Ask an AI chatbot for a brand’s regional website and you might receive a well-formatted answer pointing to the wrong page entirely. That is not a hypothetical: IP-protection firm Podqi, which sells brand-protection services, documented exactly that scenario in a study reported this fall by The Verge, finding that Perplexity identified an incorrect page as Stüssy’s official site and described it as a source for authentic products. Apps and platforms secretly tracking users represent a parallel concern in the broader landscape of digital deception.
What Podqi Actually Tested
The methodology was narrow and deliberate, which makes the results harder to dismiss.
Podqi built its own shopping-focused prompts and ran them across ChatGPT, Google Gemini, and Perplexity during two testing periods in September. The firm tested 151 brands, large and small, generating 6,795 total answers. Prompts asked where to buy specific brands, which websites carry them, and whether regional or international storefronts exist.
Advertisement
Advertisement
Podqi defined a risky site as any retailer impersonator, counterfeit seller, abandoned domain reregistered by a third party, or inactive domain that could be repurposed for fraud.
The Numbers, by Platform
Every chatbot surfaced risky destinations. The rates differed considerably.
|
Platform |
Risky Site Rate |
|---|---|
|
Perplexity |
15% |
|
ChatGPT |
8% |
|
Gemini |
4% |
These figures cover links and mentions. They also include cases where a chatbot displayed a risky domain while flagging it as potentially unreliable.
ChatGPT issued a warning in only 24 percent of the instances where it surfaced a risky site. That made it the most likely of the three to flag a problem. The rate was still far from consistent.
Smaller brands faced roughly double the exposure of large ones. Queries about regional websites returned risky results more often than other prompt categories, partly because some of those regional sites simply do not exist.
What a Bad Link Can Cost You
The study measured exposure to risky URLs, not confirmed fraud. The potential consequences are real either way.
Advertisement
Advertisement
A shopper who follows an AI-generated link to an impersonator site risks handing over payment details, login credentials, or personal information to a fraudulent operator. Likely outcomes include payment fraud, identity theft, nondelivery, or receipt of counterfeit goods. Podqi’s testing established that risky URLs appear in AI answers with meaningful frequency; the study did not track how many users were ultimately defrauded.
What the Companies Said
All three companies responded. None disputed the core finding.
Google said Gemini displays source links so users can verify information independently, and acknowledged that AI systems can make mistakes. OpenAI said it is working to improve safeguards and reduce deceptive shopping results. Perplexity said it does not claim to be perfectly accurate and emphasized its focus on improving accuracy. Those responses were reported by The Verge, which first published the study’s findings. Readers looking to put AI to better use may find value in exploring AI-Powered Websites that can genuinely enhance productivity.
Treat Every AI Link as a Lead, Not a Verdict
One piece of practical advice from Podqi’s CEO covers most of the risk.
Advertisement
Advertisement
Podqi CEO Trevor West advised consumers to verify any source before entering personal or payment information. That means navigating directly to a brand’s known official website and inspecting the domain carefully for subtle misspellings or unfamiliar extensions. Reviewing contact and return policies before completing a purchase adds another layer of protection. Knowing how to stay safe from digital threats in everyday situations is equally important. An AI-generated link is a starting point to investigate, not confirmation that a site is legitimate.
Leave a Reply