Cases have been identified in which AI chatbots direct users to fake sales sites or websites that have stopped operating when recommending online shopping malls. As AI-driven product searches and purchases spread, concerns are rising that users' personal and payment information could be exposed to risk.

On Nov. 12, The Verge reported that intellectual property protection firm Podqi said an investigation of ChatGPT, Gemini and Perplexity found all 3 services included risky shopping sites in their answers.

Podqi entered shopping questions about 151 brands, including Lululemon, Birkenstock, Stussy and Le Creuset, in 2 rounds in September and analysed a total of 6,795 responses. The study found risky sites appeared in about 9 percent of all responses. By service, Perplexity was highest at 15 percent, followed by ChatGPT at 8 percent and Gemini at 4 percent.

The risky sites included fake shopping malls impersonating brands, sellers of counterfeit goods, former brand domains re-registered by third parties and websites that have stopped operating. The investigation identified 718 risky domains in total, comprising 351 fake sites and 367 inactive sites.

In particular, 57 percent of the responses in which risky sites appeared recommended the site as a place to buy. Another 34 percent mentioned the sites without any explanation, while only 9 percent warned of the risk. ChatGPT provided warnings in 24 percent of the responses that presented risky sites, the highest rate among the 3 services.

Differences also emerged by question type. For questions about overseas delivery or official sellers by country, the rate of exposure to risky sites reached 16 percent. Cases were also identified in which fake websites were directed for brands that do not have official regional stores.

Google stressed that Gemini recorded the lowest rate of exposure to risky sites in the study. OpenAI said it will improve safety measures for shopping search results, while Perplexity said it continues to focus on improving accuracy.