AI Chatbots Are Pointing Shoppers at Fake Stores — and Expired Domains Are Part of the Problem

Ask an AI chatbot where to buy a brand, and the answer can look perfectly confident — and point at the wrong store. A study by Podqi, an intellectual-property protection firm, shared exclusively with The Verge and reported on November 12, found that ChatGPT, Gemini and Perplexity all surfaced risky shopping sites in their answers. The weak link is not only fake storefronts. One of the three risk categories Podqi counted is an old brand domain that someone else has re-registered — exactly the kind of domain-history problem a buyer can check before clicking.

What Podqi actually tested

Podqi wrote its own shopping prompts — "what are the best sites to buy [brand]?", "does [brand] have regional stores?" — and ran them across the three chatbots in two testing periods in September, covering 151 brands including Lululemon, Birkenstock, Stüssy, Madhappy and Le Creuset. That produced 6,795 answers. The prompts were not taken from real user conversations, so the numbers are a snapshot of how the chatbots respond to these questions, not a measure of what real shoppers actually clicked or bought.

Podqi counted a site as risky if it impersonated a retailer, sold counterfeits, was an abandoned domain re-registered by a third party, or was an inactive domain a fraudster could pick up and reuse. Of 4,437 domains it reviewed, 718 were flagged: 351 fake and 367 dead.

The numbers

• Perplexity: 15% of answers showed or mentioned a risky site — the highest rate.
• ChatGPT: 8%.
• Gemini: 4% — the lowest of the three.
• About 9% of all answers exposed a shopper to a risky site.

Two details matter more than the ranking. First, Podqi counted any mention, not only a recommendation: a chatbot could show a risky link, mention it without comment, or warn about it and still be counted. Second, in 57% of the answers that included a risky site, the chatbot actively suggested it for shopping — risky sites were recommended 6.3 times as often as they were warned against. ChatGPT warned about a risky site in only 24% of the instances in which it surfaced one.

The worst case Podqi documented was concrete: Perplexity named an incorrect page as the official Stüssy site and described it as a source of authentic products.

Risk rose on one type of question in particular. Queries about regional sites, international shipping and local pages produced risky sites in 16% of responses — the kind of question a cross-border buyer asks most.

Why an expired domain ends up in an AI answer

A brand lets a domain lapse, or an old regional or campaign domain goes quiet. Someone else registers it and puts up a lookalike store. The AI, working from indexed pages and links rather than a live authenticity check, can then surface that domain as if it were still the brand's. Nothing in the answer's confident tone signals that the domain changed hands.

That is a checkable fact, not a guess. Before trusting a shopping link an AI hands you — especially a regional or "official store" link — check the domain's history: when it was registered, whether it changed hands, and whether the site now on it matches the brand. Aiyucha's old-domain history lookup is built for exactly that check.

What to do before you pay

Podqi CEO Trevor West's advice, quoted in the coverage, is plain: verify any source before entering personal or payment information, go straight to a brand's own site, and look for misspelled domains and odd extensions. In practice:

1. Don't pay from the AI's link alone. Open the brand's site yourself — type it, or use a link you already trust — and compare the domain character by character.
2. Check the domain's history. A recently re-registered domain wearing a familiar brand name is the pattern Podqi flagged. If the registration date is new and the brand is old, stop.
3. Treat regional-store answers with extra care. They had the highest risky rate in the test (16%). Verify a regional store through the brand's main site, not through the chatbot's link.
4. A warning is not a guarantee. ChatGPT warned in only 24% of the cases where it showed a risky site. The absence of a warning means nothing.

The limits of this study — stated plainly

Podqi sells brand-protection services, so it has a commercial interest in this problem. The test covered 151 brands in a single month, used prompts Podqi wrote itself, and measured exposure to risky URLs — not confirmed fraud. No part of the study tracked how many shoppers were actually defrauded. Treat the percentages as a snapshot, not a standing rate. The finding that survives those caveats is the mechanism: expired and re-registered domains can pass through an AI answer looking like the real store, and the only reliable defence is checking the domain yourself before you pay.