SnapDragon uses AI to hunt counterfeit listings, but says sellers can return, sometimes within hours

An Edinburgh company, SnapDragon, uses AI to search online listings for signs of counterfeit goods, according to a BBC News report published on 26 September. The same report says a takedown may not last: SnapDragon says sellers can return with new accounts or on different platforms, "sometimes within hours".

The 90-second clip, headlined "Using AI to detect fake goods online", is described by the BBC as being about "How brands are using new tools to try and keep up with scammers." It comes from an episode of the BBC programme Tech Now called "Hunting the Fakes", in which, according to the BBC's listing, Ammie Sekhon "explores the tech being deployed to identify and remove counterfeit products online".

What the BBC report shows

The clip opens by comparing an authentic item with a fake, in a report that centres on copies of products from the brand DFYNE; the BBC's thumbnail shows black leggings with a hole in the back. One speaker says the authentic item "has stretch", while the copy "has created a hole and burst" after a single wear. A speaker says that on any platform, "within five minutes you'll get an ad for a DFYNE-like product".

The report's narration puts the difficulty plainly: the fakes can be easy to find online, but "Working out who's behind them is much harder."

How SnapDragon says it works

In the clip, a speaker describes SnapDragon's search as built on "key terms, which might be word terms", adding: "They can be misspellings, product descriptions, images, logos."

SnapDragon's own website describes the same approach in more detail. It lists several AI modules, including:

• Text analysis, which it says uses "Natural Language Processing (NLP) to identify listings of relevance to your brand".
• Logo detection, which it says "Searches through hundreds of thousands of images every single day looking for matches".
• Optical character recognition, which it says examines listing images "looking for snippets of text in those images and symbols".
• Seller network analysis, intended to "Identify repeat offenders" and "Build evidence that links them to multiple accounts".

The company presents this as a mix of software and people, describing "Automated risk intelligence meets human-led enforcement" and a "Hybrid AI + human workflow". On its homepage it claims an "89% enforcement rate", which that page does not define, and says it has enforced "the removal of hundreds of thousands of infringements".

Sellers adapting

Much of the report is about sellers adapting too. A speaker says sellers "blur the logo on the products so that they can avoid our AI detection", and points to a listing where "they misspell DFYNE". The narration adds: "And the deception is changing again."

In one case shown, a speaker says a seller's videos were identified as AI-generated from their metadata, and confirms that this led to a successful takedown. But the report adds that such takedowns may not last, since SnapDragon says sellers can reappear on new accounts or other platforms, sometimes within hours.

One speaker says it is "probably fair to say that counterfeiters are one step ahead of the game, and they're always innovating". Near the end, the narration says: "As AI gets better at finding fakes, counterfeiters are using it too."

What's not known

The clip gives no figures for how many fake DFYNE listings were found or removed, and its captions do not name the people speaking. SnapDragon's detection and enforcement figures come from its own marketing pages and have not been independently verified here. The full Tech Now episode, which the BBC says is 24 minutes long, may carry more detail than the clip.

On our reading, the clip describes a contest rather than a fix: the detection methods it shows are met by workarounds it also shows or describes, from blurred logos and misspelt brand names to sellers returning on new accounts.