Big tech’s blue screen programming extends to retail—study shows ChatGPT and Gemini are actively steering you toward scam domains and depopulation data harvesters

The erosion of trust in mainstream institutions has fueled a parallel economy of alternative information, where AI chatbots trained on sanitized “authoritative” datasets often mirror the biases of their corporate creators. This Podqi study, conducted in September 2026 and reported by The Verge, underscores a critical vulnerability: as chatbots become primary gateways to the internet, their propensity to recommend fraudulent storefronts threatens both consumer safety and the credibility of AI itself. The timing is crucial, with holiday shopping approaching and AI adoption accelerating.

Chatbots flunk retail veracity test

A new study from brand protection firm Podqi, tested across 6,795 responses from three leading AI chatbots, reveals a systemic failure to distinguish legitimate retailers from imposters. Perplexity, ChatGPT and Google Gemini all recommended risky websites—defined as counterfeit stores, abandoned domains, or re-registered dead sites—when prompted with shopping queries for 151 brands, including Lululemon, Birkenstock and Le Creuset. Perplexity proved the worst, linking to dangerous sites in 15% of its answers, while ChatGPT and Gemini followed at 8% and 4% respectively. The research, conducted in September 2026, examined five standardized questions, such as “What are the best websites to buy products from this brand?” and “Does this brand have regional online stores?”

The anatomy of a risky recommendation

Beyond raw frequency, the study’s most troubling finding lies in how chatbots presented these threats. Among responses containing a risky URL, 57% actively endorsed the site as a place to shop, while 34% displayed it without comment. Only 9% warned users away. ChatGPT offered warnings in 24% of its risky responses, the best among the three, but still a minority. This confident delivery, lacking the user-controlled ranking of traditional search engines, fosters false trust—a vulnerability scammers can exploit.

Small brands and regional queries most at risk

Smaller, niche brands appeared twice as likely to have their storefronts supplanted by fakes compared to global corporations. The risk spiked when queries targeted regional storefronts: 16% of answers for such prompts included dangerous links, versus 6-7% for other question types. In one case, Perplexity identified an unrelated webpage as the official site for Stüssy, calling it “the best starting point for guaranteed authentic items.” The study also netted 367 inactive domains, representing a latent hazard—scammers can re-register them later.

How AI companies responded

OpenAI, Google and Perplexity acknowledged the issue. A Google spokesperson noted Gemini performed best and emphasized source citations for verification. OpenAI stated it is building safeguards to suppress deceptive sites. Perplexity highlighted its strong accuracy benchmarks while conceding no system is perfect. Encouragingly, the aggregate risky link rate fell from 12% to 7% between the two September testing waves, suggesting some progress.

Keep reading

Unknown's avatar

Author: HP McLovincraft

Seeker of rabbit holes. Pessimist. Libertine. Contrarian. Your huckleberry. Possibly true tales of sanity-blasting horror also known as abject reality. Prepare yourself. Veteran of a thousand psychic wars. I have seen the fnords. Deplatformed on Tumblr and Twitter.

Leave a Reply