2/3 of Branded Beauty on Resale Sites May Be Fake—AI Test Shows Promise

Retailers and marketplace operators know that counterfeit beauty erodes buyer trust and increases disputes, returns, and safety complaints. Grover Lab's experiment with Google Gemini 3.6 Flash is a practical stress test of whether e-commerce platforms could add AI verification at listing or checkout. Given an estimated two-thirds of branded cosmetics on eBay, TikTok Shop, and Vinted are fake, this matters for category integrity and conversion.

The key development is an exploratory field test by Grover Lab, a research group that builds low-cost tools for detecting fake medicines, which applied Google's Gemini 3.6 Flash model with Thinking enabled to a small set of Rhode Peptide Lip Tint packages. The lab purchased authentic and counterfeit versions of the product, photographed six views per item—four long sides of the cardboard box and the front and back of the tube—and asked the model a direct question: 'Is this tube of Rhode Peptide Lip Tint authentic?' The experiment is a proof of concept for a consumer-friendly counterfeit screen, and it lands in a market where studies estimate that two-thirds of branded makeup and skincare products sold through eBay, TikTok Shop, and Vinted are fake.

“

Given an estimated two-thirds of branded cosmetics on eBay, TikTok Shop, and Vinted are fake, this matters for category integrity and conversion.

The context is a long-running problem with serious health and economic stakes. Counterfeit cosmetics are not merely inferior copies; they can contain heavy metals, bacteria, and other contaminants that pose direct risks to users. Meanwhile, the fraudsters behind these products have become skilled at reproducing packaging details, making fakes virtually indistinguishable from genuine items at a glance. The standard countermeasure has been a mix of expert inspection, brand authentication programs, and occasionally laboratory testing. Grover Lab's question is whether widely available multimodal models can shift some of that inspection burden to ordinary buyers using just a phone camera and a chatbot.

The market implications are significant even if the results are preliminary. For brands, a cheap, accessible authentication step could reduce brand dilution and safety complaints, but it also raises the stakes: if a general AI model can identify fakes, counterfeiters may use the same technology to close the gap. For marketplaces, integrating AI verification into listings or post-purchase checks could reduce disputes and improve trust, but it would require reliability thresholds and privacy controls. For consumers, the promise is appealing—before buying a discounted lip tint, snap a few photos and get a quick authenticity read. However, the experiment involves only three packages of a single product and one model, and the provided excerpt does not include the full verdicts or confidence scores, so it cannot be treated as validation.