AI2 views

AI Chatbots Risk Recommending Fake Stores in Product Searches

A new study testing ChatGPT, Gemini, and Perplexity found that AI chatbots can steer users toward fraudulent online stores when answering product inquiries. Researchers queried the three systems about 151 brands and analyzed 6,795 total responses to identify risky recommendations.

Perplexity displayed potentially dangerous websites in 15% of responses, significantly higher than ChatGPT's 8% and Gemini's 4%. However, ChatGPT partially mitigated this risk by warning users about sketchy sites in 24% of cases—the highest warning rate among the three. Gemini and Perplexity offered fewer safety alerts, meaning users could unknowingly visit fraudulent retailers without being alerted to the danger.

The findings highlight a growing vulnerability in relying on chatbots for shopping recommendations. Even minor percentages translate to substantial numbers of problematic suggestions across millions of queries, making this a legitimate concern for both users and the platforms themselves.