Your Product Photo Can’t Prove Anything Anymore

What broke, exactly

A product photo used to carry an implicit contract: this is what you get. The lens pointed at something real, and the file that came out the other end was proof of that. Shoppers didn’t examine each image for signs of fakery. They didn’t need to — fakery was visually obvious. Warped fingers, uncanny shadows, a background that didn’t quite sit right. Your eye caught it without trying.

That contract is gone. In a controlled September 2025 study, adults asked to sort real product photos from AI-generated ones got it right 49% of the time on real images and 52% of the time on AI ones — statistically indistinguishable from a coin flip. Worse: 42% of the same participants said they felt confident or extremely confident in their ability to tell the difference. That’s the actual danger. It isn’t that customers can’t spot AI images. It’s that they’re sure they can, right up until they buy against one that lied to them.

That study is US-sourced — no India-specific version exists yet in the public record — but there’s no reason to expect Indian shoppers are meaningfully better at this than anyone else, and the platform and regulatory context landing on Indian sellers right now is very much local.

Why the platforms won’t rescue you

The instinct is to assume Amazon or Flipkart will draw a hard line and settle this. They haven’t, and on current policy they won’t. Both marketplaces’ 2026 seller rules explicitly permit AI-generated and AI-enhanced product imagery, as long as the image accurately represents the physical product. Background replacement, lighting correction, lifestyle-scene generation around a real unit — all allowed. Fabricating features, scale, or review imagery — not allowed, but that’s a floor, not a trust signal. “Not banned” is doing a lot of work for sellers who assumed “not banned” meant “not competing with.”

Meanwhile the regulatory floor under all of this has moved from theoretical to live. India’s IT Rules amendment on synthetically generated information — SGI, in the ministry’s language — took effect on 20 February 2026, requiring visual labels on AI-generated or AI-modified content, with embedded metadata where feasible. That rule targets the intermediaries offering AI-generation tools, not marketplace listings directly — the connection to your catalog is an inference, not settled law, so don’t take it as “my listing photos must carry a label” today. But the direction is unambiguous.

On the advertising side, ASCI — India’s ad self-regulator — published a draft three-tier AI-disclosure framework on 8 May 2026, with medium-risk content (which explicitly includes synthetic product visualization) requiring a “created/enhanced using AI” disclosure. Public consultation closed 13 June 2026. It is not finalized yet — treat it as coming, not arrived — but it’s aimed squarely at the kind of AI hero shots running in paid ads right now.

The timeline that put you here

Ten months. A camera company and a phone company both shipped hardware that cryptographically proves a photo is real, at the same time a research firm proved shoppers can no longer tell the difference by eye, at the same time two separate Indian regulatory tracks started requiring AI images to say so. None of that is coincidence — it’s the same underlying shift showing up in hardware, research, and law simultaneously.

What it costs you, specifically

Run the numbers on the skincare brand from the opening. ₹18,000 for a half-day shoot covering six SKUs works out to roughly ₹3,000 per hero image, fully loaded — studio time, photographer, retouching. A competitor generating a comparable hero shot with an AI tool spends ₹500 to ₹1,500 per image. On a mobile thumbnail, at 300 pixels wide, in a search results grid, the two images are functionally identical to the shopper’s eye. You spent 2-6x more and the platform gave you zero visible credit for it.

That’s the direct cost. There’s a second, quieter one: 56% of Indian online sellers report using some AI tool in their business already, with 43% using it specifically for listing and content creation. If you’re still shooting everything by hand, you’re no longer the differentiated one — you’re the minority who “didn’t bother with AI,” unless you can prove the shoot happened and that it matters. “We use real photography” stopped being a claim that carries weight on its own. It needs evidence attached to it now.

The AI-image failure mode nobody prices in

There’s a third cost that shows up downstream, not on the invoice. Generative image tools are known to subtly alter garment drape, color, or proportion when producing a “model wearing the product” shot from a flat-lay. If that shift is big enough, two things happen: the platform can reject the listing for misrepresentation under the same accuracy rule that permits AI images in the first place, and — worse — customers who bought against the AI-rendered image get a product that doesn’t quite match what they saw, which shows up in your return rate and your COD reject rate before it shows up anywhere you’re actually tracking it.

This is the actual asymmetry sellers miss: AI imagery isn’t risky because platforms ban it. It’s risky because the failure mode is invisible until a return or a compliance flag makes it visible, and by then it’s already cost you the sale and the trust.

What still counts as proof

Here’s the reframe. You cannot win an argument about which pixels look more real — that argument is over, and chance-level detection settled it. What you can still do is document a process a generated image cannot fake: a named photographer, a dated shoot, a studio location, unedited source files, behind-the-scenes footage. None of that requires a customer to zoom into a JPEG and spot an artifact. It requires you to have kept the receipts.

This is also, not coincidentally, the direction the hardware and the law are both moving. C2PA-capable cameras cryptographically attach exactly this kind of provenance record to a photo at the moment of capture. No major Indian marketplace currently surfaces a Content Credentials badge to shoppers — adoption is effectively nil in the Indian retail pipeline right now, so don’t go chasing a certification with no visible payoff today. But the underlying discipline — name the photographer, date the shoot, keep the RAW files, disclose the AI-assisted ones — is something you can start doing this week on owned channels, ahead of the badge existing anywhere a shopper can see it.

Real photography vs. AI-only: the honest comparison

Neither column wins outright. The table below is what the trade-off actually looks like once you price in compliance risk, not just per-image cost.

The honest reading: AI imagery wins on cost and speed, real photography wins on documentable proof and regulatory safety, and neither wins on visual differentiation alone. The scarce asset isn’t the image. It’s what you can prove about the image.

Should you keep shooting real, go AI, or split the catalog?