Shopping with AI? ChatGPT, Gemini and other AI chatbots can lead you to fake websites, study finds

Are you someone who uses AI to perform everyday tasks, whether it is responding to emails or shopping online? Beware! A new report has revealed that ChatGPT, Google Gemini and Perplexity can direct users to fake shopping websites, counterfeit sellers and risky online destinations while answering product-related queries. The security firm Podqi examined 6,795 AI-generated answers involving 151 brands and found that all three chatbots recommended or mentioned potentially unsafe websites during the tests. Perplexity performed the worst, with 15 per cent of its responses containing risky links, followed by ChatGPT at 8 per cent and Gemini at just 4 per cent.

Perplexity shows the highest number of risky websites

According to research by intellectual property protection firm Podqi, Perplexity performed the worst among the three AI platforms tested: ChatGPT, Google Gemini and Perplexity. Data published on Podqi’s website showed that Perplexity provided links to risky websites in 15 per cent of its responses, followed by ChatGPT at 8 per cent. Google Gemini performed the best, with only 4 per cent of its responses containing links to risky websites.

Not only that, but the company also pointed out that ChatGPT performed better when it came to warning users about potentially risky websites, as it issued warnings in 24 per cent of the instances where such websites appeared.

Also read: When will Outlook and other Microsoft 365 services be back? Here is what Windows maker says

Smaller brands face a greater risk from AI shopping results

The Verge, which cited Podqi, said that the company tested the chatbots using questions about 151 companies, including Lululemon, Birkenstock, Stüssy, Madhappy and Le Creuset. The researchers asked the chatbots questions about the best websites to buy a specific brand, where to find its products online, and whether the brand shipped products internationally.

Sharing the findings, Podqi stated that the smaller brands were more vulnerable to the results. Queries about larger companies produced risky website references around half as often.

Researchers also found that questions about regional websites were more likely to generate unsafe results. In some cases, chatbots even suggested fake regional websites for brands that did not have any.

What does the Podqi CEO have to say?

Podqi CEO and co-founder Trevor West said that people should always check the source of information provided by AI chatbots before sharing personal or credit card details. Podqi also found that some AI chatbots showed risky websites more often for certain searches, especially when users searched for regional brand websites. In some cases, AI tools even suggested fake websites for brands that did not have official regional websites. For example, Perplexity wrongly identified a website as Stüssy's official website and described it as a safe place to find authentic products.

Also read: Nvidia Jensen Huang exit after affairs: Elon Musk, Donald Trump, Tim Sweeney and others react

AI companies react

Google said that it takes information quality seriously and was happy that Gemini performed best in Podqi's test. It also said that Gemini can make mistakes and provides links to sources so users can check the information themselves.

OpenAI said that its goal is to help people find reliable shopping websites. It is working to improve its safety measures so that fake or misleading websites are less likely to appear in shopping results. It also thanked researchers for sharing examples that can help improve ChatGPT.

Perplexity said that it has performed well in accuracy tests for four years. However, it also accepted that its AI is not always 100 per cent accurate.