Research by IP‑protection firm Podqi shows that Gemini, Perplexity and ChatGPT present risky URLs in 4‑15% of shopping‑related answers, exposing consumers to potentially fraudulent retail sites.
Risks posed by AI chatbots in online shopping
Podqi evaluated three popular chatbots – Google’s Gemini, Perplexity and OpenAI’s ChatGPT – by asking them a set of shopping‑related questions about 151 brands, ranging from Lululemon to Le Creuset. The test revealed that all three models occasionally supplied links to counterfeit or abandoned domains that could be repurposed by malicious actors.
The study defines a “risky” URL as any link that either impersonates a legitimate retailer, sells knock‑off goods, has been re‑registered after expiration, or is completely inactive. Users who follow such links risk fraud, data theft, or exposure to low‑quality products.
Detailed findings of the study
Across 6,795 generated answers, Gemini displayed risky URLs in 4% of cases, the lowest among the three. Perplexity performed worst, with 15% of its responses containing a dubious link. ChatGPT fell in the middle at 8%.
ChatGPT was the most proactive in issuing warnings, flagging a risky site in 24% of the instances where it appeared, whereas Gemini and Perplexity rarely warned users.
Types of risky sites identified
Podqi categorized the problematic links into three groups: (1) sites that masquerade as official brand stores, (2) previously owned domains now controlled by third parties, and (3) dead sites that could be re‑registered. Queries about larger brands such as Lululemon produced risky links about half as often as those about smaller labels.
Responses from the AI providers
Google’s spokesperson Jessica Levis Boord noted that Gemini achieved the best performance and that Google continuously tests and improves its models, emphasizing the display of source links to aid verification.
OpenAI’s Taya Christianson said the company is enhancing safeguards to reduce deceptive site appearances, while acknowledging that the technology is not infallible.
Perplexity’s representative Jesse Dwyer reiterated the company’s focus on research accuracy, but admitted that no AI can guarantee 100% correctness.
Implications for consumers and brands
As the holiday shopping season approaches, more shoppers are likely to rely on chatbots for product discovery and price comparison. The presence of fake URLs could lead to financial loss, erode confidence in AI‑driven commerce, and create reputational damage for brands.
Podqi advises users to verify any link before entering personal or payment information. Brands are encouraged to monitor AI‑generated recommendations and consider implementing URL‑validation tools to protect their customers.
TechTimes Perspective
TechTimes believes the study highlights a critical gap in the way AI chatbots are integrated into e‑commerce workflows. While chatbots promise convenience, their occasional reliance on unverified sources can undermine user trust. For Vietnamese retailers, the findings suggest a need to audit AI‑generated referrals and possibly adopt link‑verification services. If major AI providers improve warning mechanisms and source transparency, the risk of Vietnamese consumers encountering fraudulent sites could diminish, fostering broader acceptance of AI‑assisted shopping in the region.