AI chatbots send shoppers to fake stores, Podqi study finds

Each of the three named risky sites somewhere in the 6,795 answers. Podqi counts three kinds: dead sites, old brand domains someone else has re-registered, and shops that impersonate a brand or sell suspected fakes. Gemini showed a risky site in 4% of its answers and ChatGPT in 8%. Perplexity’s rate was 15%, the highest of the three. When ChatGPT showed a risky site, it warned the shopper about it in 24% of those answers, the best rate of the three.

For every warning against a risky site there were 6.3 recommendations of one, Podqi’s own report adds. The question that drew the most risky sites asked for a brand’s regional websites. Lists are risky by design, Podqi argues: two-thirds of risky answers to the question asking for a brand’s top websites recommended the risky site as a place to buy. Between the first and second rounds, the share of risky answers fell from 12% to 7%. Google’s AI Overviews warned the shopper in each of their seven risky answers.

ChatGPT’s answers are also turning into ad space: its Go plan card said “This plan may include ads” when we read it on 31 October, as our ChatGPT page records.

Since the question asking for a brand’s regional websites showed a risky site in 16% of answers, any brand whose shoppers ask whether it ships to their country is exposed on the riskiest question in the study.