Study finds ChatGPT, Gemini and Perplexity steer shoppers to fake retail websites

Shopping bots recommended impostor sites

AI chatbots regularly send shoppers to fake or otherwise risky retail websites, according to a study by the brand-protection firm Podqi that was shared exclusively with The Verge. Testing ChatGPT, Gemini and Perplexity on shopping questions, the firm found all three cited or displayed sites that impersonate real retailers, sell counterfeit goods, or occupy domains that could be exploited by third parties.

Across 6,795 collected answers, Perplexity surfaced a risky site in 15 percent of responses, the worst showing of the three. ChatGPT did so in 8 percent of answers, while Google's Gemini performed best at 4 percent, according to The Verge.

How the test worked

Podqi ran its prompts over two testing periods in September, covering 151 brands from large companies such as Lululemon, Birkenstock and Le Creuset to smaller labels including Stüssy and Madhappy, the latter being a Podqi client. Questions took forms such as asking which websites are best for buying a given brand, where a brand can be found online, and whether a brand ships internationally or runs regional websites.

The firm wrote its own prompts rather than drawing on real user conversations, and tracked responses using a tool that monitors chatbot answers across services. A site was classed as risky if it impersonated a legitimate retailer, sold knockoffs, was an expired brand domain since re-registered by someone else, or was a dead domain that another party could purchase and repurpose. A mention counted whether the model recommended a risky site, displayed one without comment, or warned the user about it.

Certain question types proved more dangerous than others. Queries about regional brand websites triggered risky results more often than other categories, and the chatbots at times invented regional sites for brands that have none. In one exchange reported by The Verge, Perplexity pointed to the wrong page as Stüssy's official website and described it as the best starting point for guaranteed authentic merchandise.

Warnings and the small-brand gap

ChatGPT stood out in one respect: when it surfaced a risky site, it attached a caution to the user 24 percent of the time, more consistently than its rivals.

Brand size also mattered. Queries about larger companies produced risky sites roughly half as often as queries about smaller brands, suggesting that less-documented retailers are more exposed to being misrepresented in chatbot answers, and that shoppers hunting for niche labels face a higher chance of being misled.

Company responses

Google said it takes information quality seriously and was pleased that Gemini performed best among the products tested. Spokesperson Jessica Levis Boord added that the company continually tests and improves its models and displays links to sources alongside Gemini's responses so people can verify answers for themselves.

OpenAI spokesperson Taya Christianson said the company's goal is for ChatGPT to connect people with reliable places to shop, and that it is continuing to strengthen safeguards so potentially deceptive sites are less likely to appear in shopping-related results.

Perplexity spokesperson Jesse Dwyer countered that Perplexity has been the most accurate AI on credible research benchmarks for four years, while acknowledging the company does not claim perfect accuracy.

Podqi CEO and cofounder Trevor West said the broader aim is consumer awareness: even when a chatbot presents a result, people should verify the source before entering personal or credit card information.

Why it matters

AI companies are actively marketing chatbots as shopping assistants for researching, comparing and discovering products and retailers. As The Verge notes, more consumers are expected to lean on them this holiday season, often for brands and stores they do not already know. When a chatbot confidently labels an impostor site as an official source for authentic goods, what was a brand-protection nuisance becomes a concrete consumer hazard: phishing, payment fraud and counterfeit purchases. The study's figures also suggest the risk is unevenly distributed, with smaller brands and their customers hit hardest. For shoppers, the practical takeaway is to treat a chatbot's retail recommendation as a lead to verify, not a vetted store.