Podqi Study Finds AI Shopping Assistants Recommend Fake Sites 6.3x More Often Than They Warn About Them

·2026.11.13 21:40

Key point

Analysis of 6,795 responses from ChatGPT, Gemini, and Perplexity shows that when risky sites appear, they are recommended as purchase sources 57% of the time, with Perplexity exhibiting the highest appearance rate at 15%.

Details

  • Podqi study of 6,795 answers from ChatGPT, Gemini, and Perplexity reveals that AI systems recommend risky sites 6.3 times more often than they warn users about them.
  • Perplexity had the highest fake-site appearance rate at 15%, followed by ChatGPT at 8% and Gemini at 4%; none of the services eliminated risky site mentions entirely.
  • When a risky site was mentioned, 57% of answers recommended it as a place to buy, while only 24% of ChatGPT's answers included a warning, leaving 76% without any caution.
  • Fake sites appeared in answers about small brands approximately twice as often as for major brands, and AI systems sometimes directed users to regional official sites that did not exist.
  • The study notes that the persuasiveness of AI answers is distinct from the safety of the URLs they contain, requiring users to verify domains against official brand lists.
  • Methodological limitations include a broad definition of risky sites, a limited sample of 151 brands, and the fact that the research firm sells brand-protection services.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.