How to Spot AI-Generated Products Before You Buy
AI-generated listings sell objects that were never made. Here are the tells, in the image, the copy, and the shop itself.
Check three things: the image, the release rate, and the catalogue breadth. AI-generated product photos fail at repeated detail - hands, teeth on a zip, evenly spaced stitches, symmetrical hardware, readable text. A shop posting dozens of intricate new designs every week is not making them. And a "maker" whose catalogue spans jewellery, tapestries, mugs and journals in one uniform style is generating, not crafting.
This is now a distinct problem from dropshipping, and it is worse in one specific way. A dropshipper sells you a real factory object described dishonestly. An AI-generated listing can sell you an object that does not exist at all - a rendered image with no physical original, fulfilled with whatever the nearest print supplier can approximate. The gap between the picture and the parcel is not disappointment. It is a different item.
What does "AI-generated" mean on a product listing?
It covers four separate things, and conflating them is why buyers get confused. In rising order of how much it should worry you:
- AI-written copy. The title and description were generated. Common, mostly harmless, but a signal about how much the seller cares.
- AI-retouched photos. A real photo of a real object, cleaned up or background-replaced. Widely accepted, and not misrepresentation on its own.
- AI-generated designs on real products. A prompted image printed onto a genuine mug, poster, or shirt. The object exists. The art was not drawn by a person.
- Fully AI-generated listing images. The photograph itself is generated. No such object was ever made or photographed. This is the one that gets you a parcel bearing no relationship to what you clicked on.
What are the visual tells in an AI-generated photo?
Image generators have got very good at overall impression and remain unreliable at repetition and structure. Zoom in. That is the whole technique.
- Repeated elements drift. Stitches that change size along a seam, chain links that merge, beads that are not the same bead twice, knit or crochet stitches that dissolve into texture in the middle of a row. Real handwork is irregular in a rhythmic way; generated handwork is irregular in a chaotic way.
- Hardware makes no mechanical sense. Zip teeth that do not interlock, a clasp with no hinge, a buckle the strap could not pass through, earring posts attached at an angle nothing could wear.
- Text falls apart. Any lettering on a label, tag, book spine, or packaging. Generators still produce plausible-shaped nonsense at small sizes.
- Hands and fingers. Still the fastest tell in any photo containing a person holding the product.
- Impossible light. Shadows in inconsistent directions, a reflection missing from a glazed or metal surface, a highlight with no light source.
- Too perfect a background. Studio scenes with a flawless, slightly dreamlike quality and no dust, no wonky edge, no domestic mess. Real makers photograph on real tables.
The two-second version
Look at the third or fourth photo rather than the first. Sellers put the most convincing image first. Generated sets often collapse by the third angle, because a generator does not have a consistent object to rotate - it is producing a new image each time, and the details drift between them. If the handle changes shape between photos, there is no handle.
What are the tells in the listing text?
Generated copy tends to describe a category rather than an object. Watch for:
- No specifics that only the maker would know. A real description says the fabric came from a 1970s curtain, that the glaze went patchy in the second firing, that this is the last one from that batch of yarn. Generated copy says "meticulously crafted with attention to detail".
- Measurements missing or vague. Anyone who made the object can measure it. "Approximately medium sized" is not a measurement.
- Adjective stacking. "Exquisite, timeless, unique, artisanal, one-of-a-kind, handcrafted" in a single sentence. Real makers describe; generated copy praises.
- No process, no problems. Handmade work has constraints - a limited amount of material, a repair, a variation between pieces. Copy that mentions none of that is describing an idea, not a thing.
- Identical structure across every listing. Open three products from the same shop. If the paragraphs sit in exactly the same order with the same rhythm and only the nouns change, it was templated or generated.
What are the shop-level tells?
The single most reliable signal is not in any individual listing. It is arithmetic.
- Count new listings against plausible working hours. An intricately illustrated design takes an illustrator days. A shop releasing thirty of them a week is not illustrating. There is no version of that output that a person produces by hand.
- Check catalogue breadth. Genuine makers specialise, because skills and equipment do not transfer. A shop selling enamel pins, tapestries, phone cases, oil-painting prints and embroidered hoodies is a design pipeline attached to print suppliers.
- Look for the person. No name, no face, no workshop, no location, no process photos, no in-progress shots. Real makers show their bench because it is the most persuasive thing they own.
- Look for stylistic fingerprints. Every generator has a house style. If forty "different artists" across a marketplace produce the same glossy, over-rendered, slightly airbrushed look, they share a tool rather than an aesthetic.
- Read reviews for the phrase "not as pictured". This is where fully generated listings finally show up in public.
What do the marketplaces actually allow?
More than most buyers assume. Etsy’s Creativity Standards place "seller-prompted AI art" inside the Designed by a seller production category, alongside original designs made by a third party. AI-assisted work is permitted on Etsy. The rule is about which label it sits under, not whether it may be sold at all.
That is a defensible position for a platform of Etsy’s scale, and it puts the burden of checking squarely on you. It also means the word "handmade" on a large open marketplace is doing less work than it appears to. We break down how those production labels read in what Etsy production labels really mean, and the wider 2026 rule changes in Etsy’s 2026 policy changes.
Does anyone actually mind?
The measurable answer is yes, consistently, across independent surveys. A June 2026 study of 1,017 US adults by SegmentOS found that when shown two identical products at the same price, 68% chose the one labelled as made by humans, and 54% said they would pay more for human-made products. Gartner reported in May 2026 that half of US consumers would prefer to give their business to brands that do not use generative AI in customer-facing content. Research from Klaviyo found only 7% of shoppers trust a brand more for visible AI-generated marketing, while 31% trust it less.
Worth being precise about what that evidence covers. Those studies measure attitudes to AI in marketing, content and service. We have not found a credible survey measuring a price premium specifically for handmade physical goods, and we are not going to invent one. What the data supports is narrower and still useful: when people can see the difference, most of them choose the human.
A generated image is a description of an object. It is not evidence that one exists.
What should you do about it?
Practically, four habits cover almost everything:
- Zoom into the second and third photos before you read a word of the description.
- Reverse image search anything you are unsure about. This still catches both generated images and stolen catalogue photos - the full method is in our dropshipping checklist.
- Message the seller and ask a question only someone holding the object could answer.
- Buy where the marketplace has already done the checking, so authenticity is not a research project you conduct on every purchase.
That last one is the structural fix. On SCRAPD a person reviews every storefront before it opens, every piece is one-of-one, and nothing is AI-generated - our reasoning is in why never AI-generated matters, and the positive signals of real craft are in how to tell if something is really handmade. For more on finding trustworthy shops anywhere, see where to buy genuinely handmade online and the rest of our buyer guides.
How can I tell if a product photo is AI-generated?
Zoom into repeated details - stitches, chain links, zip teeth, beads - and check whether they stay consistent. Look for text that dissolves into nonsense, hardware that could not mechanically work, and shadows without a light source. Then compare the third and fourth photos: if details change between angles, no single object was photographed.
Is it against the rules to sell AI-generated products?
Not on most large marketplaces. Etsy’s Creativity Standards explicitly place seller-prompted AI art under the "Designed by a seller" production category, so AI-assisted work is permitted as long as it is attributed correctly. The responsibility for checking what you are buying largely falls to the buyer.
What is the difference between an AI-generated listing and dropshipping?
A dropshipper sells a real factory-made object while describing it misleadingly. A fully AI-generated listing may sell an object that was never made at all, with the image produced by a model and the order fulfilled by whatever a print supplier can approximate. The second usually produces a bigger gap between the photo and the parcel.
Do buyers care whether a product was made by a human?
Independent surveys say most do. A June 2026 SegmentOS study of 1,017 US adults found 68% chose a product labelled as human-made when shown two identical options at the same price, and 54% said they would pay more for human-made products.
What should I ask a seller if I suspect an AI listing?
Ask for a fresh photo of the actual item next to something ordinary, such as a coin or a handwritten note with today’s date. A maker with the piece on their bench can send it within minutes. A shop selling a generated image cannot send it at all.