AI Chatbots Won’t Stop Recommending Scam Websites to People Trying to Shop

New research reveals AI chatbots such as ChatGPT, Gemini, and Perplexity continue to recommend risky, scam websites to consumers seeking to shop online. The study, conducted by Podqi and reported by The Verge, tested the leading chatbot platforms on questions related to top retailers and their shipping capabilities, using 151 different brands ranging from small to major companies like Lululemon and Birkenstock.

Perplexity showed risky sites in 15 percent of its answers, while ChatGPT had an 8 percent rate and Gemini the lowest at 4 percent. Although the chatbots sometimes warned users about the risky sites, 57 percent of the answers that showed a risky site still recommended shopping there, and 34 percent did so without comment. Smaller brands were particularly prone to being replaced by scam sites in AI responses.

Podqi CEO Trevor West advised consumers to verify sources before providing personal or financial information. Similar studies have found AI chatbots can also recommend cloned versions of defunct websites designed to steal banking information. The issue of AI's susceptibility to manipulation is further highlighted by a BBC journalist demonstrating how easily he could deceive ChatGPT into making untrue statements about himself.

The prevalence of fake and unreliable information online, including in AI responses, poses a significant challenge in discerning trustworthy content.