A new study has highlighted growing AI shopping risks, finding that ChatGPT, Google Gemini and Perplexity sometimes direct users towards potentially unsafe online stores. The research, conducted by intellectual property protection firm Podqi, found that around 9% of 6,795 chatbot answers included at least one risky website, raising concerns about how reliably artificial intelligence tools identify legitimate retailers.
The websites identified in the study included fake online storefronts, suspected counterfeit sellers and inactive or re-registered domains previously associated with established brands. Some of these websites were presented as potential places to purchase products, creating risks for shoppers who rely on AI assistants to find authentic goods.
The findings are particularly important as more consumers use AI-powered services to research products, compare retailers and discover online shopping options.
Perplexity, ChatGPT and Gemini Compared
The Podqi study found differences in how frequently the three AI services surfaced potentially risky websites.
Perplexity had the highest rate, with risky websites appearing in 15% of its answers. ChatGPT followed with 8%, while Google Gemini recorded the lowest rate at 4%.
However, the research counted a website whenever a chatbot included it in an answer, regardless of whether it recommended the retailer, mentioned it without comment or specifically warned users against visiting it.
ChatGPT was the most likely to issue a warning. It flagged potential risks in 24% of its answers that included a risky website.
Across all three services, 57% of answers containing a risky website recommended it as a place to shop. Another 34% mentioned the website without providing a warning, while only 9% explicitly advised users to avoid it.
These figures highlight an important issue: identifying a website in an AI-generated response does not necessarily mean the service has verified its authenticity.
Study Identifies Hundreds of Risky Shopping Domains
Podqi examined 4,437 domains during its research and classified 718 as potentially risky.
The identified domains included 351 fake websites and 367 inactive websites. Such domains can create problems for consumers, particularly when they resemble the official websites of popular brands.
Fake storefronts may use familiar branding, product photographs and website designs to appear legitimate. Shoppers who cannot distinguish these websites from genuine retailers could risk losing money or receiving counterfeit products.
Inactive domains also create uncertainty because previously legitimate brand-related web addresses may expire, change ownership or be registered by unrelated parties.
The study’s findings underline the importance of website verification when using AI assistants for shopping recommendations.
Regional Shopping Questions Carry Higher Risk
One of the most significant findings concerned questions about international shipping and regional online stores.
According to Podqi, these questions produced risky websites in 16% of answers, the highest rate among the five types of prompts tested.
The chatbots sometimes suggested regional websites that were not actually operated by the brands they claimed to represent.
In one example, Perplexity reportedly identified the wrong website as the official Stüssy online store and recommended it as a source of authentic products.
The research also found that questions about larger brands surfaced risky websites around half as often as questions involving smaller brands.
This suggests that shoppers searching for lesser-known labels or regional versions of international retailers may face greater uncertainty when relying on AI-generated recommendations.
For consumers, the AI shopping risks become more significant when a chatbot suggests a local storefront without confirming whether it is officially associated with the brand.
How Podqi Conducted the Research
Podqi developed the prompts used in the study and collected chatbot responses using data from Gauge, an AI search tracking company.
The tests were conducted in two rounds during September and covered 151 brands, including Lululemon, Birkenstock, Stüssy, Le Creuset and Madhappy.
The researchers used logged-out versions of ChatGPT, Gemini and Perplexity to gather responses to shopping-related questions.
The results were based on 6,795 answers and should be understood within the limits of the specific brands, prompts and testing conditions used.
They do not establish that every recommendation from these services is unsafe. Instead, they demonstrate that potentially misleading shopping websites can appear in AI-generated responses, even when users are looking for legitimate retailers.
What OpenAI, Google and Perplexity Said
The companies behind the three services acknowledged the importance of accuracy and the possibility of incorrect information.
Google spokesperson Jessica Levis Boord said Gemini can make mistakes and noted that Google provides source links so users can check the information themselves.
OpenAI spokesperson Taya Christianson said the company is working to improve safeguards and reduce the chances of deceptive websites appearing in shopping-related results.
Perplexity spokesperson Jesse Dwyer referred to the service’s performance on other research benchmarks and said the company does not claim perfect accuracy.
These responses highlight an ongoing challenge for AI companies: improving the reliability of recommendations while helping users independently verify information.
How Shoppers Can Avoid Fake Online Stores
The findings come ahead of the holiday shopping season, when consumers often search for discounts, limited-stock products and international shipping options.
Podqi co-founder and CEO Trevor West advised shoppers to verify websites before entering personal information or payment details.
The company recommends starting from a brand’s main official website when searching for regional stores. Consumers can then follow links provided by the brand rather than relying entirely on chatbot recommendations.
Shoppers should also check website addresses carefully, review retailer contact information and look for independent evidence that a seller is authorised to distribute the products.
AI Shopping Needs Better Verification
The study shows that AI assistants can make online shopping research easier, but their recommendations should not automatically be treated as verified endorsements.
The AI shopping risks identified by Podqi demonstrate why consumers should independently check unfamiliar retailers before making purchases.
As ChatGPT, Gemini and Perplexity become more widely used for product discovery, improving website verification will remain an important challenge for AI developers. Until then, shoppers should use these tools as starting points for research rather than relying on them as the final authority on whether an online store is trustworthy.
