AI-Generated People in Amazon Product Images: The New Disclosure Rule and What Actually Converts

Amazon told sellers on July 22 that every product image containing a photorealistic AI-generated person must now carry a specific metadata tag before upload. Miss it, and your listing risks suppression. The rule applies to main images, secondary slots, A+ Content, and video — across every Amazon marketplace worldwide.

This isn't a suggestion. It follows New York's synthetic performer disclosure law (GBL § 396-b), which carries $1,000 fines for a first violation and $5,000 per offense after that. Amazon is enforcing compliance through metadata scanning at the upload layer, and it will display a visible disclosure label on affected listings.

After optimizing 14,000+ hero images, I can tell you this rule will catch a lot of sellers off guard — not because it's hard to follow, but because most sellers don't know which of their images contain AI-generated people in the first place. The line between "AI-enhanced photo of a real model" and "fully AI-generated person" is blurrier than you think, and getting it wrong in either direction costs you money.

Here's exactly what the rule requires, how to comply in under 10 minutes per image, and — the part nobody else is covering — whether you should be using AI-generated people in your Amazon product images at all.

What Is the Amazon Synthetic Performer Disclosure Requirement?

Amazon's synthetic performer disclosure requirement mandates that sellers tag any product image or video containing a photorealistic person generated entirely by AI with the keyword contains-synthetic-performer in the file's XMP metadata before uploading it to Seller Central. Amazon will then display a disclosure indicator on the listing where applicable.

The requirement went live on July 22, 2026, and applies globally across all Amazon stores. It covers:

• Main (hero) images in slots 1-7
• A+ Content images and modules
• A+ Content videos and product videos
• Sponsored ad creative that features AI-generated people

The legal trigger is New York's General Business Law § 396-b, signed by Governor Hochul in December 2025 and effective June 9, 2026. The law defines a "synthetic performer" as a digitally created asset made using generative AI or algorithms that creates the impression of a human performer who isn't recognizable as any real person.

Amazon's implementation goes further than the law requires. New York's statute applies to advertisements shown to New York consumers. Amazon is applying the metadata requirement to all listings globally — a preemptive move that signals this will become standard across more states and countries.

What Images Need the Tag (And What Doesn't)

This is where most sellers will trip up. The rule sounds simple, but the edge cases matter.

You MUST tag images that contain:

• A photorealistic person created entirely by AI (Midjourney, DALL-E, Stable Diffusion, Flux, etc.)
• AI-generated hands holding your product where the hands look real
• AI-generated faces or bodies in lifestyle scenes
• Composite images where the background scene is real but the person was generated

You do NOT need to tag images that contain:

• Real people whose photos were edited with AI — even heavy retouching, background swaps, or lighting adjustments on a real model's photo don't trigger the requirement
• Images with no people at all — product-only shots, even with AI-generated backgrounds or scenes
• Non-photorealistic characters — cartoon mascots, illustrated people, stylized graphics
• Characters from movies, TV, or video games used in licensed product imagery

The critical distinction: the person must be generated entirely by AI and look photorealistic. A real model photographed against an AI-generated kitchen background doesn't need the tag. An AI-generated "model" wearing your product in a studio-style shot does.

Here's a quick self-test: Could a reasonable viewer believe they're looking at a photograph of a real person? If yes, and that person was AI-generated, tag it.

How to Add the Contains-Synthetic-Performer Metadata Tag

The technical implementation takes under 10 minutes per image. Here are three methods, from simplest to most scalable.

Method 1: Windows File Properties (No Software Needed)

1. Right-click the image file
2. Select Properties
3. Go to the Details tab
4. Find the Tags field under the Description section
5. Type contains-synthetic-performer exactly (case-sensitive, hyphenated)
6. Click Apply, then OK
7. Upload the tagged file to Seller Central

This works for JPEG and TIFF files. It writes to the XMP dc:subject field, which is what Amazon's scanner reads.

Method 2: Adobe Photoshop or Bridge

1. Open the image in Photoshop
2. Go to File > File Info
3. Click the IPTC Core panel
4. In the Keywords field, add contains-synthetic-performer
5. Click OK and save the file
6. Upload to Seller Central

In Adobe Bridge, select the image, open the Metadata panel, and add the keyword to the IPTC Keywords field.

Method 3: ExifTool (Batch Processing for Large Catalogs)

For sellers with dozens or hundreds of images to tag, ExifTool handles batch processing via command line:

exiftool -Subject+="contains-synthetic-performer" *.jpg

This appends the tag to every JPEG in the directory without overwriting existing metadata. Run it against your lifestyle image folder before your next bulk upload.

Verification step: After tagging, confirm the metadata stuck. In ExifTool: exiftool -Subject filename.jpg. You should see contains-synthetic-performer listed under Subject. Amazon's upload scanner checks this field at ingest — if it's missing, your image passes through untagged and you're non-compliant.

What Happens If You Don't Tag

Amazon hasn't published specific enforcement penalties beyond its standard listing suppression framework. Based on the pattern I've seen across Amazon's 2026 image policy enforcement, expect this rollout:

1. Phase 1 (now): Soft enforcement. Amazon scans for the tag but primarily relies on sellers self-reporting. Untagged AI people won't immediately trigger suppression.
2. Phase 2 (likely Q4 2026): Active detection. Amazon's vision models already detect AI-generated content through C2PA metadata and pixel analysis. Adding "synthetic person" detection to this pipeline is a small step.
3. Phase 3: Listing suppression for non-compliance, similar to how off-white backgrounds and text-on-hero violations are handled now.

The $1,000/$5,000 fine structure comes from the New York law, not Amazon. But Amazon's enforcement mechanism — listing suppression — is arguably worse for your business than a fine. A suppressed listing on a $30 AOV product doing 50 units/day costs you $1,500/day in lost revenue.

AI-Generated People vs Real Photography: What the Conversion Data Actually Shows

Here's the question the disclosure rule forces every seller to ask: Is it worth using AI-generated people in my listing images?

The answer is category-dependent, and most sellers get it wrong.

Where AI-Generated People Outperform Real Photography

Product-in-context lifestyle scenes where the person is secondary to the product — a blurred figure using a kitchen appliance, hands holding a skincare bottle, a silhouette jogging with wireless earbuds — are the sweet spot. The person provides scale and context without being the focus. AI handles this well because visual accuracy demands are lower. Nobody is scrutinizing the runner's face; they're looking at how the earbuds sit.

In categories like home goods, kitchen appliances, electronics, and outdoor gear, AI lifestyle images with background figures have tested on par with or slightly above real photography in A/B tests I've run with clients. The reason is speed and volume: you can test 6 scene concepts in a day with AI versus one concept per $3,000 studio session.

Demographic variation at scale is another clear win. If you sell a product used by a wide demographic — different ages, ethnicities, body types — and you want your image stack to reflect that, AI lets you produce 15 model variations for under $50. The equivalent real photography session runs $8,000-$15,000 with multiple models, wardrobe, and studio time.

Where AI-Generated People Tank Your Conversion Rate

Apparel, fashion, and anything worn on the body. This is the danger zone. AI-generated models consistently underperform real photography in these categories — and the data isn't close.

One case study from a controlled test showed conversion rates dropping measurably when switching from real model photography to AI-generated models for clothing. The problem: AI distorts fabric texture, stitching, drape, and proportions in ways that look "off" even when you can't pinpoint why. Shoppers notice. Return rates spiked from 28% to 41% in one apparel brand's test because the AI model made the garment look different from reality.

Beauty and skincare where the model's skin is the product. If you're selling foundation, concealer, or skincare, the model's skin texture IS the demonstration. AI-generated skin looks too perfect — uniformly smooth, uncanny. Real models with real skin texture convert better because they build trust. Shoppers have learned to spot "too perfect" as "probably fake."

Any image where the person IS the selling point. Think fitness equipment where the model's physique demonstrates results, or ergonomic products where you need to show real posture and real body mechanics. AI-generated people can't authentically demonstrate what your product does to a real human body.

The Hybrid Play That Actually Works

The highest-performing approach I've seen across 200+ brands isn't all-AI or all-real. It's this:

1. Hero image (Slot 1): Product only, real photography, no people. This is where compliance risk is highest and conversion impact from people is lowest.
2. Slots 2-4: Real product photography — infographics, detail shots, dimensions. No people needed.
3. Slots 5-6: Lifestyle scenes with context. AI-generated backgrounds around a real product photo. If people appear, they're secondary (hands, blurred figures, partial shots).
4. Slot 7: Social proof, comparison, or warranty/guarantee graphic. No people needed.
5. A+ Content: This is where AI-generated people can work — larger format, lower scrutiny, and the brand story context makes lifestyle imagery with AI models more acceptable.

This structure avoids the disclosure tag for your main image stack entirely (no AI-generated people in slots 1-7), confines AI people to A+ Content where the format supports them, and keeps your image stack hierarchy focused on the product itself.

The Real Cost Math: Disclosure Tag Changes the Equation

Before July 22, the decision to use AI-generated people was purely economic. Now there's a compliance cost and a trust cost to factor in.

The compliance cost is minimal per image — 5-10 minutes of metadata tagging. But at scale (50+ SKUs with lifestyle images across image stacks and A+ Content), that's a meaningful workflow addition. And the ongoing audit burden is real: every time you refresh creative, you need to verify which images contain AI people and ensure they're tagged.

The trust cost is the bigger variable. Amazon will display a disclosure indicator on listings with tagged images. We don't yet know exactly how this label will look or where it will appear. But the precedent from other platforms (Instagram, YouTube) suggests it will be visible enough that some shoppers will notice.

Here's the math that should guide your decision:

The row most sellers should focus on is the last one: real product photography composited into AI-generated lifestyle scenes with no people. No disclosure tag needed. No compliance risk. No trust cost. And the conversion data backs it up — lifestyle context around a real product outperforms product-only shots by 5-15% without any of the regulatory overhead.

If your category genuinely benefits from showing people (apparel, baby products, fitness), invest in real model photography. The $200-600/image cost is a rounding error compared to the revenue impact of a conversion rate that's 15-20% higher than an AI alternative that now carries a disclosure label.

Common Mistakes Sellers Will Make With the New Disclosure Rule

I'm already seeing these in client audits. Avoid all five.

1. Not auditing existing images. If you uploaded AI-generated lifestyle images with people before July 22, they're non-compliant. Amazon hasn't set a retroactive deadline, but the rule applies to all images on the platform, not just new uploads. Audit your catalog now.

2. Tagging images that don't need it. Over-tagging wastes time and puts unnecessary disclosure labels on your listing. If a real model posed for your photo and you used AI to swap the background, that does NOT require the tag. Only fully AI-generated people require it.

3. Using the wrong metadata field. The tag must go in the XMP dc:subject field (which maps to "Keywords" or "Tags" in most image editors). Putting it in the Description, Caption, or Comment field doesn't work — Amazon's scanner reads the subject field specifically.

4. Forgetting A+ Content images. Sellers remember to tag their 7 listing image slots but forget that A+ Content modules, Brand Story images, and product videos also fall under the requirement. Any visual content on your listing with an AI-generated photorealistic person needs the tag.

5. Assuming the rule only applies to US listings. Amazon is enforcing this globally, across all marketplaces. Even if New York's law technically only applies to advertisements reaching New York consumers, Amazon's implementation is worldwide. If you sell on Amazon UK, DE, JP, or any other marketplace, your images need the tag.

What This Means for Your Creative Workflow Going Forward

The synthetic performer disclosure rule is the first of what will be many AI transparency requirements hitting ecommerce. California, Illinois, and the EU all have AI disclosure legislation in various stages. Amazon, as usual, is implementing a single global standard ahead of the regulatory patchwork.

For your creative production workflow, this means three practical changes:

Add a metadata tagging step to your image QA process. Before any image goes to Seller Central, it passes through a compliance check: Does it contain an AI-generated person? If yes, is it tagged? This takes 30 seconds per image with a documented workflow.

Shift AI investment toward scenes, not people. The highest-ROI use of AI in product photography has always been environment generation — placing real products in AI-generated kitchens, bathrooms, offices, and outdoor settings. The disclosure rule makes this advantage even clearer. AI backgrounds with no people = no tag required, no label on your listing, and no compliance risk.

Build a model photography relationship for categories that need people. If you sell apparel, beauty, baby products, fitness equipment, or anything where showing real human use is critical to conversion, now is the time to lock in a reliable photography partner. The sellers who default to AI-generated people because it's cheaper will now carry a visible disclosure label. The ones who invest in real photography will not. That's a competitive advantage you can buy for $200-600 per image.

Should I Use AI-Generated People in My Amazon Listing Images?

Only if the person is secondary to the product, your category doesn't require body-accurate representation, and you're willing to carry the disclosure label. For most sellers, AI-generated scenes without people deliver the same lifestyle context at the same cost without the compliance burden.

How Do I Know If My Existing Images Have AI-Generated People?

Check your creative production records. If you used Midjourney, DALL-E, Stable Diffusion, Flux, or similar tools to generate any image that includes a photorealistic human figure, it needs the tag. If you're unsure, err on the side of tagging — an unnecessary disclosure label is better than a suppressed listing.

Does This Apply to Amazon's Own AI Creative Studio?

Yes. If you used Amazon's AI Creative Studio to generate lifestyle images that include photorealistic people, those images need the contains-synthetic-performer tag. Amazon's own tools are not exempt from the disclosure requirement.

Will Amazon Automatically Detect Untagged AI-Generated People?

Not yet at scale, but they will. Amazon already uses C2PA metadata scanning and pixel-level AI detection for other image compliance checks. Adding synthetic person detection to this pipeline is technically straightforward. The current reliance on self-reporting is a grace period, not a permanent policy.

Does a "Disclosure Label" on My Listing Hurt Conversion?

We don't have Amazon-specific data yet since the rule just launched. But early data from other platforms suggests the impact is small (1-3% conversion dip) for products where the person is contextual, and larger (5-8% dip) for products where the person is central to the purchasing decision. The safest play is to avoid the label entirely by using real photography for people and AI for everything else.

The Three Actions to Take This Week

1. Audit your catalog. Pull every listing image and A+ Content asset. Flag any image containing an AI-generated photorealistic person. If you used the AI lifestyle workflow I've documented, check your Midjourney and Flux outputs specifically.

2. Tag what needs tagging. Use the ExifTool batch method if you have more than 20 images. Use the Windows Properties method for smaller catalogs. Verify the tag stuck before re-uploading.

3. Shift your creative strategy. For future productions, default to AI scenes without people and real photography when people are needed. This isn't just about compliance — it's about conversion. The hybrid approach where AI handles environments and real photography handles products and people consistently outperforms full-AI at a fraction of the cost of full-studio.

The sellers who treat this disclosure rule as a minor metadata chore will miss the strategic signal. Amazon is telling you — through policy, not just best practices — that AI-generated people carry a cost that AI-generated scenes do not. Build your creative workflow around that reality and you'll outperform the sellers who are still figuring out how to add an XMP tag six months from now.