AI Chatbots Sent Shoppers to Fake Stores in Up to 15% of Answers, Study Finds

Photo: Hippopx (public domain). Illustrative image.

What happened. Ask an AI chatbot where to buy a brand online and the answer can include a fake store, a counterfeit seller, or an old brand domain that now belongs to a stranger. That is the finding of a study by Podqi, a company that protects brands’ intellectual property, shared exclusively with The Verge and reported on 12 November 2026. Podqi asked ChatGPT, Google’s Gemini and Perplexity the questions shoppers actually type — which websites are best for buying a given brand, where to find it online, whether it ships abroad through regional sites — across 151 brands, from Lululemon, Birkenstock and Le Creuset to streetwear labels such as Stüssy and Madhappy. Across two testing rounds in September it collected 6,795 answers, and about 9% of them exposed a shopper to a risky site: 15% of Perplexity’s answers, 8% of ChatGPT’s and 4% of Gemini’s.

What counts as risky. Podqi’s definition was broad by design. A site qualified if it impersonated a real retailer, sold counterfeits, was an expired brand domain re-registered by someone else, or was a dead site that anyone could buy and reuse for fraud. Of the 4,437 domains in the answers, 718 were flagged: 351 fake and 367 dead. The worst single case Podqi documented was Perplexity naming an incorrect page as Stüssy’s official website and describing it as a source of authentic products. Shoppers asking about smaller brands were the most exposed: queries about them surfaced risky sites roughly twice as often as queries about big names.

Recommended more often than warned against. The uncomfortable part is not just that risky sites appear — it is how they appear. Podqi says they were recommended about 6.3 times more often than they were warned against, and that in 57% of the answers that included one, the chatbot actively suggested it for shopping. ChatGPT was the most cautious of the three and still attached a warning to only 24% of the risky sites it surfaced. There was one improving trend: the overall rate fell from 12% in the first September round to 7% in the second. The timing of the study matters for readers: Black Friday falls on 27 November and Cyber Monday on 30 November, when millions of shoppers will visit unfamiliar stores — exactly the situation these answers feed into. It is the consumer end of the problem our guide to AI safety describes: confident answers are not the same as checked ones.

What the companies say. All three companies responded to The Verge. Google said it was pleased Gemini recorded the lowest rate in the test and pointed to the source links Gemini displays, which let people check its answers. OpenAI said it wants ChatGPT to direct shoppers to trustworthy retailers and is working to reduce how often deceptive sites appear in shopping results. Perplexity pushed back through spokesperson Jesse Dwyer, defending the company’s record on accuracy benchmarks and noting it has never claimed to be 100% accurate.

The caveats. This is one firm’s test, and the firm sells brand protection, so it benefits when impersonation looks like a serious threat. Podqi wrote its own prompts instead of analysing real user conversations; one of the 151 brands tested, Madhappy, is a Podqi client; and its headline rates count a risky site that was merely mentioned — even one the chatbot warned about — the same as one it recommended. The study also did not track whether anyone clicked a risky link, bought anything, or lost money, and it captures models that change every few weeks. None of that makes the finding disappear — 718 flagged domains is a concrete count, not a vibe — but it does mean the percentages describe this test, not a universal error rate for AI shopping. Our standing advice applies before you trust any answer with your money: evaluate the tool on your own real tasks first, and learn how to spot AI-generated media before the sale season makes fakes profitable.

What shoppers should do. Podqi’s chief executive, Trevor West, gave the simple version: always verify the source before entering personal or payment details. In practice that means typing the brand’s own address yourself instead of following a chatbot’s link, checking the domain for misspellings and odd extensions, being extra careful with “regional” stores and deals that look too good, and paying by card so a bad purchase can be disputed. The chatbots are useful shopping assistants. This study is a reminder that a link inside a confident answer has not been vetted just because the sentence around it sounds sure.

Sources

• Podqi study of AI shopping answers (151 brands, 6,795 answers, September 2026), shared exclusively with and first reported by The Verge, 12 November 2026: https://www.gadgetreview.com/ai-chatbots-are-recommending-fake-shopping-websites-to-users-new-study-finds
• AI Europe — “Podqi study finds ChatGPT, Gemini and Perplexity send shoppers to fake stores” (includes the 718-domain breakdown and the 6.3x recommendation figure): https://www.europesays.com/ai/238781/
• The Outpost — “AI chatbots point shoppers at fake stores, study finds”: https://theoutpost.ai/news-story/ai-chatbots-keep-pointing-shoppers-at-fake-stores-a-study-finds-33488/
• FreshSynapse — “What Is AI Safety? A Plain-English Guide to the Risks, the Research and the Debate”: https://freshsynapse.com/articles/what-is-ai-safety/