Study: ChatGPT, Gemini, and Perplexity sometimes recommend fake retail websites
A study by Podqi found that ChatGPT, Gemini, and Perplexity sometimes link to fraudulent or risky websites when responding to shopping queries. Perplexity performed worst (15% of responses), Gemini best (4%).
Intellectual property protection firm Podqi tested the ChatGPT, Gemini, and Perplexity chatbots using shopping queries related to 151 brands (e.g. Lululemon, Birkenstock, Stüssy, Le Creuset). From 6795 responses collected over two testing waves in September, it found that all three chatbots occasionally referenced "risky" websites — meaning fake sites imitating real retailers, counterfeit shops, repurchased old brand domains, or inactive domains that anyone else could buy.
According to the study, risky websites appeared least often with Gemini (4% of responses), in 8% of ChatGPT's responses, and most often with Perplexity (15% of responses). According to Podqi, ChatGPT performed best at flagging the riskiness of a linked website — it did so for 24% of its risky recommendations. According to the study, smaller brands were at roughly twice the risk compared to large companies. Podqi created the queries itself, did not use real user conversations, and tracked the responses using its own tool.
A Google spokesperson said the company takes testing seriously and is pleased with Gemini's good result, adding that the model displays source links alongside its answers so users can verify them. An OpenAI spokesperson said the company continues to improve protection against fraudulent sites in shopping results. A Perplexity spokesperson denied that the company claims 100% accuracy, and said that it focuses on accuracy more than anyone else in the long run.
According to Podqi CEO Trevor West, users should always verify the source of a recommended website before entering personal information or a payment card. According to the article, the topic is particularly relevant ahead of the upcoming Christmas shopping season, when more consumers will be shopping through AI chatbots, including for lesser-known brands.
Why it matters
For ordinary users, this represents a concrete risk when shopping through AI assistants — a recommended link may not lead to the brand's real website, but to a fraudulent copy or an inactive domain that anyone could own. For companies and brands, especially smaller ones, it poses a risk of reputational damage and lost revenue to counterfeiters, without being able to directly influence it.
Two audiences, two different impacts
What this means
For individuals
When shopping based on a chatbot's recommendation (ChatGPT, Gemini, Perplexity), there is a risk that the link leads to a fraudulent or inactive website posing as the real brand, especially for smaller companies.
For a business
According to the study, smaller brands are roughly twice as likely as large companies to have a risky or fraudulent website shown by AI chatbots next to their name, which poses a risk to reputation and revenue.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.