Study finds chatbots surface risky retail websites

Podqi tested ChatGPT, Gemini and Perplexity with shopping-related prompts about 151 brands.[1] The testing took place in two periods in September 2026.[2] Podqi said it created its own prompts rather than using real user conversations, an important limitation when applying the results to shoppers’ everyday use.[1] The firm analyzed 6,795 answers and reviewed 4,437 domains.[2] Its report was shared exclusively with The Verge.[1] The Verge’s report is dated November 12, 2026.[1]

Across the study, 9 percent of answers showed a risky website.[2] Podqi flagged 718 of the reviewed domains as risky: 351 were fake and 367 were dead.[2] The company defined a risky site as one that impersonated a brand, offered suspected counterfeits, or was dead and unavailable for shopping.[2] Perplexity had the highest rate among the tested engines, showing a risky site in 15 percent of its answers.[1] ChatGPT showed or mentioned one in 8 percent of answers, while Gemini did so in 4 percent.[1]

Podqi counted recommendations, unqualified mentions and warnings as instances in which an engine surfaced a risky site.[1] Among answers that showed one, 57 percent recommended it as a place to shop; 34 percent mentioned it without comment, and 9 percent warned shoppers away.[2] Podqi also found risky sites were recommended 6.3 times as often as they were warned against.[2] ChatGPT warned users about a risky site in 24 percent of the instances in which it surfaced one.[1] These figures describe Podqi’s tested answers, not real-world shopper behavior.[1]

Questions about a brand’s regional websites were the riskiest prompt category: 16 percent of answers showed a risky site.[2] Queries related to larger companies had risky sites in answers about half as often.[1] As one example, Podqi found that Perplexity identified the wrong page as Stüssy’s official website and called it a starting point for guaranteed authentic items.[1] Google said it displays source links alongside responses to help people verify information.[1] OpenAI spokesperson Taya Christianson said the company is working to make potentially deceptive websites less likely to appear in shopping results.[1] The report said the issue could be especially relevant during the upcoming holiday shopping season.[1]