Podqi finds risky shopping websites in chatbot answers

Text
iHOLYSHITVISUAL / TECHOnline phenomenaiHOLYSHIT / ILLUSTRATION
iHolyShit illustration

A 6,795-answer test separates recommendations from mentions and warnings about questionable shopping destinations.

An AI shopping answer can look reassuring while pointing toward the wrong storefront. A study by brand-protection company Podqi found risky websites in roughly nine percent of 6,795 responses from three chatbots. Its category included suspected impersonators and counterfeit sellers as well as inactive domains. The concern is the confidence a polished answer can lend to a destination that a shopper has never visited.

Podqi worked with AI-search company Gauge on two rounds of testing in September 2026, asking about 151 brands. The Verge reported the findings on November 12. The experiment used researcher-written questions and logged-out consumer tools. Podqi sells brand-protection services, a commercial interest worth keeping in view alongside the results. This was an examination of answers and destinations, with a defined set of shopping questions.

The study distinguished recommending a risky site, mentioning it without comment and warning about it. Those are different behaviors, even though all put the address into an answer. Risky sites appeared in about 15 percent of Perplexity responses, eight percent of ChatGPT responses and four percent of Gemini responses. In responses to The Verge, the companies addressed safeguards, potential errors and accuracy. The experiment measured chatbot output rather than shoppers’ eventual financial losses.

Think of the assistant as someone drawing a convenient map of a shopping mall. A neat map does not establish who owns every shop. Similarly, a clear paragraph can connect a brand name, an address and the word official without making that combination trustworthy. A useful separation is to let the assistant help identify what you want, then use the brand’s own established website to find a regional store or authorized seller.

For shoppers, choosing a product and choosing whom to pay are separate decisions. Comparing dimensions, colors and intended uses is one task; deciding whether an unfamiliar merchant should receive an order is another. A moment spent checking the domain or returning through a familiar brand entrance can interrupt a misleading path. For companies building shopping assistants, merchant identity needs to be as legible as the recommendation itself. Convenience matters most when it leads to the right checkout.

Original report published: November 13, 2026, 01:00 Beijing time / November 12, 2026, 12:00 PM EST.

Share this article

What do you think of this story?

Anonymous feedback, up to 5 reactions per article.

Anonymous comments

Share your thoughts without signing in.

Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.

0/500

No comments yet. Be the first to share.

Sources

Podqi finds risky shopping websites in chatbot answers|iHolyShit