and main images, then A+ modules and brand store pages.

- Move tagging upstream. Metadata belongs in asset production, not a post-upload cleanup pass. Teams working with agencies or freelancers can make identification of synthetic elements a delivery requirement and allocate tagging responsibility and liability in the contract.

- Keep creative records. Because the statute turns partly on intent and knowledge, briefs and prompts specifying a mannequin, cartoon, or abstract avatar document that an asset was never meant to read as a human performer.

- Choose the creative lane per category. Product-only imagery, real models with proper releases, and synthetic models carry different costs and different disclosure outcomes. AI-generated backgrounds behind a real model sit outside the synthetic performer trigger, which makes hybrid production a strategy rather than a compromise. New York's Fashion Workers Act separately requires written consent before creating or using an AI digital replica of a model.

- Test the route, not the label. Brand Registry sellers can compare product-only, human-model, and synthetic-model creative through Amazon's experiment tooling. The indicator itself is not a variable a seller controls.

Where the patchwork goes next

New York is first rather than last. TikTok's Shop guidance directs sellers to label content fully or substantially created or altered by AI, and YouTube began automatically applying AI labels in May when its systems detect significant photorealistic AI use. California's AI Transparency Act, as amended by AB 853, becomes operative August 2, with large online platform duties beginning January 1, 2027 that include a prohibition on knowingly stripping compliant provenance data from uploaded content. Provenance metadata is built to travel with the file, so an untagged asset can still carry a signal downstream.

Federal direction remains unsettled. Tech Policy Press counts 109 state AI laws enacted through July 1 of this year, while the December 2025 executive order on a national AI framework created a Justice Department litigation task force to challenge state AI rules and directed Commerce to review them. No federal statute currently preempts state disclosure requirements, and firms advising advertisers have generally recommended continued compliance with state law while the question is litigated.

For marketing teams, the question has shifted from whether AI imagery is cheaper to which images can afford a label. That is a brand decision, and it is a better one to make before shoppers see the indicator than after.