AI Chatbots Fake Stores: Study Finds Shoppers at Risk

AI chatbots can send shoppers to fake stores, a new study has found. Ask an AI chatbot where to buy a product online, and the answer can point you to a fake store, according to research by Podqi, a firm that protects brands’ intellectual property, which tested the shopping answers of ChatGPT, Google’s Gemini and Perplexity.

The study, shared exclusively with The Verge and reported on November 12, found risky websites in about 9 per cent of 6,795 answers. Perplexity showed a risky site in 15 per cent of its answers, ChatGPT in 8 per cent and Gemini in 4 per cent.

The results land just before the holiday shopping season, when many shoppers turn to unfamiliar retailers in search of lower prices or stock that is hard to find elsewhere.

How the AI chatbots fake stores test worked

Podqi wrote its own shopping prompts and ran them in two rounds in September 2026, using data from AI search tracking company Gauge. The prompts asked the three chatbots where to buy products from 151 brands, including Lululemon, Birkenstock, Stussy, Le Creuset and Madhappy. The chatbots’ answers named 4,437 websites. Podqi reviewed each one and classed 718 as risky, including 351 fake sites and 367 inactive ones.

A site counted as risky if it impersonated a real retailer, sold suspected counterfeits, used an old brand domain that someone else had re-registered, or sat inactive in a way that a fraudster could buy and reuse.

The worst single case Podqi documented was concrete. Perplexity named the wrong page as Stussy’s official store and described it as a source of authentic products. Questions about international shipping and regional storefronts carried the highest risk, surfacing risky sites in 16 per cent of answers.

There was one improving trend. The overall rate fell from 12 per cent in the first September round to 7 per cent in the second.

What happened when a risky site appeared

The danger is not only that a risky site shows up. It is what the chatbot does with it. In 57 per cent of the answers that included a risky site, the chatbot recommended it as a place to buy. Another 34 per cent mentioned it without comment, and only 9 per cent warned the shopper away.

ChatGPT gave warnings most often, flagging the risk in 24 per cent of the answers in which it surfaced a risky site.

The companies responded in measured terms. Google spokesperson Jessica Levis Boord said Gemini can make mistakes and that Google shows source links so users can check its answers. OpenAI spokesperson Taya Christianson said the company is improving its safeguards to reduce the chances of deceptive websites appearing in shopping results. Perplexity spokesperson Jesse Dwyer pointed to the company’s performance on other research benchmarks and said the service does not claim perfect accuracy.

Podqi co-founder and chief executive Trevor West advised shoppers to verify a website before entering personal information or payment details. For regional stores, the firm recommends starting from the brand’s main website.

Why Pakistani shoppers should take note

The warning matters in Pakistan too. Online shopping fraud rises sharply around sale events, and shoppers here often meet unfamiliar sellers for the first time during discount seasons. The National Cyber Crime Investigation Agency warned shoppers about scams ahead of the 11-11 sale. With Black Friday falling on November 27 and Cyber Monday on November 30, many Pakistanis will be searching for deals in the coming weeks, and some will ask a chatbot where to buy.

The practical advice is simple: start from the brand’s own website, check that the domain matches the brand, and do not enter card details on a store a chatbot suggested until you have confirmed it is genuine.

Readers should also keep the study in perspective. Podqi sells brand-protection services, so it has a commercial interest in this problem, and one of the brands tested, Madhappy, is a Podqi client. The firm wrote its own prompts rather than studying real user conversations, and it counted any mention of a risky site, even one the chatbot warned about, the same as a recommendation. The study did not track whether any shopper clicked a risky link, bought anything or lost money.

Even with those limits, the pattern is clear. AI chatbots can be confident and wrong at the same time, and a shopping answer that looks helpful can lead to a store that does not exist. OpenAI is also building business profiles into ChatGPT, which will put more shopping inside the chatbot. Google’s Gemini is being woven into Gmail, raising fresh questions about how AI handles personal data. Anthropic is working on a system to prove which AI model produced a given answer.

Frequently asked questions

Which chatbot showed the most fake stores?

Perplexity, at 15 per cent of its answers. ChatGPT showed risky sites in 8 per cent of answers and Gemini in 4 per cent.

How many answers did the Podqi study test?

It tested 6,795 answers across ChatGPT, Gemini and Perplexity, covering 151 brands, in two rounds in September 2026.

What counts as a risky shopping site?

A site that impersonates a real retailer, sells suspected counterfeits, uses an expired brand domain re-registered by someone else, or sits inactive in a way a fraudster could buy and reuse.

Did the chatbots warn shoppers about risky sites?

Rarely. Only 9 per cent of answers that included a risky site warned the shopper away, while 57 per cent recommended the site as a place to buy.

How can shoppers stay safe when using AI chatbots?

Verify the website before entering personal or payment details, and start from the brand’s own main website, especially for regional stores.