AI Chatbots Steer Shoppers to Risky Retail Sites in Up to 15% of Answers
Updated
Updated · The Verge · Nov 12
AI Chatbots Steer Shoppers to Risky Retail Sites in Up to 15% of Answers
3 articles · Updated · The Verge · Nov 12
Summary
6,795 shopping-related chatbot answers tested by Podqi showed risky URLs across ChatGPT, Gemini and Perplexity, with fake stores, hijacked old domains and dead sites appearing in as many as 15% of conversations.
Perplexity performed worst at 15%, ChatGPT hit 8%, and Gemini 4%; ChatGPT warned users about risky sites 24% of the time, more often than the others.
151 brands were included in the September study, and smaller companies were hit harder, with large-brand queries producing risky links about half as often; prompts about regional websites triggered more bad results.
The findings raise consumer-fraud concerns ahead of the holiday shopping season as AI companies push chatbots for product discovery, while Google, OpenAI and Perplexity said they are improving accuracy and safeguards.