AI Chatbots Are Recommending Fake Retail Websites, Study Finds

A study testing Gemini, Perplexity, and ChatGPT found that in as many as 15 percent of conversations, chatbots displayed 'risky' URLs—raising fresh concerns about AI reliability as shopping assistants.

Study Overview

A new study examining the shopping recommendations produced by leading AI chatbots has found that the systems frequently point users toward fraudulent or otherwise untrustworthy retail websites. Researchers tested Gemini, Perplexity, and ChatGPT across a range of shopping-related queries and discovered that in as many as 15 percent of conversations, the chatbots surfaced what the study classified as "risky" URLs.

The findings arrive at a moment when consumers are increasingly turning to AI assistants for product research and purchase decisions. In 2026, conversational commerce has become a mainstream behavior, with a substantial share of online shoppers using chatbots to compare prices, identify alternatives, and locate retailers—making the accuracy and safety of AI-generated links a matter of direct financial risk.

Why the Results Matter

When a chatbot recommends a URL, users often treat it as a trusted referral. But large language models can generate plausible-looking links that do not correspond to legitimate merchants, or may reproduce URLs associated with scam storefronts. In the worst case, a shopper could be directed to a site designed to harvest payment details or personal information.

The 15 percent figure is particularly notable because it reflects everyday, benign shopping queries rather than adversarial prompting. In other words, the risk emerges during ordinary use—precisely the context in which users are least likely to be on guard.

The Broader Context of AI Trust

The study adds to a growing body of evidence that AI chatbots, despite rapid improvements in reasoning and retrieval, remain unreliable as arbiters of factual information—especially when they generate links rather than retrieve verified ones. As AI assistants become embedded in search, shopping, and browser workflows throughout 2026, the boundary between a helpful recommendation and a harmful one can be difficult for users to detect.

Regulators and platform operators have begun responding. Several jurisdictions are examining whether AI-generated recommendations fall under existing consumer protection and advertising rules, while chatbot developers are investing in link verification, source citation, and real-time retrieval to reduce hallucinated URLs.

What Users Can Do

Until link verification becomes more robust, experts advise treating chatbot-suggested URLs with caution: confirm the retailer's domain independently, avoid entering payment information on unfamiliar sites, and prefer chatbots that cite verifiable sources for their recommendations.

Key Takeaways

• Testing of Gemini, Perplexity, and ChatGPT found risky URLs in up to 15 percent of shopping conversations.
• The issue arises during ordinary use, not just adversarial prompting.
• AI-generated links pose direct financial and privacy risks as conversational commerce grows in 2026.
• Verification tools and regulatory scrutiny are expanding, but user vigilance remains essential.

via The Verge AI