All three platforms let you use AI in product images, but none of them lets AI change the product itself. Google Merchant Center wants AI-made images to keep the metadata tag that says so, and the main image must still show the exact item you sell. Amazon’s main image has to be a real photo of the product on a pure white background, and since July 2026 it wants a metadata tag on any photorealistic AI-generated person. Instagram doesn’t make you label AI still images, but it may add an “AI info” label on its own.

AI image tools have made a ₹20,000 product shoot feel optional. Upload one phone photo of your kurta or pickle jar, type “on a wooden table near a window, morning light”, and you get something that looks like a catalogue. For a small seller listing a few hundred products, that’s tempting. The trouble is that each platform has its own rules, they changed again this year, and the penalty for getting it wrong ranges from a rejected image to a suppressed listing to an angry customer asking for a refund.
The rule all three quietly agree on
Strip away the platform jargon and every policy comes back to one question: does the picture match what arrives at the customer’s door?
Changing the scene around a real product is mostly fine. Changing the product is where you get into trouble. If AI “improves” the stitching on a handbag, makes a cotton saree look like silk, or adds a lid your steel container doesn’t have, you’ve moved from styling into misrepresentation. In India that isn’t only a platform matter. The Central Consumer Protection Authority’s 2022 guidelines on misleading advertisements apply to e-commerce and social media too, and the CCPA can impose penalties of up to ₹10 lakh (Press Information Bureau). I’m not a lawyer and this isn’t legal advice, but I’d treat “the AI made it look like that” as a very weak defence.
Google Merchant Center: keep the AI tag inside the file
Google is fairly relaxed about AI images in Shopping, with one firm condition. Its image_link help page says all images created with generative AI must contain metadata showing they were AI generated, such as the IPTC Digital Source Type value trainedAlgorithmicMedia, and that you mustn’t remove those tags. It also lists compositeSynthetic for images that mix real and synthetic elements, which is what you get when a real product photo is dropped onto an AI background.
The same page is strict about what the main image shows. It must clearly display the exact item being sold. Line art, mockups and illustrations that don’t depict the physical product get disapproved, except in a few categories like software and vehicle parts. Promotional text, “Free delivery” stickers and added logos on the image are out as well.
There’s an easy mistake to make here. Many image compression tools and WordPress optimisation plugins strip metadata to save a few kilobytes, so a correctly tagged AI image can lose its tag before Google ever crawls it. If your product feed pulls images from your website, check your optimiser’s settings for anything like “strip EXIF” or “remove metadata”, and test one image after upload. Also note that Google has announced a minimum size of 500 x 500 pixels for all product images from 31 January 2027, so don’t let an AI tool export small thumbnails.
Amazon: leave the main image alone
Amazon’s product image requirements in Seller Central are the strictest of the three for the first image. It has to be a professional photograph of the actual product on a pure white background, without props, text, graphics or inset images, and it has to match the product title. A drawing or a rendered mockup doesn’t qualify. Category style guides add more rules on top, and in clothing, for example, the image guidelines spell out that sketches and drawings aren’t allowed as main images. Your logged-in Seller Central version always wins over any blog post, including this one.
The secondary image slots are where AI lifestyle scenes can live, as long as the product in them is your real product, shown accurately.
The newer rule is about people. In July 2026, after a New York law on “synthetic performers” in advertising took effect, Amazon told sellers to add metadata to product images, videos and A+ content that contain photorealistic AI-generated people before uploading them (Quartz report). The tag is the keyword contains-synthetic-performer in the file’s XMP subject field, and Amazon says it may show shoppers an indicator when that tag is present. It doesn’t apply to real models whose photos were retouched with AI, or to characters from films and games. So if your A+ page has an AI-generated “happy customer” holding your product, that file needs the tag. A real model on an AI background doesn’t.
Instagram: the label may turn up on its own
Meta’s rules are lighter for still images. According to its help page on AI labels, Facebook, Instagram and Threads may require an “AI info” label when content has photorealistic video or realistic-sounding audio that was digitally created or altered. That requirement doesn’t apply to images. But an image can still get a label if Meta’s systems detect AI signals in it, and Meta has said it reads the industry C2PA and IPTC markers that tools from Google, OpenAI, Adobe and others embed (Meta newsroom).
In practice that means the same metadata Google wants you to keep is what makes Instagram show the label. I wouldn’t fight it. A small “AI info” tag on a styled flat-lay is far less damaging than a customer finding out in the comments.
There’s an Indian layer here too. The IT Rules amendment notified by MeitY on 10 February 2026, in force from 20 February, requires “synthetically generated information” that appears real to be prominently labelled, and large platforms have to ask users to declare such content (Business Today). Most of the duties sit with the platforms, but it does mean you should answer honestly when an app asks whether your post uses AI.
A made-up Jaipur bedsheet seller
Take a hypothetical block-print bedsheet business in Jaipur that sells on its own Shopify store, on Amazon and through Instagram. The owner shoots each design flat on the floor of the workshop and uses an AI tool to place it on a neatly made bed in a sunlit room.
On Amazon, the floor shot cleaned up on pure white stays as the main image, and the bedroom scene goes into slot three. For Google Shopping, the bedroom image keeps its IPTC tag, which means switching off metadata stripping in the store’s image app. On Instagram the post may get an “AI info” label, and that’s fine. Where it goes wrong is the one design where the AI tool quietly shifted the indigo to a brighter royal blue. Nobody notices until three customers return it, all saying the colour doesn’t match. That’s the check no platform will do for you.
What I’d use AI for, and what I wouldn’t
My honest view: AI backgrounds are a good deal for small sellers, and fully AI-generated products are a bad one. A real product photo on a generated background gives you most of the visual lift at a fraction of a shoot’s cost. Generating the product from a text prompt saves even more time, but you lose control of exactly the details customers buy on, like weave, colour, size and finish. The trade-off is that AI scenes can look a bit too perfect, and some shoppers trust a slightly imperfect real photo more. Try both on a few listings before redoing your whole catalogue.
Before any AI image goes live, I’d run these checks:
- Put the AI image next to the real product in daylight and compare colour, pattern, logo placement, size and any small parts.
- Make sure the scene doesn’t imply things that aren’t included, like a matching pillow cover sold separately.
- Keep the main Amazon and Google images as real photos on clean backgrounds.
- Check the final uploaded file still carries its AI metadata. Free tools like ExifTool can show it.
- Tag any photorealistic AI person for Amazon with
contains-synthetic-performerbefore upload. - Save the original photo and the prompt you used, in case a platform or customer questions an image later.
One last thing. These policies have moved twice in two years, so recheck the official help pages every few months rather than trusting a rule you read once.
If you sell online, our pieces on ChatGPT shopping results for small sellers and Instagram’s original content rule cover the other side of getting your products seen, and testing AI writing tools applies the same check-before-publishing habit to product copy. For help reviewing your own listings, contact the SyncNeck team.
