AI
Study says shopping AI chatbots recommend fake retail websites
After testing Gemini, Perplexity, and ChatGPT, brand-protection company Podqi found that across 6,795 shopping-related responses, all three chatbots displayed risky URLs, at rates of 4%, 15%, and 8%, respectively. Risky websites included fake retail sites, sites selling counterfeits, re-registered old domains, and inactive sites.
In September, over two testing periods, Podqi tested ChatGPT, Gemini, and Perplexity using shopping questions it designed itself. The questions centered on 151 brands of varying sizes, such as Lululemon, Birkenstock, Stüssy, Madhappy, and Le Creuset. Prompts included asking for the preferred website for buying a brand, online purchasing channels, whether the brand ships internationally, and what regional websites it has.
The study collected 6,795 responses in total. Podqi defined risky URLs as fake websites impersonating real retailers or selling counterfeit goods, old brand domains re-registered by others, and dead sites that are inactive but could be purchased and reused by others.
The results showed that Perplexity performed the worst, displaying risky websites in 15% of responses; ChatGPT was at 8%; and Gemini was the lowest at 4%. “Mentions” in the statistics included active recommendations, display without explanation, and warnings alerting users to risks. ChatGPT warned users about the risky websites it displayed at the highest rate, 24%. Podqi also found that risky websites were more likely to appear in queries related to smaller brands, with major brands at roughly half that rate.
Trevor West, CEO and co-founder of Podqi, advised consumers that even when a chatbot provides results, they should verify the source before entering personal or credit-card information. Google spokesperson Jessica Levis Boord said Google values information quality, continuously tests and improves its models, and displays source links.