Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarThe Verge→ original

AI Chatbots Are Sending Shoppers to Fake Stores: Perplexity in 15% of Answers, Gemini in 4%

Ask an AI chatbot where to buy sneakers, and the answer may point to a lookalike site, a counterfeit storefront or a lapsed brand domain. Podqi, a company that protects brands from counterfeits, asked ChatGPT, Gemini and Perplexity about 151 brands and analyzed 6,795 answers. A risky link appeared in 15% of Perplexity's answers, 8% of ChatGPT's and 4% of Gemini's. That makes it worth checking a store address from an AI answer before entering card details.

AI-processed from The Verge; edited by Hamidun News
AI Chatbots Are Sending Shoppers to Fake Stores: Perplexity in 15% of Answers, Gemini in 4%
Source: The Verge. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Verge ↗
Editor-in-Chief
Key metric #1Major brands had about half as many risky links as smaller ones
Key metric #2ChatGPT warned about a questionable site in only 24% of cases where it found one
Key metric #3Chatbots performed worst on questions about regional brand sites

Ask a chatbot where to buy Stüssy sneakers or a Le Creuset pot, and you will most likely get a confident answer with a link. The problem is that the link does not always lead to a real store. A new study by Podqi, a company that protects brands from counterfeits, found that ChatGPT, Gemini and Perplexity sometimes name lookalike sites, counterfeit shops and abandoned domains that someone else has bought.

The share of such answers ranged from 4% for Gemini to 15% for Perplexity. A "risky site" here means any of these three categories, and the percentages show how often a bot mentioned such an address at least once. Before holiday sales, when people look for in-stock goods and the best price, this can cost real money.

What was tested and how

Podqi ran questions about 151 brands through three chatbots, exclusively for The Verge. The list ranged from giants to small labels and included Lululemon, Birkenstock, Stüssy, Le Creuset and Madhappy, which is itself a Podqi client. The questions were simple: "What are the best sites to buy [brand]?", "Where can I find [brand] online?" and "Does [brand] ship internationally, and what are its regional sites?" The test ran in two stages in September and produced 6,795 answers in total.

There were three types of risky addresses. The first is fake stores that pass themselves off as genuine or sell counterfeits. The second is old brand domains that someone else has re-registered. The third is dead sites that no longer work but whose address anyone could buy and use. Imagine a brand that moved to a new address and abandoned the old one. The bot keeps treating it as official, while in reality another person already owns it.

Results by chatbot

Perplexity performed worst: a risky site appeared in 15% of its answers. ChatGPT showed 8%, and Gemini did best with 4%. "Showed" means any mention: a recommendation, a link without comment or a warning to the user. On warnings, ChatGPT looks better: it added a flag to a risky site it found in 24% of cases.

The problem hits smaller brands hardest. In answers about large companies, risky sites appeared roughly half as often. Well-known brands may have more reliable sources for search to rely on, while the picture for smaller ones is blurrier and fakes slip through more easily. This is a hypothesis; the study does not examine the causes.

Chatbots handled questions about regional sites worst of all. Sometimes they named such sites even for brands that do not have them. In one conversation, Perplexity identified an incorrect page as Stüssy's official site and added that it was "the best starting point for guaranteed authentic products." That sounds convincing, which is exactly why it is dangerous: a person sees one confident recommendation rather than a list of ten links.

What companies say and what to do

A Google representative noted that Gemini shows source links next to its answers so people can verify the information. OpenAI said it continues to improve its protections so that questionable sites appear less often in shopping results. Perplexity replied that it does not claim 100% accuracy but focuses on it more intensely than others.

The study has limitations. Podqi wrote the questions itself rather than drawing them from real user conversations. The report's commissioner sells brand protection, so it has an interest in showing the scale of the problem. The test covers three chatbots and one period. Finally, "risky" includes simply dead domains, not only active fraudulent stores, so 15% does not equal 15% of answers involving direct fraud.

The practical takeaways are these. For shoppers: open a site from an AI answer as you would an unfamiliar one, check the domain and reviews, and enter card details only after that. This matters especially for regional versions of stores and for unfamiliar brands. For brands: monitor old and forgotten domains, do not leave them without an owner, and regularly check what chatbots say about you. For teams building their own AI shopping assistants: add checks for link freshness and domain ownership, and show warnings about questionable sources far more often than in 24% of cases.

Story

Key Questions Answered

What was tested and how?

Podqi ran questions about 151 brands through three chatbots, exclusively for The Verge. The list ranged from giants to small labels and included Lululemon, Birkenstock, Stüssy, Le Creuset and Madhappy, which is itself a Podqi client. The questions were simple: "What are the best sites to buy [brand]?", "Where can I find [brand] online?" and "Does [brand] ship internationally, and what are its regional sites?" The test ran in two stages in September and produced 6,795 answers in total.

What are the key details on Results by chatbot?

Perplexity performed worst: a risky site appeared in 15% of its answers. ChatGPT showed 8%, and Gemini did best with 4%. "Showed" means any mention: a recommendation, a link without comment or a warning to the user. On warnings, ChatGPT looks better: it added a flag to a risky site it found in 24% of cases. The problem hits smaller brands hardest. In answers about large companies, risky sites appeared roughly half as often.

What companies say and what to do?

A Google representative noted that Gemini shows source links next to its answers so people can verify the information. OpenAI said it continues to improve its protections so that questionable sites appear less often in shopping results. Perplexity replied that it does not claim 100% accuracy but focuses on it more intensely than others. The study has limitations. Podqi wrote the questions itself rather than drawing them from real user conversations.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…