AI shopping answers cite fake retailer sites, study finds

Imagine a shopper asks a chatbot where to buy a jacket from a brand they trust. The chatbot answers with a link. On Perplexity, one answer in seven points to a fake or risky retailer site. That is what a new study found.

The Verge reported the study this week. IP protection firm Podqi shared the results with The Verge before anyone else. Podqi asked ChatGPT, Gemini and Perplexity shopping questions. The questions used 151 small and large brands, including Lululemon, Birkenstock, Stüssy, Madhappy and Le Creuset. Madhappy is a Podqi client. Two test periods ran in September. Podqi wrote the prompts itself. It did not use real user conversations.

What the three chatbots did

The test covered 6,795 answers. Podqi marked a site "risky" in four cases. A fake store impersonating a real retailer. A site selling knockoffs. An old brand domain reregistered by someone else. A dead site someone else could buy and reuse.

Google's Gemini cited or showed a risky site in 4 percent of answers. That was the lowest of the three. Perplexity did it in 15 percent of answers. That was the highest. ChatGPT sat in the middle at 8 percent. The count included answers that showed a risky site with a warning. ChatGPT warned the user 24 percent of the time it showed a risky site. That was the best warning rate of the three.

Small brands and regional queries did worse

Podqi said questions about smaller brands got risky sites about twice as often. Questions about regional sites also did worse. The chatbots sometimes listed fake regional sites for brands that have no regional presence at all. In one Perplexity answer, the bot named the wrong page as Stüssy's official website. It called the page "the best starting point for guaranteed authentic items".

What the companies said

Google and OpenAI gave statements to The Verge. Google spokesperson Jessica Levis Boord said Gemini "can make mistakes". She said Google shows links alongside answers so people can verify. OpenAI spokesperson Taya Christianson said OpenAI is improving its safeguards. Perplexity did not provide a statement in the article. Podqi CEO and co-founder Trevor West told The Verge that shoppers should verify the source before they enter payment information.

If your brand sells online, this is a direct attack on your citations in AI search. A chatbot that names an imposter site first can hand a buyer to someone else. That is before the buyer opens your own URL. A clean citation path has to be maintained per brand, per region, per chatbot. The three tested here already disagree by a factor of nearly four. That is the discipline a GEO program buys. It is not optional for retailers going into a holiday season. More shoppers are asking a chatbot before they search.

Source

AI chatbots are recommending fake retail websites, study finds, The Verge, by Mia Sato.