AI chatbots recommending fake retail websites: study finds shoppers may face risky links
How the test was conducted
Podqi, an intellectual property protection firm, asked ChatGPT, Gemini and Perplexity shopping-related questions about 151 small and large companies. The brands included Lululemon, Birkenstock, Stüssy, Madhappy and Le Creuset.
The company created its own prompts rather than using real user conversations. It ran the test during two periods in September and tracked chatbot responses with a tool designed to monitor AI answers.
Risky links appeared across platforms
Across 6,795 responses, all three chatbots cited or displayed risky sites. Podqi defined them as fake retailer websites, knockoff sellers, old brand domains registered by someone else, or inactive sites that could be reused.
Gemini had the lowest rate, with risky sites linked or mentioned in 4 percent of answers. ChatGPT’s rate was 8 percent, while Perplexity’s reached 15 percent. ChatGPT warned users about a surfaced risky site in 24 percent of instances.
Shopping prompts and consumer risks
Podqi found risky sites appeared more often when prompts asked about regional brand websites than in other prompt categories. The results suggest that chatbot shopping guidance may direct users toward unreliable destinations.
The issue could matter more during the holiday shopping season, when people look for available products and lower prices. Podqi CEO and cofounder Trevor West said shoppers should verify a source before entering personal or credit card information.
Companies respond to findings
Google spokesperson Jessica Levis Boord said Gemini performed best among the products tested and that Google works to test and improve its models. She added that Gemini identifies itself as AI and displays source links to help users verify information.
OpenAI spokesperson Taya Christianson said the company aims to connect people with reliable places to shop and is improving safeguards against deceptive sites in shopping results. She said the company welcomes examples from researchers that can help improve the experience.
Key points
- Podqi found risky websites in answers from all three tested chatbots.
- Perplexity had the highest rate of risky sites, at 15 percent of answers.
- Gemini had the lowest rate, at 4 percent; ChatGPT’s rate was 8 percent.
- Podqi advised shoppers to check sources before sharing personal or payment information.