ChatGPT, Gemini and Perplexity sometimes send shoppers to fake store websites, a study found
Medium: worth knowing before holiday shopping, though it's one company's test using its own questions.
What happened
A study by Podqi, a firm that helps brands protect their names, asked ChatGPT, Gemini and Perplexity shopping questions about 151 brands, The Verge reports. Across 6,795 answers, all three sometimes showed risky sites: fake stores, knockoff sellers, old brand addresses now owned by someone else, or dead sites someone could buy. Perplexity did it in 15 percent of answers, ChatGPT 8 percent and Gemini 4 percent. Questions about bigger brands hit risky sites about half as often.
What's actually going on
When you ask a chatbot where to buy something, it often searches the web, reads what it finds and writes a short answer with links. Unlike a plain list of search results, the AI picks a few sites for you and describes them in a confident voice. So a fake store can come across as a recommendation. In one test chat, Perplexity named the wrong page as Stüssy's official website and called it "the best starting point for guaranteed authentic items."
Here's the weak spot. Reading a page doesn't tell a chatbot who really runs it, and a good copy of a store can read just like the real one. Old web addresses a brand gave up can be bought by someone else and still look familiar. The bots also sometimes listed regional sites for brands that don't have any, and questions about regional sites turned up risky links more often than other kinds of questions.
A few things to keep in mind. These numbers count every time a risky site showed up, even when the bot warned about it. ChatGPT warned 24 percent of the time, the best of the three. Podqi wrote its own questions rather than using real shoppers' chats, and one tested brand, Madhappy, is its client. Google said it shows source links so people can check. OpenAI said it's improving its safeguards. Perplexity said it doesn't claim to be 100 percent accurate.
The word you'll hear
Citation
The link a chatbot shows next to its answer to point to where the info came from, so you can check it yourself, though the link itself can lead somewhere wrong.
In your life
Using a chatbot to brainstorm gifts is probably still fine. I'd treat any store link it gives you as a lead, not a stamp of approval. Most answers in this test were clean, but one bad link is enough when your card number is involved.
Be careful about
A fake store can look official, and a chatbot may even call it the real one. Podqi's cofounder, Trevor West, says to verify a site before typing in personal or card information. Smaller brands looked riskier in this test.
Where to get it
There's nothing to download. One small habit: before buying from a store a chatbot suggests, compare its web address to the one on the brand's packaging, receipt or social media page. If it's even slightly off, close the tab.
Go deeper
AI chatbots are recommending fake retail websites, study finds
The Verge, November 12, 2026
How was this brief?
Get one of these every morning
One story, about five minutes, in your inbox by 7 AM Eastern. No spam, and unsubscribing takes one tap.