Podqi study says AI tools recommend fake or problematic stores
🎧 Listen to this story
A study conducted by Podqi says ChatGPT, Gemini and Perplexity can direct users to fake, unreliable or inactive stores. In tests carried out in September 2026 with 151 brands in fashion, beauty, supplements, lifestyle and home goods, the tools generated 6,795 responses and mentioned 4,437 websites. Of those, 351 imitated official brands or sold seemingly illegal or dubious-quality products; another 367 recommendations led to suspended, under-construction or inaccessible sites.
According to Podqi, Perplexity performed worst, with problematic sites in 15% of recommendations. ChatGPT reached 8%, while Gemini came in at 4%. OpenAI told The Verge that it is improving safety measures; Google said Gemini performed best among the products tested and that it displays sources to help users verify information. Perplexity also disputed the assessment, saying it has consistently been accurate in reliable reference searches.
Why it matters · editorial interpretation
Podqi’s reported results suggest that users may need to verify stores recommended by assistants before buying, especially when a recommendation leads to imitative or inaccessible websites. The companies’ responses indicate that the scale of the problem depends on the evaluation method and the safety measures in place.
Sources
ResearchSecuritySociety