A new study by IP protection firm Podqi reveals that major AI chatbots frequently direct users to fraudulent retail sites when asked about purchasing products. Researchers tested ChatGPT, Gemini, and Perplexity with questions regarding 151 brands, including inquiries about top buying websites and international shipping regions. The experiment ran during two separate periods in September. The results showed that the models often generated links to imposters or flagged domains as safe when they were actually risky. This occurred across all three tested systems despite their different underlying technologies.
The finding matters because it exposes a direct safety gap in how these tools handle e-commerce queries. Users relying on these assistants for shopping advice face an unverified risk of being steered toward phishing sites or scam operations without warning. The models appear to hallucinate specific URLs or fail to distinguish between legitimate merchant pages and knock-off sites during their training or retrieval processes. This creates a tangible threat to consumer security that extends beyond general misinformation.
- The study tested 151 small and large companies.
- Questions focused on brand websites and international shipping.
- All three major models showed similar vulnerability patterns.
