Study finds shopping chatbots sometimes point to fake or hijacked store sites

What the test found

As infosecu.technews.tw reported, a study by Podqi, a company that protects intellectual property, found that shoppers who lean on chatbots can be steered to fake retail sites or other risky links. Podqi gave the results exclusively to The Verge.

The test covered ChatGPT, Google Gemini and Perplexity. All three, at times, recommended or displayed suspicious sites when answering shopping questions.

How the test was run

Podqi ran the test in two phases in September, using 151 brands, both large and small. These included Lululemon, Birkenstock, Stüssy, Madhappy and Le Creuset. The chatbots were asked things like which site is best for buying a brand, where to find it online, and whether it ships internationally.

In all, the team analyzed 6,795 answers. It defined a high-risk site as one of four kinds:

• a site posing as an official retailer
• a page selling counterfeits
• an old brand web address that someone else has registered
• a dead site that could be taken over by bad actors

Who did better, and who did worse

Perplexity, an AI search app, had the highest share of answers with a risky site, at 15%. ChatGPT came next at 8%, and Gemini was lowest at 4%.

ChatGPT was the only one the study credited with a notable safeguard: it warned users about a suspicious site 24% of the time. Even so, all three tools sometimes listed or mentioned risky links.

Where the traps are thickest

Podqi found that small brands were more likely than big companies to trigger risky sites. Questions about a brand's regional official site were also especially prone to bad results.

In some cases the chatbots invented regional sites for brands that have none. Podqi's chief executive and co-founder, Trevor West, urged shoppers to confirm a source is genuine before typing in personal or credit card information.