The Counterfeit Was Called Blue 365 Cream

TL;DR

• A shopper who bought Aestura cream on Coupang in June received counterfeits; the listing showed "Aestura cream" as a single item and "Blue 365 Cream" as a discounted multipack. A brand-name keyword monitor never sees the second listing.
• Sellers scrape genuine product pages and use generative AI to mass-produce descriptions, reviews and promo videos, so every text signal in a listing is under the counterfeiter's control. Unauthorized overseas K-brand trademark registrations rose from 4,045 in 2023 to 10,020 in 2025.
• The fake still has to look like the product. Shade, typeface, back-panel layout and logo geometry are vision problems, and Temu's monitoring database reportedly draws on 47 million images against 9.5 million keywords.
• Image matching identifies a listing; a test purchase identifies the route. The parcel carries the shipping origin, carrier, payment counterparty, packaging and lot codes that no product page contains.

Kim Cho-hee bought five containers of Aestura soothing cream on Coupang in June. She had bought the same cream five months earlier, so she had something to hold it against, and the comparison was not subtle. The blue of the containers was a different shade. The typeface was wrong. There was Chinese writing on the back.

She got a refund. When she went back to the listing, the seller was gone.

The part that should interest anyone running brand protection is how the listing was written. Select the product as a single item and it read "Aestura cream." Select the heavily discounted multipack and the same product was called "Blue 365 Cream." Aestura is known for its blue containers.

A keyword monitor watching for the string "Aestura" never sees the multipack.

The text stopped being evidence

That rename is not a clumsy dodge. It is the whole method, and the rest of the listing is built to match.

Korean reporting describes online sellers scraping the genuine product pages, then using generative AI to mass-produce multilingual descriptions, fake customer reviews and promotional videos at whatever volume they need. The goods then move in small quantities through social media and YouTube broadcasts rather than in bulk through a single storefront. The Trademark Police Division under Korea's Ministry of Intellectual Property arrested the head of a Vietnam-based operation running exactly this playbook, with some 3 billion won — about $2.2 million — in counterfeit goods behind it.

Consider what that leaves for a text-based detection system to work with. The product name is a string the counterfeiter chose and can change at will. The description is machine-generated and unique to each listing, so it matches no fingerprint. The reviews are synthetic and positive. The seller account is disposable, as Kim Cho-hee found when she went looking for it.

The same pressure is showing up one level higher, on the trademarks themselves. Suspected unauthorized overseas trademark registrations tracked by the Korea Intellectual Property Protection Agency went from 4,045 in 2023 to 10,020 in 2025. Squatting a K-brand name in a foreign register attacks the identifier from the opposite end: now the counterfeiter has a registration to point at, and the string in the listing is arguably theirs.

Every text signal in the chain is under the counterfeiter's control.

What the fake cannot hide

The one thing a counterfeit still has to do is look like the product. That is not a preference — it is the entire commercial proposition. A fake that does not resemble the original is unsellable.

And that is where these fakes fail. The Aestura counterfeit was caught on shade, typeface and back-panel text, by a shopper with a genuine unit on the table. FugenBio, which makes the skincare brand cepoLAB, has published side-by-side photographs so consumers can do the same comparison themselves. In China, a store called OnlyYoung borrowed the logo and sign colours of the Korean retailer Olive Young — a visual imitation, not a textual one, and one that no keyword list would ever catch.

Those tells are all vision problems. Shade of blue, letterform, layout of a back panel, the geometry of a logo: this is what image models are genuinely good at, and it is the one category of signal the counterfeiter is economically forced to leave in place.

The platforms appear to know it. When Temu signed a cooperation agreement with KOIPA in July, the release described its proactive monitoring database as covering more than 15,000 brands and drawing on "over 47 million images and 9.5 million keywords." Five times as many images as keywords. Coupang's own three-stage process names artificial intelligence and monitoring as the detection layer, sitting between seller onboarding checks and post-report takedown.

That is the right direction. It is also where most programmes stop, and stopping there costs more than it looks.

A picture identifies the listing. Buying one identifies the route.

Image matching tells you a listing is probably counterfeit. It tells you almost nothing about who made it, where it shipped from, which payment rail it settled on, what else that operation is selling, or which other brands it is hitting at the same time. Those facts are not in the listing, because the listing is marketing copy written by the seller — and increasingly written by a model on the seller's behalf.

They are in the parcel.

A test purchase is the only step that converts a suspicious listing into physical evidence. What arrives carries a return address, a shipping origin, a carrier, a payment counterparty, packaging, batch and lot markings, and an actual object that can be photographed, measured and sent to a lab. Kim Cho-hee's five containers were an accidental test purchase, and they produced more usable information than the listing ever held — including the country-of-origin tell on the back panel. The listing itself evaporated within weeks.

Singapore's Health Sciences Authority ordered a recall of a product sold as "Medicube Pink Collagen Capsule Cream" after a cancer-causing substance was found in it; APR, which makes the genuine product, then determined the recalled item was counterfeit. Nothing in that sequence was visible from a product page. It required somebody to have the physical unit and test it.

Coupang's stated next step points the same way, in careful corporate phrasing: the platform wants to expand cooperation so that "product identification data and distribution characteristics held by brands can be factored into the detection process." Distribution characteristics are not observable from a listing. A brand only learns those characteristics by handling goods that came off the network, which means buying them.

Run enough of those purchases across enough listings and the individual results start connecting. Two listings with different seller names, different product names and separately AI-written descriptions ship from the same warehouse, or refund through the same account, or arrive in packaging from the same converter. That is the point at which you stop removing listings and start seeing an operation. The detection layer finds the candidates; the purchases draw the graph.

What is actually being counted

Look at how the progress in this story gets reported. Counterfeit products blocked from online platforms worldwide: 160,000 in 2023, 210,000 in 2025. Blocked listings of counterfeit Korean-brand cosmetics on overseas platforms: 16,774 in 2023, 36,116 in 2025. Korea's Intellectual Property Ministry opened a reporting centre in July and will launch a brand certification system in October.

Every one of those figures counts listings. Not sellers, not shipments, not networks.

A listing is the cheapest thing in this entire system to replace. The seller who sold Kim Cho-hee five counterfeit creams did not need to rebuild a factory or find a new supplier when the account disappeared; a new listing costs a new name, and the name is already a variable — "Aestura cream" one day, "Blue 365 Cream" the next. Blocked-listing counts can rise steadily while the number of operations behind those counts stays flat, and the metric gives no way to tell those two worlds apart.

The OECD puts the global trade in Korean brand counterfeits at roughly 11 trillion won a year as of 2024, against about 7 trillion won of direct sales losses and some 14,000 jobs. Set that against the takedown totals and the mismatch is the story. Nobody in the public numbers is counting how many distinct operations sit behind 210,000 removals — whether it is thousands or a few dozen running at scale.

That number exists. It is just not recoverable from listings, which is the only place anyone is currently looking. It is recoverable from parcels: from what the packages have in common when you buy across the network and compare what shows up.

References

• Korea JoongAng Daily, Kang Bo-Hyun, "What's in a name? Less than shoppers might think as counterfeiters cash in on Korean brand popularity", 10 September 2026: https://www.koreajoongangdaily.com/business/whats-in-a-name-less-than-shoppers-might-think-as-counterfeiters-cash-in-on-korean-brand-popularity/12866647
• Seoul Economic Daily, Kim Sun-young, "Coupang Pilots Brand Hotline to Fight Counterfeit K-Beauty", 13 September 2026: https://en.sedaily.com/finance/2026/09/13/coupang-pilots-brand-hotline-to-fight-counterfeit-k-beauty
• Temu and the Korea Intellectual Property Protection Agency, "Temu Signs MOU With Korean Intellectual Property Protection Agency to Strengthen IP Protection for Korean Brands", 1 July 2026: https://www.prnewswire.com/apac/news-releases/temu-signs-mou-with-korean-intellectual-property-protection-agency-to-strengthen-ip-protection-for-korean-brands-302814769.html
• Europol and EUIPO, "Intellectual Property Crime and its Link to Other Serious Crimes: Focus on Poly-Criminality", June 2020: https://www.europol.europa.eu/publications-events/publications/intellectual-property-crime-and-its-link-to-other-serious-crimes-focus-poly-criminality

Image: "Myeongdong street" by Izzatfikry99, licensed under CC BY 4.0, via Wikimedia Commons; cropped. Myeongdong in Seoul is the retail district most closely associated with the K-beauty brands now being copied abroad.

Source

• Korea JoongAng Daily: What's in a name? Less than shoppers might think as counterfeiters cash in on Korean brand popularity