An IP protection firm conducted research examining how three major AI chatbots handle shopping-related inquiries, testing responses across 151 brands ranging from small fashion labels to larger homeware companies. The study involved over 6,700 responses generated during testing in September, with researchers asking the chatbots questions about where to purchase specific brands and their regional websites.
Results showed that all three AI systems occasionally directed users toward problematic websites, defined as counterfeit retailer sites, reregistered expired domains, and inactive pages. Google’s Gemini showed the lowest incidence rate at 4 percent of answers containing risky sites, while ChatGPT appeared in the middle range at 8 percent. Perplexity had the highest rate at 15 percent of answers. However, ChatGPT demonstrated the most user protection by providing warnings about risky sites in 24 percent of cases where they were surfaced. Smaller brands appeared more vulnerable to this issue, with queries about larger companies encountering risky sites roughly half as often as those about smaller retailers.
The findings carry implications as consumers increasingly turn to AI assistants for shopping guidance, particularly heading into the holiday shopping season when purchasing activity typically peaks. Researchers noted that certain prompt types triggered risky results more frequently, such as questions about regional brand websites. In some cases, AI systems recommended fake regional websites for brands that have no actual regional presence.
Both Google and OpenAI indicated commitment to improving their systems. A Google spokesperson highlighted Gemini’s performance and noted the inclusion of source links to aid verification. OpenAI stated it continues enhancing safeguards to reduce the appearance of deceptive sites in shopping contexts. A Perplexity representative emphasized the company’s focus on accuracy, though acknowledged that the systems are not error-free.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI